rocksdb/db/version_set.cc

5865 lines
215 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/version_set.h"
#include <algorithm>
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
#include <array>
#include <cinttypes>
#include <cstdio>
#include <list>
#include <map>
2014-01-30 00:26:43 +01:00
#include <set>
#include <string>
#include <unordered_map>
#include <vector>
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
#include "compaction/compaction.h"
#include "db/blob/blob_fetcher.h"
#include "db/blob/blob_file_cache.h"
#include "db/blob/blob_file_reader.h"
#include "db/blob/blob_index.h"
#include "db/internal_stats.h"
#include "db/log_reader.h"
#include "db/log_writer.h"
#include "db/memtable.h"
#include "db/merge_context.h"
#include "db/merge_helper.h"
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
2016-07-20 18:49:03 +02:00
#include "db/pinned_iterators_manager.h"
#include "db/table_cache.h"
#include "db/version_builder.h"
#include "db/version_edit_handler.h"
#include "file/filename.h"
#include "file/random_access_file_reader.h"
#include "file/read_write_util.h"
#include "file/writable_file_writer.h"
#include "monitoring/file_read_sample.h"
#include "monitoring/perf_context_imp.h"
#include "monitoring/persistent_stats_history.h"
#include "options/options_helper.h"
#include "rocksdb/env.h"
#include "rocksdb/merge_operator.h"
#include "rocksdb/write_buffer_manager.h"
#include "table/format.h"
#include "table/get_context.h"
#include "table/internal_iterator.h"
#include "table/merging_iterator.h"
#include "table/meta_blocks.h"
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
#include "table/multiget_context.h"
#include "table/plain/plain_table_factory.h"
#include "table/table_reader.h"
#include "table/two_level_iterator.h"
#include "test_util/sync_point.h"
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
#include "util/cast_util.h"
#include "util/coding.h"
#include "util/stop_watch.h"
#include "util/string_util.h"
#include "util/user_comparator_wrapper.h"
namespace ROCKSDB_NAMESPACE {
namespace {
// Find File in LevelFilesBrief data structure
// Within an index range defined by left and right
int FindFileInRange(const InternalKeyComparator& icmp,
const LevelFilesBrief& file_level,
const Slice& key,
uint32_t left,
uint32_t right) {
auto cmp = [&](const FdWithKeyRange& f, const Slice& k) -> bool {
return icmp.InternalKeyComparator::Compare(f.largest_key, k) < 0;
};
const auto &b = file_level.files;
return static_cast<int>(std::lower_bound(b + left,
b + right, key, cmp) - b);
}
Status OverlapWithIterator(const Comparator* ucmp,
const Slice& smallest_user_key,
const Slice& largest_user_key,
InternalIterator* iter,
bool* overlap) {
InternalKey range_start(smallest_user_key, kMaxSequenceNumber,
kValueTypeForSeek);
iter->Seek(range_start.Encode());
if (!iter->status().ok()) {
return iter->status();
}
*overlap = false;
if (iter->Valid()) {
ParsedInternalKey seek_result;
Status s = ParseInternalKey(iter->key(), &seek_result,
false /* log_err_key */); // TODO
if (!s.ok()) return s;
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
if (ucmp->CompareWithoutTimestamp(seek_result.user_key, largest_user_key) <=
0) {
*overlap = true;
}
}
return iter->status();
}
// Class to help choose the next file to search for the particular key.
// Searches and returns files level by level.
// We can search level-by-level since entries never hop across
// levels. Therefore we are guaranteed that if we find data
// in a smaller level, later levels are irrelevant (unless we
// are MergeInProgress).
class FilePicker {
public:
FilePicker(std::vector<FileMetaData*>* files, const Slice& user_key,
const Slice& ikey, autovector<LevelFilesBrief>* file_levels,
unsigned int num_levels, FileIndexer* file_indexer,
const Comparator* user_comparator,
const InternalKeyComparator* internal_comparator)
: num_levels_(num_levels),
curr_level_(static_cast<unsigned int>(-1)),
returned_file_level_(static_cast<unsigned int>(-1)),
hit_file_level_(static_cast<unsigned int>(-1)),
search_left_bound_(0),
search_right_bound_(FileIndexer::kLevelMaxIndex),
#ifndef NDEBUG
files_(files),
#endif
level_files_brief_(file_levels),
is_hit_file_last_in_level_(false),
curr_file_level_(nullptr),
user_key_(user_key),
ikey_(ikey),
file_indexer_(file_indexer),
user_comparator_(user_comparator),
internal_comparator_(internal_comparator) {
#ifdef NDEBUG
(void)files;
#endif
// Setup member variables to search first level.
search_ended_ = !PrepareNextLevel();
if (!search_ended_) {
// Prefetch Level 0 table data to avoid cache miss if possible.
for (unsigned int i = 0; i < (*level_files_brief_)[0].num_files; ++i) {
auto* r = (*level_files_brief_)[0].files[i].fd.table_reader;
if (r) {
r->Prepare(ikey);
}
}
}
}
int GetCurrentLevel() const { return curr_level_; }
FdWithKeyRange* GetNextFile() {
while (!search_ended_) { // Loops over different levels.
while (curr_index_in_curr_level_ < curr_file_level_->num_files) {
// Loops over all files in current level.
FdWithKeyRange* f = &curr_file_level_->files[curr_index_in_curr_level_];
hit_file_level_ = curr_level_;
is_hit_file_last_in_level_ =
curr_index_in_curr_level_ == curr_file_level_->num_files - 1;
int cmp_largest = -1;
// Do key range filtering of files or/and fractional cascading if:
// (1) not all the files are in level 0, or
// (2) there are more than 3 current level files
// If there are only 3 or less current level files in the system, we skip
// the key range filtering. In this case, more likely, the system is
// highly tuned to minimize number of tables queried by each query,
// so it is unlikely that key range filtering is more efficient than
// querying the files.
if (num_levels_ > 1 || curr_file_level_->num_files > 3) {
// Check if key is within a file's range. If search left bound and
// right bound point to the same find, we are sure key falls in
// range.
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
assert(curr_level_ == 0 ||
curr_index_in_curr_level_ == start_index_in_curr_level_ ||
user_comparator_->CompareWithoutTimestamp(
user_key_, ExtractUserKey(f->smallest_key)) <= 0);
int cmp_smallest = user_comparator_->CompareWithoutTimestamp(
user_key_, ExtractUserKey(f->smallest_key));
if (cmp_smallest >= 0) {
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
cmp_largest = user_comparator_->CompareWithoutTimestamp(
user_key_, ExtractUserKey(f->largest_key));
}
// Setup file search bound for the next level based on the
// comparison results
if (curr_level_ > 0) {
file_indexer_->GetNextLevelIndex(curr_level_,
curr_index_in_curr_level_,
cmp_smallest, cmp_largest,
&search_left_bound_,
&search_right_bound_);
}
// Key falls out of current file's range
if (cmp_smallest < 0 || cmp_largest > 0) {
if (curr_level_ == 0) {
++curr_index_in_curr_level_;
continue;
} else {
// Search next level.
break;
}
}
}
#ifndef NDEBUG
// Sanity check to make sure that the files are correctly sorted
if (prev_file_) {
if (curr_level_ != 0) {
int comp_sign = internal_comparator_->Compare(
prev_file_->largest_key, f->smallest_key);
assert(comp_sign < 0);
} else {
// level == 0, the current file cannot be newer than the previous
// one. Use compressed data structure, has no attribute seqNo
assert(curr_index_in_curr_level_ > 0);
assert(!NewestFirstBySeqNo(files_[0][curr_index_in_curr_level_],
files_[0][curr_index_in_curr_level_-1]));
}
}
prev_file_ = f;
#endif
returned_file_level_ = curr_level_;
if (curr_level_ > 0 && cmp_largest < 0) {
// No more files to search in this level.
search_ended_ = !PrepareNextLevel();
} else {
++curr_index_in_curr_level_;
}
return f;
}
// Start searching next level.
search_ended_ = !PrepareNextLevel();
}
// Search ended.
return nullptr;
}
// getter for current file level
// for GET_HIT_L0, GET_HIT_L1 & GET_HIT_L2_AND_UP counts
unsigned int GetHitFileLevel() { return hit_file_level_; }
// Returns true if the most recent "hit file" (i.e., one returned by
// GetNextFile()) is at the last index in its level.
bool IsHitFileLastInLevel() { return is_hit_file_last_in_level_; }
private:
unsigned int num_levels_;
unsigned int curr_level_;
unsigned int returned_file_level_;
unsigned int hit_file_level_;
int32_t search_left_bound_;
int32_t search_right_bound_;
#ifndef NDEBUG
std::vector<FileMetaData*>* files_;
#endif
autovector<LevelFilesBrief>* level_files_brief_;
bool search_ended_;
bool is_hit_file_last_in_level_;
LevelFilesBrief* curr_file_level_;
unsigned int curr_index_in_curr_level_;
unsigned int start_index_in_curr_level_;
Slice user_key_;
Slice ikey_;
FileIndexer* file_indexer_;
const Comparator* user_comparator_;
const InternalKeyComparator* internal_comparator_;
#ifndef NDEBUG
FdWithKeyRange* prev_file_;
#endif
// Setup local variables to search next level.
// Returns false if there are no more levels to search.
bool PrepareNextLevel() {
curr_level_++;
while (curr_level_ < num_levels_) {
curr_file_level_ = &(*level_files_brief_)[curr_level_];
if (curr_file_level_->num_files == 0) {
// When current level is empty, the search bound generated from upper
// level must be [0, -1] or [0, FileIndexer::kLevelMaxIndex] if it is
// also empty.
assert(search_left_bound_ == 0);
assert(search_right_bound_ == -1 ||
search_right_bound_ == FileIndexer::kLevelMaxIndex);
// Since current level is empty, it will need to search all files in
// the next level
search_left_bound_ = 0;
search_right_bound_ = FileIndexer::kLevelMaxIndex;
curr_level_++;
continue;
}
// Some files may overlap each other. We find
// all files that overlap user_key and process them in order from
// newest to oldest. In the context of merge-operator, this can occur at
// any level. Otherwise, it only occurs at Level-0 (since Put/Deletes
// are always compacted into a single entry).
int32_t start_index;
if (curr_level_ == 0) {
// On Level-0, we read through all files to check for overlap.
start_index = 0;
} else {
// On Level-n (n>=1), files are sorted. Binary search to find the
// earliest file whose largest key >= ikey. Search left bound and
// right bound are used to narrow the range.
if (search_left_bound_ <= search_right_bound_) {
if (search_right_bound_ == FileIndexer::kLevelMaxIndex) {
search_right_bound_ =
static_cast<int32_t>(curr_file_level_->num_files) - 1;
}
// `search_right_bound_` is an inclusive upper-bound, but since it was
// determined based on user key, it is still possible the lookup key
// falls to the right of `search_right_bound_`'s corresponding file.
// So, pass a limit one higher, which allows us to detect this case.
start_index =
FindFileInRange(*internal_comparator_, *curr_file_level_, ikey_,
static_cast<uint32_t>(search_left_bound_),
static_cast<uint32_t>(search_right_bound_) + 1);
if (start_index == search_right_bound_ + 1) {
// `ikey_` comes after `search_right_bound_`. The lookup key does
// not exist on this level, so let's skip this level and do a full
// binary search on the next level.
search_left_bound_ = 0;
search_right_bound_ = FileIndexer::kLevelMaxIndex;
curr_level_++;
continue;
}
} else {
// search_left_bound > search_right_bound, key does not exist in
2015-04-25 11:14:27 +02:00
// this level. Since no comparison is done in this level, it will
// need to search all files in the next level.
search_left_bound_ = 0;
search_right_bound_ = FileIndexer::kLevelMaxIndex;
curr_level_++;
continue;
}
}
start_index_in_curr_level_ = start_index;
curr_index_in_curr_level_ = start_index;
#ifndef NDEBUG
prev_file_ = nullptr;
#endif
return true;
}
// curr_level_ = num_levels_. So, no more levels to search.
return false;
}
};
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
class FilePickerMultiGet {
private:
struct FilePickerContext;
public:
FilePickerMultiGet(MultiGetRange* range,
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
autovector<LevelFilesBrief>* file_levels,
unsigned int num_levels, FileIndexer* file_indexer,
const Comparator* user_comparator,
const InternalKeyComparator* internal_comparator)
: num_levels_(num_levels),
curr_level_(static_cast<unsigned int>(-1)),
returned_file_level_(static_cast<unsigned int>(-1)),
hit_file_level_(static_cast<unsigned int>(-1)),
range_(range),
batch_iter_(range->begin()),
batch_iter_prev_(range->begin()),
upper_key_(range->begin()),
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
maybe_repeat_key_(false),
current_level_range_(*range, range->begin(), range->end()),
current_file_range_(*range, range->begin(), range->end()),
level_files_brief_(file_levels),
is_hit_file_last_in_level_(false),
curr_file_level_(nullptr),
file_indexer_(file_indexer),
user_comparator_(user_comparator),
internal_comparator_(internal_comparator) {
for (auto iter = range_->begin(); iter != range_->end(); ++iter) {
fp_ctx_array_[iter.index()] =
FilePickerContext(0, FileIndexer::kLevelMaxIndex);
}
// Setup member variables to search first level.
search_ended_ = !PrepareNextLevel();
if (!search_ended_) {
// REVISIT
// Prefetch Level 0 table data to avoid cache miss if possible.
// As of now, only PlainTableReader and CuckooTableReader do any
// prefetching. This may not be necessary anymore once we implement
// batching in those table readers
for (unsigned int i = 0; i < (*level_files_brief_)[0].num_files; ++i) {
auto* r = (*level_files_brief_)[0].files[i].fd.table_reader;
if (r) {
for (auto iter = range_->begin(); iter != range_->end(); ++iter) {
r->Prepare(iter->ikey);
}
}
}
}
}
int GetCurrentLevel() const { return curr_level_; }
// Iterates through files in the current level until it finds a file that
// contains at least one key from the MultiGet batch
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
bool GetNextFileInLevelWithKeys(MultiGetRange* next_file_range,
size_t* file_index, FdWithKeyRange** fd,
bool* is_last_key_in_file) {
size_t curr_file_index = *file_index;
FdWithKeyRange* f = nullptr;
bool file_hit = false;
int cmp_largest = -1;
if (curr_file_index >= curr_file_level_->num_files) {
// In the unlikely case the next key is a duplicate of the current key,
// and the current key is the last in the level and the internal key
// was not found, we need to skip lookup for the remaining keys and
// reset the search bounds
if (batch_iter_ != current_level_range_.end()) {
++batch_iter_;
for (; batch_iter_ != current_level_range_.end(); ++batch_iter_) {
struct FilePickerContext& fp_ctx = fp_ctx_array_[batch_iter_.index()];
fp_ctx.search_left_bound = 0;
fp_ctx.search_right_bound = FileIndexer::kLevelMaxIndex;
}
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
return false;
}
// Loops over keys in the MultiGet batch until it finds a file with
// atleast one of the keys. Then it keeps moving forward until the
// last key in the batch that falls in that file
while (batch_iter_ != current_level_range_.end() &&
(fp_ctx_array_[batch_iter_.index()].curr_index_in_curr_level ==
curr_file_index ||
!file_hit)) {
struct FilePickerContext& fp_ctx = fp_ctx_array_[batch_iter_.index()];
f = &curr_file_level_->files[fp_ctx.curr_index_in_curr_level];
Slice& user_key = batch_iter_->ukey_without_ts;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
// Do key range filtering of files or/and fractional cascading if:
// (1) not all the files are in level 0, or
// (2) there are more than 3 current level files
// If there are only 3 or less current level files in the system, we
// skip the key range filtering. In this case, more likely, the system
// is highly tuned to minimize number of tables queried by each query,
// so it is unlikely that key range filtering is more efficient than
// querying the files.
if (num_levels_ > 1 || curr_file_level_->num_files > 3) {
// Check if key is within a file's range. If search left bound and
// right bound point to the same find, we are sure key falls in
// range.
int cmp_smallest = user_comparator_->CompareWithoutTimestamp(
user_key, false, ExtractUserKey(f->smallest_key), true);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
assert(curr_level_ == 0 ||
fp_ctx.curr_index_in_curr_level ==
fp_ctx.start_index_in_curr_level ||
cmp_smallest <= 0);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
if (cmp_smallest >= 0) {
cmp_largest = user_comparator_->CompareWithoutTimestamp(
user_key, false, ExtractUserKey(f->largest_key), true);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
} else {
cmp_largest = -1;
}
// Setup file search bound for the next level based on the
// comparison results
if (curr_level_ > 0) {
file_indexer_->GetNextLevelIndex(
curr_level_, fp_ctx.curr_index_in_curr_level, cmp_smallest,
cmp_largest, &fp_ctx.search_left_bound,
&fp_ctx.search_right_bound);
}
// Key falls out of current file's range
if (cmp_smallest < 0 || cmp_largest > 0) {
next_file_range->SkipKey(batch_iter_);
} else {
file_hit = true;
}
} else {
file_hit = true;
}
if (cmp_largest == 0) {
// cmp_largest is 0, which means the next key will not be in this
// file, so stop looking further. However, its possible there are
// duplicates in the batch, so find the upper bound for the batch
// in this file (upper_key_) by skipping past the duplicates. We
// leave batch_iter_ as is since we may have to pick up from there
// for the next file, if this file has a merge value rather than
// final value
upper_key_ = batch_iter_;
++upper_key_;
while (upper_key_ != current_level_range_.end() &&
user_comparator_->CompareWithoutTimestamp(
batch_iter_->ukey_without_ts, false,
upper_key_->ukey_without_ts, false) == 0) {
++upper_key_;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
break;
} else {
if (curr_level_ == 0) {
// We need to look through all files in level 0
++fp_ctx.curr_index_in_curr_level;
}
++batch_iter_;
}
if (!file_hit) {
curr_file_index =
(batch_iter_ != current_level_range_.end())
? fp_ctx_array_[batch_iter_.index()].curr_index_in_curr_level
: curr_file_level_->num_files;
}
}
*fd = f;
*file_index = curr_file_index;
*is_last_key_in_file = cmp_largest == 0;
if (!*is_last_key_in_file) {
// If the largest key in the batch overlapping the file is not the
// largest key in the file, upper_ley_ would not have been updated so
// update it here
upper_key_ = batch_iter_;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
return file_hit;
}
FdWithKeyRange* GetNextFile() {
while (!search_ended_) {
// Start searching next level.
if (batch_iter_ == current_level_range_.end()) {
search_ended_ = !PrepareNextLevel();
continue;
} else {
if (maybe_repeat_key_) {
maybe_repeat_key_ = false;
// Check if we found the final value for the last key in the
// previous lookup range. If we did, then there's no need to look
// any further for that key, so advance batch_iter_. Else, keep
// batch_iter_ positioned on that key so we look it up again in
// the next file
// For L0, always advance the key because we will look in the next
// file regardless for all keys not found yet
if (current_level_range_.CheckKeyDone(batch_iter_) ||
curr_level_ == 0) {
batch_iter_ = upper_key_;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
}
// batch_iter_prev_ will become the start key for the next file
// lookup
batch_iter_prev_ = batch_iter_;
}
MultiGetRange next_file_range(current_level_range_, batch_iter_prev_,
current_level_range_.end());
size_t curr_file_index =
(batch_iter_ != current_level_range_.end())
? fp_ctx_array_[batch_iter_.index()].curr_index_in_curr_level
: curr_file_level_->num_files;
FdWithKeyRange* f;
bool is_last_key_in_file;
if (!GetNextFileInLevelWithKeys(&next_file_range, &curr_file_index, &f,
&is_last_key_in_file)) {
search_ended_ = !PrepareNextLevel();
} else {
if (is_last_key_in_file) {
// Since cmp_largest is 0, batch_iter_ still points to the last key
// that falls in this file, instead of the next one. Increment
// the file index for all keys between batch_iter_ and upper_key_
auto tmp_iter = batch_iter_;
while (tmp_iter != upper_key_) {
++(fp_ctx_array_[tmp_iter.index()].curr_index_in_curr_level);
++tmp_iter;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
maybe_repeat_key_ = true;
}
// Set the range for this file
current_file_range_ =
MultiGetRange(next_file_range, batch_iter_prev_, upper_key_);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
returned_file_level_ = curr_level_;
hit_file_level_ = curr_level_;
is_hit_file_last_in_level_ =
curr_file_index == curr_file_level_->num_files - 1;
return f;
}
}
// Search ended
return nullptr;
}
// getter for current file level
// for GET_HIT_L0, GET_HIT_L1 & GET_HIT_L2_AND_UP counts
unsigned int GetHitFileLevel() { return hit_file_level_; }
// Returns true if the most recent "hit file" (i.e., one returned by
// GetNextFile()) is at the last index in its level.
bool IsHitFileLastInLevel() { return is_hit_file_last_in_level_; }
const MultiGetRange& CurrentFileRange() { return current_file_range_; }
private:
unsigned int num_levels_;
unsigned int curr_level_;
unsigned int returned_file_level_;
unsigned int hit_file_level_;
struct FilePickerContext {
int32_t search_left_bound;
int32_t search_right_bound;
unsigned int curr_index_in_curr_level;
unsigned int start_index_in_curr_level;
FilePickerContext(int32_t left, int32_t right)
: search_left_bound(left), search_right_bound(right),
curr_index_in_curr_level(0), start_index_in_curr_level(0) {}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
FilePickerContext() = default;
};
std::array<FilePickerContext, MultiGetContext::MAX_BATCH_SIZE> fp_ctx_array_;
MultiGetRange* range_;
// Iterator to iterate through the keys in a MultiGet batch, that gets reset
// at the beginning of each level. Each call to GetNextFile() will position
// batch_iter_ at or right after the last key that was found in the returned
// SST file
MultiGetRange::Iterator batch_iter_;
// An iterator that records the previous position of batch_iter_, i.e last
// key found in the previous SST file, in order to serve as the start of
// the batch key range for the next SST file
MultiGetRange::Iterator batch_iter_prev_;
MultiGetRange::Iterator upper_key_;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
bool maybe_repeat_key_;
MultiGetRange current_level_range_;
MultiGetRange current_file_range_;
autovector<LevelFilesBrief>* level_files_brief_;
bool search_ended_;
bool is_hit_file_last_in_level_;
LevelFilesBrief* curr_file_level_;
FileIndexer* file_indexer_;
const Comparator* user_comparator_;
const InternalKeyComparator* internal_comparator_;
// Setup local variables to search next level.
// Returns false if there are no more levels to search.
bool PrepareNextLevel() {
if (curr_level_ == 0) {
MultiGetRange::Iterator mget_iter = current_level_range_.begin();
if (fp_ctx_array_[mget_iter.index()].curr_index_in_curr_level <
curr_file_level_->num_files) {
batch_iter_prev_ = current_level_range_.begin();
upper_key_ = batch_iter_ = current_level_range_.begin();
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
return true;
}
}
curr_level_++;
// Reset key range to saved value
while (curr_level_ < num_levels_) {
bool level_contains_keys = false;
curr_file_level_ = &(*level_files_brief_)[curr_level_];
if (curr_file_level_->num_files == 0) {
// When current level is empty, the search bound generated from upper
// level must be [0, -1] or [0, FileIndexer::kLevelMaxIndex] if it is
// also empty.
for (auto mget_iter = current_level_range_.begin();
mget_iter != current_level_range_.end(); ++mget_iter) {
struct FilePickerContext& fp_ctx = fp_ctx_array_[mget_iter.index()];
assert(fp_ctx.search_left_bound == 0);
assert(fp_ctx.search_right_bound == -1 ||
fp_ctx.search_right_bound == FileIndexer::kLevelMaxIndex);
// Since current level is empty, it will need to search all files in
// the next level
fp_ctx.search_left_bound = 0;
fp_ctx.search_right_bound = FileIndexer::kLevelMaxIndex;
}
// Skip all subsequent empty levels
do {
++curr_level_;
} while ((curr_level_ < num_levels_) &&
(*level_files_brief_)[curr_level_].num_files == 0);
continue;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
// Some files may overlap each other. We find
// all files that overlap user_key and process them in order from
// newest to oldest. In the context of merge-operator, this can occur at
// any level. Otherwise, it only occurs at Level-0 (since Put/Deletes
// are always compacted into a single entry).
int32_t start_index = -1;
current_level_range_ =
MultiGetRange(*range_, range_->begin(), range_->end());
for (auto mget_iter = current_level_range_.begin();
mget_iter != current_level_range_.end(); ++mget_iter) {
struct FilePickerContext& fp_ctx = fp_ctx_array_[mget_iter.index()];
if (curr_level_ == 0) {
// On Level-0, we read through all files to check for overlap.
start_index = 0;
level_contains_keys = true;
} else {
// On Level-n (n>=1), files are sorted. Binary search to find the
// earliest file whose largest key >= ikey. Search left bound and
// right bound are used to narrow the range.
if (fp_ctx.search_left_bound <= fp_ctx.search_right_bound) {
if (fp_ctx.search_right_bound == FileIndexer::kLevelMaxIndex) {
fp_ctx.search_right_bound =
static_cast<int32_t>(curr_file_level_->num_files) - 1;
}
// `search_right_bound_` is an inclusive upper-bound, but since it
// was determined based on user key, it is still possible the lookup
// key falls to the right of `search_right_bound_`'s corresponding
// file. So, pass a limit one higher, which allows us to detect this
// case.
Slice& ikey = mget_iter->ikey;
start_index = FindFileInRange(
*internal_comparator_, *curr_file_level_, ikey,
static_cast<uint32_t>(fp_ctx.search_left_bound),
static_cast<uint32_t>(fp_ctx.search_right_bound) + 1);
if (start_index == fp_ctx.search_right_bound + 1) {
// `ikey_` comes after `search_right_bound_`. The lookup key does
// not exist on this level, so let's skip this level and do a full
// binary search on the next level.
fp_ctx.search_left_bound = 0;
fp_ctx.search_right_bound = FileIndexer::kLevelMaxIndex;
current_level_range_.SkipKey(mget_iter);
continue;
} else {
level_contains_keys = true;
}
} else {
// search_left_bound > search_right_bound, key does not exist in
// this level. Since no comparison is done in this level, it will
// need to search all files in the next level.
fp_ctx.search_left_bound = 0;
fp_ctx.search_right_bound = FileIndexer::kLevelMaxIndex;
current_level_range_.SkipKey(mget_iter);
continue;
}
}
fp_ctx.start_index_in_curr_level = start_index;
fp_ctx.curr_index_in_curr_level = start_index;
}
if (level_contains_keys) {
batch_iter_prev_ = current_level_range_.begin();
upper_key_ = batch_iter_ = current_level_range_.begin();
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
return true;
}
curr_level_++;
}
// curr_level_ = num_levels_. So, no more levels to search.
return false;
}
};
} // anonymous namespace
VersionStorageInfo::~VersionStorageInfo() { delete[] files_; }
Version::~Version() {
assert(refs_ == 0);
// Remove from linked list
prev_->next_ = next_;
next_->prev_ = prev_;
// Drop references to files
for (int level = 0; level < storage_info_.num_levels_; level++) {
for (size_t i = 0; i < storage_info_.files_[level].size(); i++) {
FileMetaData* f = storage_info_.files_[level][i];
assert(f->refs > 0);
f->refs--;
if (f->refs <= 0) {
assert(cfd_ != nullptr);
uint32_t path_id = f->fd.GetPathId();
assert(path_id < cfd_->ioptions()->cf_paths.size());
vset_->obsolete_files_.push_back(
ObsoleteFileInfo(f, cfd_->ioptions()->cf_paths[path_id].path));
}
}
}
}
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
int FindFile(const InternalKeyComparator& icmp,
const LevelFilesBrief& file_level,
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
const Slice& key) {
return FindFileInRange(icmp, file_level, key, 0,
static_cast<uint32_t>(file_level.num_files));
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
}
void DoGenerateLevelFilesBrief(LevelFilesBrief* file_level,
const std::vector<FileMetaData*>& files,
Arena* arena) {
assert(file_level);
assert(arena);
size_t num = files.size();
file_level->num_files = num;
char* mem = arena->AllocateAligned(num * sizeof(FdWithKeyRange));
file_level->files = new (mem)FdWithKeyRange[num];
for (size_t i = 0; i < num; i++) {
Slice smallest_key = files[i]->smallest.Encode();
Slice largest_key = files[i]->largest.Encode();
// Copy key slice to sequential memory
size_t smallest_size = smallest_key.size();
size_t largest_size = largest_key.size();
mem = arena->AllocateAligned(smallest_size + largest_size);
memcpy(mem, smallest_key.data(), smallest_size);
memcpy(mem + smallest_size, largest_key.data(), largest_size);
FdWithKeyRange& f = file_level->files[i];
f.fd = files[i]->fd;
f.file_metadata = files[i];
f.smallest_key = Slice(mem, smallest_size);
f.largest_key = Slice(mem + smallest_size, largest_size);
}
}
static bool AfterFile(const Comparator* ucmp,
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
const Slice* user_key, const FdWithKeyRange* f) {
// nullptr user_key occurs before all keys and is therefore never after *f
return (user_key != nullptr &&
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
ucmp->CompareWithoutTimestamp(*user_key,
ExtractUserKey(f->largest_key)) > 0);
}
static bool BeforeFile(const Comparator* ucmp,
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
const Slice* user_key, const FdWithKeyRange* f) {
// nullptr user_key occurs after all keys and is therefore never before *f
return (user_key != nullptr &&
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
ucmp->CompareWithoutTimestamp(*user_key,
ExtractUserKey(f->smallest_key)) < 0);
}
bool SomeFileOverlapsRange(
const InternalKeyComparator& icmp,
bool disjoint_sorted_files,
const LevelFilesBrief& file_level,
const Slice* smallest_user_key,
const Slice* largest_user_key) {
const Comparator* ucmp = icmp.user_comparator();
if (!disjoint_sorted_files) {
// Need to check against all files
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
for (size_t i = 0; i < file_level.num_files; i++) {
const FdWithKeyRange* f = &(file_level.files[i]);
if (AfterFile(ucmp, smallest_user_key, f) ||
BeforeFile(ucmp, largest_user_key, f)) {
// No overlap
} else {
return true; // Overlap
}
}
return false;
}
// Binary search over file list
uint32_t index = 0;
if (smallest_user_key != nullptr) {
// Find the leftmost possible internal key for smallest_user_key
InternalKey small;
small.SetMinPossibleForUserKey(*smallest_user_key);
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
index = FindFile(icmp, file_level, small.Encode());
}
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
if (index >= file_level.num_files) {
// beginning of range is after all files, so no overlap.
return false;
}
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
return !BeforeFile(ucmp, largest_user_key, &file_level.files[index]);
}
namespace {
class LevelIterator final : public InternalIterator {
public:
// @param read_options Must outlive this iterator.
LevelIterator(TableCache* table_cache, const ReadOptions& read_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options,
const InternalKeyComparator& icomparator,
const LevelFilesBrief* flevel,
const SliceTransform* prefix_extractor, bool should_sample,
HistogramImpl* file_read_hist, TableReaderCaller caller,
bool skip_filters, int level, RangeDelAggregator* range_del_agg,
const std::vector<AtomicCompactionUnitBoundary>*
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
compaction_boundaries = nullptr,
bool allow_unprepared_value = false)
: table_cache_(table_cache),
read_options_(read_options),
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
file_options_(file_options),
icomparator_(icomparator),
user_comparator_(icomparator.user_comparator()),
flevel_(flevel),
prefix_extractor_(prefix_extractor),
file_read_hist_(file_read_hist),
should_sample_(should_sample),
caller_(caller),
skip_filters_(skip_filters),
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
allow_unprepared_value_(allow_unprepared_value),
file_index_(flevel_->num_files),
level_(level),
range_del_agg_(range_del_agg),
pinned_iters_mgr_(nullptr),
compaction_boundaries_(compaction_boundaries) {
// Empty level is not supported.
assert(flevel_ != nullptr && flevel_->num_files > 0);
}
~LevelIterator() override { delete file_iter_.Set(nullptr); }
void Seek(const Slice& target) override;
void SeekForPrev(const Slice& target) override;
void SeekToFirst() override;
void SeekToLast() override;
void Next() final override;
bool NextAndGetResult(IterateResult* result) override;
void Prev() override;
bool Valid() const override { return file_iter_.Valid(); }
Slice key() const override {
assert(Valid());
return file_iter_.key();
}
Slice value() const override {
assert(Valid());
return file_iter_.value();
}
Status status() const override {
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 11:44:14 +02:00
return file_iter_.iter() ? file_iter_.status() : Status::OK();
}
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
bool PrepareValue() override {
return file_iter_.PrepareValue();
}
inline bool MayBeOutOfLowerBound() override {
assert(Valid());
return may_be_out_of_lower_bound_ && file_iter_.MayBeOutOfLowerBound();
}
inline IterBoundCheck UpperBoundCheckResult() override {
if (Valid()) {
return file_iter_.UpperBoundCheckResult();
} else {
return IterBoundCheck::kUnknown;
}
}
void SetPinnedItersMgr(PinnedIteratorsManager* pinned_iters_mgr) override {
pinned_iters_mgr_ = pinned_iters_mgr;
if (file_iter_.iter()) {
file_iter_.SetPinnedItersMgr(pinned_iters_mgr);
}
}
bool IsKeyPinned() const override {
return pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled() &&
file_iter_.iter() && file_iter_.IsKeyPinned();
}
bool IsValuePinned() const override {
return pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled() &&
file_iter_.iter() && file_iter_.IsValuePinned();
}
private:
// Return true if at least one invalid file is seen and skipped.
bool SkipEmptyFileForward();
void SkipEmptyFileBackward();
void SetFileIterator(InternalIterator* iter);
void InitFileIterator(size_t new_file_index);
const Slice& file_smallest_key(size_t file_index) {
assert(file_index < flevel_->num_files);
return flevel_->files[file_index].smallest_key;
}
bool KeyReachedUpperBound(const Slice& internal_key) {
return read_options_.iterate_upper_bound != nullptr &&
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
user_comparator_.CompareWithoutTimestamp(
Iterator with timestamp (#6255) Summary: Preliminary support for iterator with user timestamp. Current implementation does not consider merge operator and reverse iterator. Auto compaction is also disabled in unit tests. Create an iterator with timestamp. ``` ... read_opts.timestamp = &ts; auto* iter = db->NewIterator(read_opts); // target is key without timestamp. for (iter->Seek(target); iter->Valid(); iter->Next()) {} for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {} delete iter; read_opts.timestamp = &ts1; // lower_bound and upper_bound are without timestamp. read_opts.iterate_lower_bound = &lower_bound; read_opts.iterate_upper_bound = &upper_bound; auto* iter1 = db->NewIterator(read_opts); // Do Seek or SeekToFirst() delete iter1; ``` Test plan (dev server) ``` $make check ``` Simple benchmarking (dev server) 1. The overhead introduced by this PR even when timestamp is disabled. key size: 16 bytes value size: 100 bytes Entries: 1000000 Data reside in main memory, and try to stress iterator. Repeated three times on master and this PR. - Seek without next ``` ./db_bench -db=/dev/shm/rocksdbtest-1000 -benchmarks=fillseq,seekrandom -enable_pipelined_write=false -disable_wal=true -format_version=3 ``` master: 159047.0 ops/sec this PR: 158922.3 ops/sec (2% drop in throughput) - Seek and next 10 times ``` ./db_bench -db=/dev/shm/rocksdbtest-1000 -benchmarks=fillseq,seekrandom -enable_pipelined_write=false -disable_wal=true -format_version=3 -seek_nexts=10 ``` master: 109539.3 ops/sec this PR: 107519.7 ops/sec (2% drop in throughput) Pull Request resolved: https://github.com/facebook/rocksdb/pull/6255 Differential Revision: D19438227 Pulled By: riversand963 fbshipit-source-id: b66b4979486f8474619f4aa6bdd88598870b0746
2020-03-07 01:21:03 +01:00
ExtractUserKey(internal_key), /*a_has_ts=*/true,
*read_options_.iterate_upper_bound, /*b_has_ts=*/false) >= 0;
}
InternalIterator* NewFileIterator() {
assert(file_index_ < flevel_->num_files);
auto file_meta = flevel_->files[file_index_];
if (should_sample_) {
sample_file_read_inc(file_meta.file_metadata);
}
const InternalKey* smallest_compaction_key = nullptr;
const InternalKey* largest_compaction_key = nullptr;
if (compaction_boundaries_ != nullptr) {
smallest_compaction_key = (*compaction_boundaries_)[file_index_].smallest;
largest_compaction_key = (*compaction_boundaries_)[file_index_].largest;
}
CheckMayBeOutOfLowerBound();
return table_cache_->NewIterator(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
read_options_, file_options_, icomparator_, *file_meta.file_metadata,
range_del_agg_, prefix_extractor_,
nullptr /* don't need reference to table */, file_read_hist_, caller_,
/*arena=*/nullptr, skip_filters_, level_,
/*max_file_size_for_l0_meta_pin=*/0, smallest_compaction_key,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
largest_compaction_key, allow_unprepared_value_);
}
// Check if current file being fully within iterate_lower_bound.
//
// Note MyRocks may update iterate bounds between seek. To workaround it,
// we need to check and update may_be_out_of_lower_bound_ accordingly.
void CheckMayBeOutOfLowerBound() {
if (read_options_.iterate_lower_bound != nullptr &&
file_index_ < flevel_->num_files) {
may_be_out_of_lower_bound_ =
user_comparator_.CompareWithoutTimestamp(
ExtractUserKey(file_smallest_key(file_index_)), /*a_has_ts=*/true,
*read_options_.iterate_lower_bound, /*b_has_ts=*/false) < 0;
}
}
TableCache* table_cache_;
const ReadOptions& read_options_;
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options_;
const InternalKeyComparator& icomparator_;
const UserComparatorWrapper user_comparator_;
const LevelFilesBrief* flevel_;
mutable FileDescriptor current_value_;
// `prefix_extractor_` may be non-null even for total order seek. Checking
// this variable is not the right way to identify whether prefix iterator
// is used.
const SliceTransform* prefix_extractor_;
HistogramImpl* file_read_hist_;
bool should_sample_;
TableReaderCaller caller_;
bool skip_filters_;
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
bool allow_unprepared_value_;
bool may_be_out_of_lower_bound_ = true;
size_t file_index_;
int level_;
RangeDelAggregator* range_del_agg_;
IteratorWrapper file_iter_; // May be nullptr
PinnedIteratorsManager* pinned_iters_mgr_;
// To be propagated to RangeDelAggregator in order to safely truncate range
// tombstones.
const std::vector<AtomicCompactionUnitBoundary>* compaction_boundaries_;
};
void LevelIterator::Seek(const Slice& target) {
// Check whether the seek key fall under the same file
bool need_to_reseek = true;
if (file_iter_.iter() != nullptr && file_index_ < flevel_->num_files) {
const FdWithKeyRange& cur_file = flevel_->files[file_index_];
if (icomparator_.InternalKeyComparator::Compare(
target, cur_file.largest_key) <= 0 &&
icomparator_.InternalKeyComparator::Compare(
target, cur_file.smallest_key) >= 0) {
need_to_reseek = false;
assert(static_cast<size_t>(FindFile(icomparator_, *flevel_, target)) ==
file_index_);
}
}
if (need_to_reseek) {
TEST_SYNC_POINT("LevelIterator::Seek:BeforeFindFile");
size_t new_file_index = FindFile(icomparator_, *flevel_, target);
InitFileIterator(new_file_index);
}
if (file_iter_.iter() != nullptr) {
file_iter_.Seek(target);
}
if (SkipEmptyFileForward() && prefix_extractor_ != nullptr &&
!read_options_.total_order_seek && !read_options_.auto_prefix_mode &&
file_iter_.iter() != nullptr && file_iter_.Valid()) {
// We've skipped the file we initially positioned to. In the prefix
// seek case, it is likely that the file is skipped because of
// prefix bloom or hash, where more keys are skipped. We then check
// the current key and invalidate the iterator if the prefix is
// already passed.
// When doing prefix iterator seek, when keys for one prefix have
// been exhausted, it can jump to any key that is larger. Here we are
// enforcing a stricter contract than that, in order to make it easier for
// higher layers (merging and DB iterator) to reason the correctness:
// 1. Within the prefix, the result should be accurate.
// 2. If keys for the prefix is exhausted, it is either positioned to the
// next key after the prefix, or make the iterator invalid.
// A side benefit will be that it invalidates the iterator earlier so that
// the upper level merging iterator can merge fewer child iterators.
size_t ts_sz = user_comparator_.timestamp_size();
Slice target_user_key_without_ts =
ExtractUserKeyAndStripTimestamp(target, ts_sz);
Slice file_user_key_without_ts =
ExtractUserKeyAndStripTimestamp(file_iter_.key(), ts_sz);
if (prefix_extractor_->InDomain(target_user_key_without_ts) &&
(!prefix_extractor_->InDomain(file_user_key_without_ts) ||
user_comparator_.CompareWithoutTimestamp(
prefix_extractor_->Transform(target_user_key_without_ts), false,
prefix_extractor_->Transform(file_user_key_without_ts),
false) != 0)) {
SetFileIterator(nullptr);
}
}
CheckMayBeOutOfLowerBound();
}
void LevelIterator::SeekForPrev(const Slice& target) {
size_t new_file_index = FindFile(icomparator_, *flevel_, target);
if (new_file_index >= flevel_->num_files) {
new_file_index = flevel_->num_files - 1;
}
InitFileIterator(new_file_index);
if (file_iter_.iter() != nullptr) {
file_iter_.SeekForPrev(target);
SkipEmptyFileBackward();
}
CheckMayBeOutOfLowerBound();
}
void LevelIterator::SeekToFirst() {
InitFileIterator(0);
if (file_iter_.iter() != nullptr) {
file_iter_.SeekToFirst();
}
SkipEmptyFileForward();
CheckMayBeOutOfLowerBound();
}
void LevelIterator::SeekToLast() {
InitFileIterator(flevel_->num_files - 1);
if (file_iter_.iter() != nullptr) {
file_iter_.SeekToLast();
}
SkipEmptyFileBackward();
CheckMayBeOutOfLowerBound();
}
void LevelIterator::Next() {
assert(Valid());
file_iter_.Next();
SkipEmptyFileForward();
}
bool LevelIterator::NextAndGetResult(IterateResult* result) {
assert(Valid());
bool is_valid = file_iter_.NextAndGetResult(result);
if (!is_valid) {
SkipEmptyFileForward();
is_valid = Valid();
if (is_valid) {
result->key = key();
result->bound_check_result = file_iter_.UpperBoundCheckResult();
// Ideally, we should return the real file_iter_.value_prepared but the
// information is not here. It would casue an extra PrepareValue()
// for the first key of a file.
result->value_prepared = !allow_unprepared_value_;
}
}
return is_valid;
}
void LevelIterator::Prev() {
assert(Valid());
file_iter_.Prev();
SkipEmptyFileBackward();
}
bool LevelIterator::SkipEmptyFileForward() {
bool seen_empty_file = false;
while (file_iter_.iter() == nullptr ||
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 11:44:14 +02:00
(!file_iter_.Valid() && file_iter_.status().ok() &&
file_iter_.iter()->UpperBoundCheckResult() !=
IterBoundCheck::kOutOfBound)) {
seen_empty_file = true;
// Move to next file
if (file_index_ >= flevel_->num_files - 1) {
// Already at the last file
SetFileIterator(nullptr);
break;
}
if (KeyReachedUpperBound(file_smallest_key(file_index_ + 1))) {
SetFileIterator(nullptr);
break;
}
InitFileIterator(file_index_ + 1);
if (file_iter_.iter() != nullptr) {
file_iter_.SeekToFirst();
}
}
return seen_empty_file;
}
void LevelIterator::SkipEmptyFileBackward() {
while (file_iter_.iter() == nullptr ||
Change and clarify the relationship between Valid(), status() and Seek*() for all iterators. Also fix some bugs Summary: Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are: * If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted. * When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not. However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours). This PR changes the convention to: * If status() is not ok, Valid() always returns false. * Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.) This does sacrifice the two use cases listed above, but siying said it's ok. Overview of the changes: * A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario. * Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed. * A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator. To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types: Iterators that didn't need changes: * status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator. * Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator. Iterators with changes (see inline comments for details): * DBIter - an overhaul: - It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back. - It had a few code paths silently discarding subiterator's status. The stress test caught a few. - The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example. - Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption. - It used to not reset status on seek for some types of errors. - Some simplifications and better comments. - Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer. * MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status. * ForwardIterator - changed to the new convention, also slightly simplified. * ForwardLevelIterator - fixed some bugs and simplified. * LevelIterator - simplified. * TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_. * BlockBasedTableIterator - minor changes. * BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid. * PlainTableIterator - some seeks used to not reset status. * CuckooTableIterator - tiny code cleanup. * ManagedIterator - fixed some bugs. * BaseDeltaIterator - changed to the new convention and fixed a bug. * BlobDBIterator - seeks used to not reset status. * KeyConvertingIterator - some small change. Closes https://github.com/facebook/rocksdb/pull/3810 Differential Revision: D7888019 Pulled By: al13n321 fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
2018-05-17 11:44:14 +02:00
(!file_iter_.Valid() && file_iter_.status().ok())) {
// Move to previous file
if (file_index_ == 0) {
// Already the first file
SetFileIterator(nullptr);
return;
}
InitFileIterator(file_index_ - 1);
if (file_iter_.iter() != nullptr) {
file_iter_.SeekToLast();
}
}
}
void LevelIterator::SetFileIterator(InternalIterator* iter) {
if (pinned_iters_mgr_ && iter) {
iter->SetPinnedItersMgr(pinned_iters_mgr_);
}
InternalIterator* old_iter = file_iter_.Set(iter);
if (pinned_iters_mgr_ && pinned_iters_mgr_->PinningEnabled()) {
pinned_iters_mgr_->PinIterator(old_iter);
} else {
delete old_iter;
}
}
void LevelIterator::InitFileIterator(size_t new_file_index) {
if (new_file_index >= flevel_->num_files) {
file_index_ = new_file_index;
SetFileIterator(nullptr);
return;
} else {
// If the file iterator shows incomplete, we try it again if users seek
// to the same file, as this time we may go to a different data block
// which is cached in block cache.
//
if (file_iter_.iter() != nullptr && !file_iter_.status().IsIncomplete() &&
new_file_index == file_index_) {
// file_iter_ is already constructed with this iterator, so
// no need to change anything
} else {
file_index_ = new_file_index;
InternalIterator* iter = NewFileIterator();
SetFileIterator(iter);
}
}
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
} // anonymous namespace
Status Version::GetTableProperties(std::shared_ptr<const TableProperties>* tp,
const FileMetaData* file_meta,
const std::string* fname) const {
auto table_cache = cfd_->table_cache();
auto ioptions = cfd_->ioptions();
Status s = table_cache->GetTableProperties(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
file_options_, cfd_->internal_comparator(), file_meta->fd, tp,
mutable_cf_options_.prefix_extractor.get(), true /* no io */);
if (s.ok()) {
return s;
}
// We only ignore error type `Incomplete` since it's by design that we
// disallow table when it's not in table cache.
if (!s.IsIncomplete()) {
return s;
}
// 2. Table is not present in table cache, we'll read the table properties
// directly from the properties block in the file.
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
std::unique_ptr<FSRandomAccessFile> file;
std::string file_name;
if (fname != nullptr) {
file_name = *fname;
} else {
file_name =
TableFileName(ioptions->cf_paths, file_meta->fd.GetNumber(),
file_meta->fd.GetPathId());
}
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
s = ioptions->fs->NewRandomAccessFile(file_name, file_options_, &file,
nullptr);
if (!s.ok()) {
return s;
}
TableProperties* raw_table_properties;
// By setting the magic number to kInvalidTableMagicNumber, we can by
// pass the magic number check in the footer.
std::unique_ptr<RandomAccessFileReader> file_reader(
new RandomAccessFileReader(
std::move(file), file_name, nullptr /* env */, io_tracer_,
nullptr /* stats */, 0 /* hist_type */, nullptr /* file_read_hist */,
nullptr /* rate_limiter */, ioptions->listeners));
s = ReadTableProperties(
file_reader.get(), file_meta->fd.GetFileSize(),
Footer::kInvalidTableMagicNumber /* table's magic number */, *ioptions,
&raw_table_properties, false /* compression_type_missing */);
if (!s.ok()) {
return s;
}
RecordTick(ioptions->stats, NUMBER_DIRECT_LOAD_TABLE_PROPERTIES);
*tp = std::shared_ptr<const TableProperties>(raw_table_properties);
return s;
}
Status Version::GetPropertiesOfAllTables(TablePropertiesCollection* props) {
Status s;
for (int level = 0; level < storage_info_.num_levels_; level++) {
s = GetPropertiesOfAllTables(props, level);
if (!s.ok()) {
return s;
}
}
return Status::OK();
}
Status Version::TablesRangeTombstoneSummary(int max_entries_to_print,
std::string* out_str) {
if (max_entries_to_print <= 0) {
return Status::OK();
}
int num_entries_left = max_entries_to_print;
std::stringstream ss;
for (int level = 0; level < storage_info_.num_levels_; level++) {
for (const auto& file_meta : storage_info_.files_[level]) {
auto fname =
TableFileName(cfd_->ioptions()->cf_paths, file_meta->fd.GetNumber(),
file_meta->fd.GetPathId());
ss << "=== file : " << fname << " ===\n";
TableCache* table_cache = cfd_->table_cache();
std::unique_ptr<FragmentedRangeTombstoneIterator> tombstone_iter;
Status s = table_cache->GetRangeTombstoneIterator(
ReadOptions(), cfd_->internal_comparator(), *file_meta,
&tombstone_iter);
if (!s.ok()) {
return s;
}
if (tombstone_iter) {
tombstone_iter->SeekToFirst();
while (tombstone_iter->Valid() && num_entries_left > 0) {
ss << "start: " << tombstone_iter->start_key().ToString(true)
<< " end: " << tombstone_iter->end_key().ToString(true)
<< " seq: " << tombstone_iter->seq() << '\n';
tombstone_iter->Next();
num_entries_left--;
}
if (num_entries_left <= 0) {
break;
}
}
}
if (num_entries_left <= 0) {
break;
}
}
assert(num_entries_left >= 0);
if (num_entries_left <= 0) {
ss << "(results may not be complete)\n";
}
*out_str = ss.str();
return Status::OK();
}
Status Version::GetPropertiesOfAllTables(TablePropertiesCollection* props,
int level) {
for (const auto& file_meta : storage_info_.files_[level]) {
auto fname =
TableFileName(cfd_->ioptions()->cf_paths, file_meta->fd.GetNumber(),
file_meta->fd.GetPathId());
// 1. If the table is already present in table cache, load table
// properties from there.
std::shared_ptr<const TableProperties> table_properties;
Status s = GetTableProperties(&table_properties, file_meta, &fname);
if (s.ok()) {
props->insert({fname, table_properties});
} else {
return s;
}
}
return Status::OK();
}
Status Version::GetPropertiesOfTablesInRange(
const Range* range, std::size_t n, TablePropertiesCollection* props) const {
for (int level = 0; level < storage_info_.num_non_empty_levels(); level++) {
for (decltype(n) i = 0; i < n; i++) {
// Convert user_key into a corresponding internal key.
InternalKey k1(range[i].start, kMaxSequenceNumber, kValueTypeForSeek);
InternalKey k2(range[i].limit, kMaxSequenceNumber, kValueTypeForSeek);
std::vector<FileMetaData*> files;
storage_info_.GetOverlappingInputs(level, &k1, &k2, &files, -1, nullptr,
false);
for (const auto& file_meta : files) {
auto fname =
TableFileName(cfd_->ioptions()->cf_paths,
file_meta->fd.GetNumber(), file_meta->fd.GetPathId());
if (props->count(fname) == 0) {
// 1. If the table is already present in table cache, load table
// properties from there.
std::shared_ptr<const TableProperties> table_properties;
Status s = GetTableProperties(&table_properties, file_meta, &fname);
if (s.ok()) {
props->insert({fname, table_properties});
} else {
return s;
}
}
}
}
}
return Status::OK();
}
Status Version::GetAggregatedTableProperties(
std::shared_ptr<const TableProperties>* tp, int level) {
TablePropertiesCollection props;
Status s;
if (level < 0) {
s = GetPropertiesOfAllTables(&props);
} else {
s = GetPropertiesOfAllTables(&props, level);
}
if (!s.ok()) {
return s;
}
auto* new_tp = new TableProperties();
for (const auto& item : props) {
new_tp->Add(*item.second);
}
tp->reset(new_tp);
return Status::OK();
}
size_t Version::GetMemoryUsageByTableReaders() {
size_t total_usage = 0;
for (auto& file_level : storage_info_.level_files_brief_) {
for (size_t i = 0; i < file_level.num_files; i++) {
total_usage += cfd_->table_cache()->GetMemoryUsageByTableReader(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
file_options_, cfd_->internal_comparator(), file_level.files[i].fd,
mutable_cf_options_.prefix_extractor.get());
}
}
return total_usage;
}
void Version::GetColumnFamilyMetaData(ColumnFamilyMetaData* cf_meta) {
assert(cf_meta);
assert(cfd_);
cf_meta->name = cfd_->GetName();
cf_meta->size = 0;
cf_meta->file_count = 0;
cf_meta->levels.clear();
cf_meta->blob_file_size = 0;
cf_meta->blob_file_count = 0;
cf_meta->blob_files.clear();
auto* ioptions = cfd_->ioptions();
auto* vstorage = storage_info();
for (int level = 0; level < cfd_->NumberLevels(); level++) {
uint64_t level_size = 0;
cf_meta->file_count += vstorage->LevelFiles(level).size();
std::vector<SstFileMetaData> files;
for (const auto& file : vstorage->LevelFiles(level)) {
uint32_t path_id = file->fd.GetPathId();
std::string file_path;
if (path_id < ioptions->cf_paths.size()) {
file_path = ioptions->cf_paths[path_id].path;
} else {
assert(!ioptions->cf_paths.empty());
file_path = ioptions->cf_paths.back().path;
}
const uint64_t file_number = file->fd.GetNumber();
files.emplace_back(
MakeTableFileName("", file_number), file_number, file_path,
static_cast<size_t>(file->fd.GetFileSize()), file->fd.smallest_seqno,
file->fd.largest_seqno, file->smallest.user_key().ToString(),
file->largest.user_key().ToString(),
file->stats.num_reads_sampled.load(std::memory_order_relaxed),
file->being_compacted, file->temperature,
file->oldest_blob_file_number, file->TryGetOldestAncesterTime(),
file->TryGetFileCreationTime(), file->file_checksum,
file->file_checksum_func_name);
files.back().num_entries = file->num_entries;
files.back().num_deletions = file->num_deletions;
level_size += file->fd.GetFileSize();
}
cf_meta->levels.emplace_back(
level, level_size, std::move(files));
cf_meta->size += level_size;
}
for (const auto& iter : vstorage->GetBlobFiles()) {
const auto meta = iter.second.get();
cf_meta->blob_files.emplace_back(
meta->GetBlobFileNumber(), BlobFileName("", meta->GetBlobFileNumber()),
ioptions->cf_paths.front().path, meta->GetBlobFileSize(),
meta->GetTotalBlobCount(), meta->GetTotalBlobBytes(),
meta->GetGarbageBlobCount(), meta->GetGarbageBlobBytes(),
meta->GetChecksumMethod(), meta->GetChecksumValue());
cf_meta->blob_file_count++;
cf_meta->blob_file_size += meta->GetBlobFileSize();
}
}
uint64_t Version::GetSstFilesSize() {
uint64_t sst_files_size = 0;
for (int level = 0; level < storage_info_.num_levels_; level++) {
for (const auto& file_meta : storage_info_.LevelFiles(level)) {
sst_files_size += file_meta->fd.GetFileSize();
}
}
return sst_files_size;
}
void Version::GetCreationTimeOfOldestFile(uint64_t* creation_time) {
uint64_t oldest_time = port::kMaxUint64;
for (int level = 0; level < storage_info_.num_non_empty_levels_; level++) {
for (FileMetaData* meta : storage_info_.LevelFiles(level)) {
assert(meta->fd.table_reader != nullptr);
uint64_t file_creation_time = meta->TryGetFileCreationTime();
if (file_creation_time == kUnknownFileCreationTime) {
*creation_time = 0;
return;
}
if (file_creation_time < oldest_time) {
oldest_time = file_creation_time;
}
}
}
*creation_time = oldest_time;
}
uint64_t VersionStorageInfo::GetEstimatedActiveKeys() const {
// Estimation will be inaccurate when:
// (1) there exist merge keys
// (2) keys are directly overwritten
// (3) deletion on non-existing keys
// (4) low number of samples
2015-12-07 19:51:08 +01:00
if (current_num_samples_ == 0) {
return 0;
}
2015-12-07 19:51:08 +01:00
if (current_num_non_deletions_ <= current_num_deletions_) {
return 0;
}
2015-12-07 19:51:08 +01:00
uint64_t est = current_num_non_deletions_ - current_num_deletions_;
uint64_t file_count = 0;
for (int level = 0; level < num_levels_; ++level) {
file_count += files_[level].size();
}
2015-12-07 19:51:08 +01:00
if (current_num_samples_ < file_count) {
// casting to avoid overflowing
return
static_cast<uint64_t>(
2015-12-12 02:33:40 +01:00
(est * static_cast<double>(file_count) / current_num_samples_)
);
} else {
return est;
}
}
double VersionStorageInfo::GetEstimatedCompressionRatioAtLevel(
int level) const {
assert(level < num_levels_);
uint64_t sum_file_size_bytes = 0;
uint64_t sum_data_size_bytes = 0;
for (auto* file_meta : files_[level]) {
sum_file_size_bytes += file_meta->fd.GetFileSize();
sum_data_size_bytes += file_meta->raw_key_size + file_meta->raw_value_size;
}
if (sum_file_size_bytes == 0) {
return -1.0;
}
return static_cast<double>(sum_data_size_bytes) / sum_file_size_bytes;
}
void Version::AddIterators(const ReadOptions& read_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& soptions,
MergeIteratorBuilder* merge_iter_builder,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
RangeDelAggregator* range_del_agg,
bool allow_unprepared_value) {
assert(storage_info_.finalized_);
for (int level = 0; level < storage_info_.num_non_empty_levels(); level++) {
AddIteratorsForLevel(read_options, soptions, merge_iter_builder, level,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
range_del_agg, allow_unprepared_value);
}
}
void Version::AddIteratorsForLevel(const ReadOptions& read_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& soptions,
MergeIteratorBuilder* merge_iter_builder,
int level,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
RangeDelAggregator* range_del_agg,
bool allow_unprepared_value) {
assert(storage_info_.finalized_);
if (level >= storage_info_.num_non_empty_levels()) {
// This is an empty level
return;
} else if (storage_info_.LevelFilesBrief(level).num_files == 0) {
// No files in this level
return;
}
bool should_sample = should_sample_file_read();
auto* arena = merge_iter_builder->GetArena();
if (level == 0) {
// Merge all level zero files together since they may overlap
for (size_t i = 0; i < storage_info_.LevelFilesBrief(0).num_files; i++) {
const auto& file = storage_info_.LevelFilesBrief(0).files[i];
merge_iter_builder->AddIterator(cfd_->table_cache()->NewIterator(
read_options, soptions, cfd_->internal_comparator(),
*file.file_metadata, range_del_agg,
mutable_cf_options_.prefix_extractor.get(), nullptr,
cfd_->internal_stats()->GetFileReadHist(0),
TableReaderCaller::kUserIterator, arena,
/*skip_filters=*/false, /*level=*/0, max_file_size_for_l0_meta_pin_,
/*smallest_compaction_key=*/nullptr,
/*largest_compaction_key=*/nullptr, allow_unprepared_value));
}
if (should_sample) {
// Count ones for every L0 files. This is done per iterator creation
// rather than Seek(), while files in other levels are recored per seek.
// If users execute one range query per iterator, there may be some
// discrepancy here.
for (FileMetaData* meta : storage_info_.LevelFiles(0)) {
sample_file_read_inc(meta);
}
}
} else if (storage_info_.LevelFilesBrief(level).num_files > 0) {
// For levels > 0, we can use a concatenating iterator that sequentially
// walks through the non-overlapping files in the level, opening them
// lazily.
auto* mem = arena->AllocateAligned(sizeof(LevelIterator));
merge_iter_builder->AddIterator(new (mem) LevelIterator(
cfd_->table_cache(), read_options, soptions,
cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level),
mutable_cf_options_.prefix_extractor.get(), should_sample_file_read(),
cfd_->internal_stats()->GetFileReadHist(level),
TableReaderCaller::kUserIterator, IsFilterSkipped(level), level,
range_del_agg,
/*compaction_boundaries=*/nullptr, allow_unprepared_value));
}
}
Status Version::OverlapWithLevelIterator(const ReadOptions& read_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options,
const Slice& smallest_user_key,
const Slice& largest_user_key,
int level, bool* overlap) {
assert(storage_info_.finalized_);
auto icmp = cfd_->internal_comparator();
auto ucmp = icmp.user_comparator();
Arena arena;
Status status;
ReadRangeDelAggregator range_del_agg(&icmp,
kMaxSequenceNumber /* upper_bound */);
*overlap = false;
if (level == 0) {
for (size_t i = 0; i < storage_info_.LevelFilesBrief(0).num_files; i++) {
const auto file = &storage_info_.LevelFilesBrief(0).files[i];
if (AfterFile(ucmp, &smallest_user_key, file) ||
BeforeFile(ucmp, &largest_user_key, file)) {
continue;
}
ScopedArenaIterator iter(cfd_->table_cache()->NewIterator(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
read_options, file_options, cfd_->internal_comparator(),
*file->file_metadata, &range_del_agg,
mutable_cf_options_.prefix_extractor.get(), nullptr,
cfd_->internal_stats()->GetFileReadHist(0),
TableReaderCaller::kUserIterator, &arena,
/*skip_filters=*/false, /*level=*/0, max_file_size_for_l0_meta_pin_,
/*smallest_compaction_key=*/nullptr,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
/*largest_compaction_key=*/nullptr,
/*allow_unprepared_value=*/false));
status = OverlapWithIterator(
ucmp, smallest_user_key, largest_user_key, iter.get(), overlap);
if (!status.ok() || *overlap) {
break;
}
}
} else if (storage_info_.LevelFilesBrief(level).num_files > 0) {
auto mem = arena.AllocateAligned(sizeof(LevelIterator));
ScopedArenaIterator iter(new (mem) LevelIterator(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
cfd_->table_cache(), read_options, file_options,
cfd_->internal_comparator(), &storage_info_.LevelFilesBrief(level),
mutable_cf_options_.prefix_extractor.get(), should_sample_file_read(),
cfd_->internal_stats()->GetFileReadHist(level),
TableReaderCaller::kUserIterator, IsFilterSkipped(level), level,
&range_del_agg));
status = OverlapWithIterator(
ucmp, smallest_user_key, largest_user_key, iter.get(), overlap);
}
if (status.ok() && *overlap == false &&
range_del_agg.IsRangeOverlapped(smallest_user_key, largest_user_key)) {
*overlap = true;
}
return status;
}
VersionStorageInfo::VersionStorageInfo(
const InternalKeyComparator* internal_comparator,
const Comparator* user_comparator, int levels,
CompactionStyle compaction_style, VersionStorageInfo* ref_vstorage,
bool _force_consistency_checks)
: internal_comparator_(internal_comparator),
user_comparator_(user_comparator),
// cfd is nullptr if Version is dummy
num_levels_(levels),
num_non_empty_levels_(0),
file_indexer_(user_comparator),
compaction_style_(compaction_style),
files_(new std::vector<FileMetaData*>[num_levels_]),
base_level_(num_levels_ == 1 ? -1 : 1),
level_multiplier_(0.0),
files_by_compaction_pri_(num_levels_),
Allowing L0 -> L1 trivial move on sorted data Summary: This diff updates the logic of how we do trivial move, now trivial move can run on any number of files in input level as long as they are not overlapping The conditions for trivial move have been updated Introduced conditions: - Trivial move cannot happen if we have a compaction filter (except if the compaction is not manual) - Input level files cannot be overlapping Removed conditions: - Trivial move only run when the compaction is not manual - Input level should can contain only 1 file More context on what tests failed because of Trivial move ``` DBTest.CompactionsGenerateMultipleFiles This test is expecting compaction on a file in L0 to generate multiple files in L1, this test will fail with trivial move because we end up with one file in L1 ``` ``` DBTest.NoSpaceCompactRange This test expect compaction to fail when we force environment to report running out of space, of course this is not valid in trivial move situation because trivial move does not need any extra space, and did not check for that ``` ``` DBTest.DropWrites Similar to DBTest.NoSpaceCompactRange ``` ``` DBTest.DeleteObsoleteFilesPendingOutputs This test expect that a file in L2 is deleted after it's moved to L3, this is not valid with trivial move because although the file was moved it is now used by L3 ``` ``` CuckooTableDBTest.CompactionIntoMultipleFiles Same as DBTest.CompactionsGenerateMultipleFiles ``` This diff is based on a work by @sdong https://reviews.facebook.net/D34149 Test Plan: make -j64 check Reviewers: rven, sdong, igor Reviewed By: igor Subscribers: yhchiang, ott, march, dhruba, sdong Differential Revision: https://reviews.facebook.net/D34797
2015-06-05 01:51:25 +02:00
level0_non_overlapping_(false),
next_file_to_compact_by_size_(num_levels_),
compaction_score_(num_levels_),
compaction_level_(num_levels_),
l0_delay_trigger_count_(0),
accumulated_file_size_(0),
accumulated_raw_key_size_(0),
accumulated_raw_value_size_(0),
accumulated_num_non_deletions_(0),
accumulated_num_deletions_(0),
2015-12-07 19:51:08 +01:00
current_num_non_deletions_(0),
current_num_deletions_(0),
current_num_samples_(0),
estimated_compaction_needed_bytes_(0),
finalized_(false),
force_consistency_checks_(_force_consistency_checks) {
if (ref_vstorage != nullptr) {
accumulated_file_size_ = ref_vstorage->accumulated_file_size_;
accumulated_raw_key_size_ = ref_vstorage->accumulated_raw_key_size_;
accumulated_raw_value_size_ = ref_vstorage->accumulated_raw_value_size_;
accumulated_num_non_deletions_ =
ref_vstorage->accumulated_num_non_deletions_;
accumulated_num_deletions_ = ref_vstorage->accumulated_num_deletions_;
2015-12-07 19:51:08 +01:00
current_num_non_deletions_ = ref_vstorage->current_num_non_deletions_;
current_num_deletions_ = ref_vstorage->current_num_deletions_;
current_num_samples_ = ref_vstorage->current_num_samples_;
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
oldest_snapshot_seqnum_ = ref_vstorage->oldest_snapshot_seqnum_;
}
}
Version::Version(ColumnFamilyData* column_family_data, VersionSet* vset,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_opt,
const MutableCFOptions mutable_cf_options,
const std::shared_ptr<IOTracer>& io_tracer,
uint64_t version_number)
: env_(vset->env_),
clock_(vset->clock_),
cfd_(column_family_data),
info_log_((cfd_ == nullptr) ? nullptr : cfd_->ioptions()->logger),
db_statistics_((cfd_ == nullptr) ? nullptr : cfd_->ioptions()->stats),
table_cache_((cfd_ == nullptr) ? nullptr : cfd_->table_cache()),
blob_file_cache_(cfd_ ? cfd_->blob_file_cache() : nullptr),
merge_operator_(
(cfd_ == nullptr) ? nullptr : cfd_->ioptions()->merge_operator.get()),
storage_info_(
(cfd_ == nullptr) ? nullptr : &cfd_->internal_comparator(),
(cfd_ == nullptr) ? nullptr : cfd_->user_comparator(),
cfd_ == nullptr ? 0 : cfd_->NumberLevels(),
cfd_ == nullptr ? kCompactionStyleLevel
: cfd_->ioptions()->compaction_style,
(cfd_ == nullptr || cfd_->current() == nullptr)
? nullptr
: cfd_->current()->storage_info(),
cfd_ == nullptr ? false : cfd_->ioptions()->force_consistency_checks),
vset_(vset),
next_(this),
prev_(this),
refs_(0),
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
file_options_(file_opt),
mutable_cf_options_(mutable_cf_options),
max_file_size_for_l0_meta_pin_(
MaxFileSizeForL0MetaPin(mutable_cf_options_)),
version_number_(version_number),
io_tracer_(io_tracer) {}
Status Version::GetBlob(const ReadOptions& read_options, const Slice& user_key,
const Slice& blob_index_slice, PinnableSlice* value,
uint64_t* bytes_read) const {
if (read_options.read_tier == kBlockCacheTier) {
return Status::Incomplete("Cannot read blob: no disk I/O allowed");
}
BlobIndex blob_index;
{
Status s = blob_index.DecodeFrom(blob_index_slice);
if (!s.ok()) {
return s;
}
}
return GetBlob(read_options, user_key, blob_index, value, bytes_read);
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
2020-11-24 06:07:01 +01:00
}
Status Version::GetBlob(const ReadOptions& read_options, const Slice& user_key,
const BlobIndex& blob_index, PinnableSlice* value,
uint64_t* bytes_read) const {
assert(value);
if (blob_index.HasTTL() || blob_index.IsInlined()) {
return Status::Corruption("Unexpected TTL/inlined blob index");
}
const auto& blob_files = storage_info_.GetBlobFiles();
const uint64_t blob_file_number = blob_index.file_number();
const auto it = blob_files.find(blob_file_number);
if (it == blob_files.end()) {
return Status::Corruption("Invalid blob file number");
}
CacheHandleGuard<BlobFileReader> blob_file_reader;
{
assert(blob_file_cache_);
const Status s = blob_file_cache_->GetBlobFileReader(blob_file_number,
&blob_file_reader);
if (!s.ok()) {
return s;
}
}
assert(blob_file_reader.GetValue());
const Status s = blob_file_reader.GetValue()->GetBlob(
read_options, user_key, blob_index.offset(), blob_index.size(),
blob_index.compression(), value, bytes_read);
return s;
}
void Version::Get(const ReadOptions& read_options, const LookupKey& k,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
PinnableSlice* value, std::string* timestamp, Status* status,
MergeContext* merge_context,
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
2018-10-24 21:29:29 +02:00
SequenceNumber* max_covering_tombstone_seq, bool* value_found,
bool* key_exists, SequenceNumber* seq, ReadCallback* callback,
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
bool* is_blob, bool do_merge) {
Slice ikey = k.internal_key();
Slice user_key = k.user_key();
assert(status->ok() || status->IsMergeInProgress());
if (key_exists != nullptr) {
// will falsify below if not found
*key_exists = true;
}
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
2016-07-20 18:49:03 +02:00
PinnedIteratorsManager pinned_iters_mgr;
uint64_t tracing_get_id = BlockCacheTraceHelper::kReservedGetId;
if (vset_ && vset_->block_cache_tracer_ &&
vset_->block_cache_tracer_->is_tracing_enabled()) {
tracing_get_id = vset_->block_cache_tracer_->NextGetId();
}
// Note: the old StackableDB-based BlobDB passes in
// GetImplOptions::is_blob_index; for the integrated BlobDB implementation, we
// need to provide it here.
bool is_blob_index = false;
bool* const is_blob_to_use = is_blob ? is_blob : &is_blob_index;
BlobFetcher blob_fetcher(this, read_options);
GetContext get_context(
user_comparator(), merge_operator_, info_log_, db_statistics_,
status->ok() ? GetContext::kNotFound : GetContext::kMerge, user_key,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
do_merge ? value : nullptr, do_merge ? timestamp : nullptr, value_found,
merge_context, do_merge, max_covering_tombstone_seq, clock_, seq,
merge_operator_ ? &pinned_iters_mgr : nullptr, callback, is_blob_to_use,
tracing_get_id, &blob_fetcher);
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
2016-07-20 18:49:03 +02:00
// Pin blocks that we read to hold merge operands
if (merge_operator_) {
pinned_iters_mgr.StartPinning();
}
FilePicker fp(
storage_info_.files_, user_key, ikey, &storage_info_.level_files_brief_,
storage_info_.num_non_empty_levels_, &storage_info_.file_indexer_,
user_comparator(), internal_comparator());
FdWithKeyRange* f = fp.GetNextFile();
while (f != nullptr) {
Cache fragmented range tombstones in BlockBasedTableReader (#4493) Summary: This allows tombstone fragmenting to only be performed when the table is opened, and cached for subsequent accesses. On the same DB used in #4449, running `readrandom` results in the following: ``` readrandom : 0.983 micros/op 1017076 ops/sec; 78.3 MB/s (63103 of 100000 found) ``` Now that Get performance in the presence of range tombstones is reasonable, I also compared the performance between a DB with range tombstones, "expanded" range tombstones (several point tombstones that cover the same keys the equivalent range tombstone would cover, a common workaround for DeleteRange), and no range tombstones. The created DBs had 5 million keys each, and DeleteRange was called at regular intervals (depending on the total number of range tombstones being written) after 4.5 million Puts. The table below summarizes the results of a `readwhilewriting` benchmark (in order to provide somewhat more realistic results): ``` Tombstones? | avg micros/op | stddev micros/op | avg ops/s | stddev ops/s ----------------- | ------------- | ---------------- | ------------ | ------------ None | 0.6186 | 0.04637 | 1,625,252.90 | 124,679.41 500 Expanded | 0.6019 | 0.03628 | 1,666,670.40 | 101,142.65 500 Unexpanded | 0.6435 | 0.03994 | 1,559,979.40 | 104,090.52 1k Expanded | 0.6034 | 0.04349 | 1,665,128.10 | 125,144.57 1k Unexpanded | 0.6261 | 0.03093 | 1,600,457.50 | 79,024.94 5k Expanded | 0.6163 | 0.05926 | 1,636,668.80 | 154,888.85 5k Unexpanded | 0.6402 | 0.04002 | 1,567,804.70 | 100,965.55 10k Expanded | 0.6036 | 0.05105 | 1,667,237.70 | 142,830.36 10k Unexpanded | 0.6128 | 0.02598 | 1,634,633.40 | 72,161.82 25k Expanded | 0.6198 | 0.04542 | 1,620,980.50 | 116,662.93 25k Unexpanded | 0.5478 | 0.0362 | 1,833,059.10 | 121,233.81 50k Expanded | 0.5104 | 0.04347 | 1,973,107.90 | 184,073.49 50k Unexpanded | 0.4528 | 0.03387 | 2,219,034.50 | 170,984.32 ``` After a large enough quantity of range tombstones are written, range tombstone Gets can become faster than reading from an equivalent DB with several point tombstones. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4493 Differential Revision: D10842844 Pulled By: abhimadan fbshipit-source-id: a7d44534f8120e6aabb65779d26c6b9df954c509
2018-10-26 04:25:00 +02:00
if (*max_covering_tombstone_seq > 0) {
// The remaining files we look at will only contain covered keys, so we
// stop here.
break;
Cache fragmented range tombstones in BlockBasedTableReader (#4493) Summary: This allows tombstone fragmenting to only be performed when the table is opened, and cached for subsequent accesses. On the same DB used in #4449, running `readrandom` results in the following: ``` readrandom : 0.983 micros/op 1017076 ops/sec; 78.3 MB/s (63103 of 100000 found) ``` Now that Get performance in the presence of range tombstones is reasonable, I also compared the performance between a DB with range tombstones, "expanded" range tombstones (several point tombstones that cover the same keys the equivalent range tombstone would cover, a common workaround for DeleteRange), and no range tombstones. The created DBs had 5 million keys each, and DeleteRange was called at regular intervals (depending on the total number of range tombstones being written) after 4.5 million Puts. The table below summarizes the results of a `readwhilewriting` benchmark (in order to provide somewhat more realistic results): ``` Tombstones? | avg micros/op | stddev micros/op | avg ops/s | stddev ops/s ----------------- | ------------- | ---------------- | ------------ | ------------ None | 0.6186 | 0.04637 | 1,625,252.90 | 124,679.41 500 Expanded | 0.6019 | 0.03628 | 1,666,670.40 | 101,142.65 500 Unexpanded | 0.6435 | 0.03994 | 1,559,979.40 | 104,090.52 1k Expanded | 0.6034 | 0.04349 | 1,665,128.10 | 125,144.57 1k Unexpanded | 0.6261 | 0.03093 | 1,600,457.50 | 79,024.94 5k Expanded | 0.6163 | 0.05926 | 1,636,668.80 | 154,888.85 5k Unexpanded | 0.6402 | 0.04002 | 1,567,804.70 | 100,965.55 10k Expanded | 0.6036 | 0.05105 | 1,667,237.70 | 142,830.36 10k Unexpanded | 0.6128 | 0.02598 | 1,634,633.40 | 72,161.82 25k Expanded | 0.6198 | 0.04542 | 1,620,980.50 | 116,662.93 25k Unexpanded | 0.5478 | 0.0362 | 1,833,059.10 | 121,233.81 50k Expanded | 0.5104 | 0.04347 | 1,973,107.90 | 184,073.49 50k Unexpanded | 0.4528 | 0.03387 | 2,219,034.50 | 170,984.32 ``` After a large enough quantity of range tombstones are written, range tombstone Gets can become faster than reading from an equivalent DB with several point tombstones. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4493 Differential Revision: D10842844 Pulled By: abhimadan fbshipit-source-id: a7d44534f8120e6aabb65779d26c6b9df954c509
2018-10-26 04:25:00 +02:00
}
if (get_context.sample()) {
sample_file_read_inc(f->file_metadata);
}
bool timer_enabled =
GetPerfLevel() >= PerfLevel::kEnableTimeExceptForMutex &&
get_perf_context()->per_level_perf_context_enabled;
StopWatchNano timer(clock_, timer_enabled /* auto_start */);
*status = table_cache_->Get(
read_options, *internal_comparator(), *f->file_metadata, ikey,
&get_context, mutable_cf_options_.prefix_extractor.get(),
cfd_->internal_stats()->GetFileReadHist(fp.GetHitFileLevel()),
IsFilterSkipped(static_cast<int>(fp.GetHitFileLevel()),
fp.IsHitFileLastInLevel()),
fp.GetHitFileLevel(), max_file_size_for_l0_meta_pin_);
// TODO: examine the behavior for corrupted key
if (timer_enabled) {
PERF_COUNTER_BY_LEVEL_ADD(get_from_table_nanos, timer.ElapsedNanos(),
fp.GetHitFileLevel());
}
if (!status->ok()) {
if (db_statistics_ != nullptr) {
get_context.ReportCounters();
}
return;
}
// report the counters before returning
if (get_context.State() != GetContext::kNotFound &&
get_context.State() != GetContext::kMerge &&
db_statistics_ != nullptr) {
get_context.ReportCounters();
}
switch (get_context.State()) {
case GetContext::kNotFound:
// Keep searching in other files
break;
case GetContext::kMerge:
// TODO: update per-level perfcontext user_key_return_count for kMerge
break;
case GetContext::kFound:
if (fp.GetHitFileLevel() == 0) {
RecordTick(db_statistics_, GET_HIT_L0);
} else if (fp.GetHitFileLevel() == 1) {
RecordTick(db_statistics_, GET_HIT_L1);
} else if (fp.GetHitFileLevel() >= 2) {
RecordTick(db_statistics_, GET_HIT_L2_AND_UP);
}
PERF_COUNTER_BY_LEVEL_ADD(user_key_return_count, 1,
fp.GetHitFileLevel());
if (is_blob_index) {
if (do_merge && value) {
constexpr uint64_t* bytes_read = nullptr;
*status =
GetBlob(read_options, user_key, *value, value, bytes_read);
if (!status->ok()) {
if (status->IsIncomplete()) {
get_context.MarkKeyMayExist();
}
return;
}
}
}
return;
case GetContext::kDeleted:
// Use empty error message for speed
*status = Status::NotFound();
return;
case GetContext::kCorrupt:
*status = Status::Corruption("corrupted key for ", user_key);
return;
case GetContext::kUnexpectedBlobIndex:
ROCKS_LOG_ERROR(info_log_, "Encounter unexpected blob index.");
*status = Status::NotSupported(
"Encounter unexpected blob index. Please open DB with "
"ROCKSDB_NAMESPACE::blob_db::BlobDB instead.");
return;
}
f = fp.GetNextFile();
}
if (db_statistics_ != nullptr) {
get_context.ReportCounters();
}
if (GetContext::kMerge == get_context.State()) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
if (!do_merge) {
*status = Status::OK();
return;
}
if (!merge_operator_) {
*status = Status::InvalidArgument(
"merge_operator is not properly initialized.");
return;
}
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// merge_operands are in saver and we hit the beginning of the key history
// do a final merge of nullptr and operands;
std::string* str_value = value != nullptr ? value->GetSelf() : nullptr;
*status = MergeHelper::TimedFullMerge(
merge_operator_, user_key, nullptr, merge_context->GetOperands(),
str_value, info_log_, db_statistics_, clock_,
nullptr /* result_operand */, true);
if (LIKELY(value != nullptr)) {
value->PinSelf();
}
} else {
if (key_exists != nullptr) {
*key_exists = false;
}
*status = Status::NotFound(); // Use an empty error message for speed
}
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
void Version::MultiGet(const ReadOptions& read_options, MultiGetRange* range,
ReadCallback* callback) {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
PinnedIteratorsManager pinned_iters_mgr;
// Pin blocks that we read to hold merge operands
if (merge_operator_) {
pinned_iters_mgr.StartPinning();
}
uint64_t tracing_mget_id = BlockCacheTraceHelper::kReservedGetId;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
if (vset_ && vset_->block_cache_tracer_ &&
vset_->block_cache_tracer_->is_tracing_enabled()) {
tracing_mget_id = vset_->block_cache_tracer_->NextGetId();
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
// Even though we know the batch size won't be > MAX_BATCH_SIZE,
// use autovector in order to avoid unnecessary construction of GetContext
// objects, which is expensive
autovector<GetContext, 16> get_ctx;
BlobFetcher blob_fetcher(this, read_options);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
for (auto iter = range->begin(); iter != range->end(); ++iter) {
assert(iter->s->ok() || iter->s->IsMergeInProgress());
get_ctx.emplace_back(
user_comparator(), merge_operator_, info_log_, db_statistics_,
iter->s->ok() ? GetContext::kNotFound : GetContext::kMerge,
iter->ukey_with_ts, iter->value, iter->timestamp, nullptr,
&(iter->merge_context), true, &iter->max_covering_tombstone_seq, clock_,
nullptr, merge_operator_ ? &pinned_iters_mgr : nullptr, callback,
&iter->is_blob_index, tracing_mget_id, &blob_fetcher);
// MergeInProgress status, if set, has been transferred to the get_context
// state, so we set status to ok here. From now on, the iter status will
// be used for IO errors, and get_context state will be used for any
// key level errors
*(iter->s) = Status::OK();
}
int get_ctx_index = 0;
for (auto iter = range->begin(); iter != range->end();
++iter, get_ctx_index++) {
iter->get_context = &(get_ctx[get_ctx_index]);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
MultiGetRange file_picker_range(*range, range->begin(), range->end());
FilePickerMultiGet fp(
&file_picker_range,
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
&storage_info_.level_files_brief_, storage_info_.num_non_empty_levels_,
&storage_info_.file_indexer_, user_comparator(), internal_comparator());
FdWithKeyRange* f = fp.GetNextFile();
Status s;
uint64_t num_index_read = 0;
uint64_t num_filter_read = 0;
uint64_t num_data_read = 0;
uint64_t num_sst_read = 0;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
while (f != nullptr) {
MultiGetRange file_range = fp.CurrentFileRange();
bool timer_enabled =
GetPerfLevel() >= PerfLevel::kEnableTimeExceptForMutex &&
get_perf_context()->per_level_perf_context_enabled;
StopWatchNano timer(clock_, timer_enabled /* auto_start */);
s = table_cache_->MultiGet(
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
read_options, *internal_comparator(), *f->file_metadata, &file_range,
mutable_cf_options_.prefix_extractor.get(),
cfd_->internal_stats()->GetFileReadHist(fp.GetHitFileLevel()),
IsFilterSkipped(static_cast<int>(fp.GetHitFileLevel()),
fp.IsHitFileLastInLevel()),
fp.GetHitFileLevel());
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
// TODO: examine the behavior for corrupted key
if (timer_enabled) {
PERF_COUNTER_BY_LEVEL_ADD(get_from_table_nanos, timer.ElapsedNanos(),
fp.GetHitFileLevel());
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
if (!s.ok()) {
// TODO: Set status for individual keys appropriately
for (auto iter = file_range.begin(); iter != file_range.end(); ++iter) {
*iter->s = s;
file_range.MarkKeyDone(iter);
}
return;
}
uint64_t batch_size = 0;
for (auto iter = file_range.begin(); s.ok() && iter != file_range.end();
++iter) {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
GetContext& get_context = *iter->get_context;
Status* status = iter->s;
// The Status in the KeyContext takes precedence over GetContext state
// Status may be an error if there were any IO errors in the table
// reader. We never expect Status to be NotFound(), as that is
// determined by get_context
assert(!status->IsNotFound());
if (!status->ok()) {
file_range.MarkKeyDone(iter);
continue;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
if (get_context.sample()) {
sample_file_read_inc(f->file_metadata);
}
batch_size++;
num_index_read += get_context.get_context_stats_.num_index_read;
num_filter_read += get_context.get_context_stats_.num_filter_read;
num_data_read += get_context.get_context_stats_.num_data_read;
num_sst_read += get_context.get_context_stats_.num_sst_read;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
// report the counters before returning
if (get_context.State() != GetContext::kNotFound &&
get_context.State() != GetContext::kMerge &&
db_statistics_ != nullptr) {
get_context.ReportCounters();
} else {
if (iter->max_covering_tombstone_seq > 0) {
// The remaining files we look at will only contain covered keys, so
// we stop here for this key
file_picker_range.SkipKey(iter);
}
}
switch (get_context.State()) {
case GetContext::kNotFound:
// Keep searching in other files
break;
case GetContext::kMerge:
// TODO: update per-level perfcontext user_key_return_count for kMerge
break;
case GetContext::kFound:
if (fp.GetHitFileLevel() == 0) {
RecordTick(db_statistics_, GET_HIT_L0);
} else if (fp.GetHitFileLevel() == 1) {
RecordTick(db_statistics_, GET_HIT_L1);
} else if (fp.GetHitFileLevel() >= 2) {
RecordTick(db_statistics_, GET_HIT_L2_AND_UP);
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
PERF_COUNTER_BY_LEVEL_ADD(user_key_return_count, 1,
fp.GetHitFileLevel());
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
file_range.MarkKeyDone(iter);
if (iter->is_blob_index) {
if (iter->value) {
constexpr uint64_t* bytes_read = nullptr;
*status = GetBlob(read_options, iter->ukey_with_ts, *iter->value,
iter->value, bytes_read);
if (!status->ok()) {
if (status->IsIncomplete()) {
get_context.MarkKeyMayExist();
}
continue;
}
}
}
file_range.AddValueSize(iter->value->size());
if (file_range.GetValueSize() > read_options.value_size_soft_limit) {
s = Status::Aborted();
break;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
continue;
case GetContext::kDeleted:
// Use empty error message for speed
*status = Status::NotFound();
file_range.MarkKeyDone(iter);
continue;
case GetContext::kCorrupt:
*status =
Status::Corruption("corrupted key for ", iter->lkey->user_key());
file_range.MarkKeyDone(iter);
continue;
case GetContext::kUnexpectedBlobIndex:
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
ROCKS_LOG_ERROR(info_log_, "Encounter unexpected blob index.");
*status = Status::NotSupported(
"Encounter unexpected blob index. Please open DB with "
"ROCKSDB_NAMESPACE::blob_db::BlobDB instead.");
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
file_range.MarkKeyDone(iter);
continue;
}
}
// Report MultiGet stats per level.
if (fp.IsHitFileLastInLevel()) {
// Dump the stats if this is the last file of this level and reset for
// next level.
RecordInHistogram(db_statistics_,
NUM_INDEX_AND_FILTER_BLOCKS_READ_PER_LEVEL,
num_index_read + num_filter_read);
RecordInHistogram(db_statistics_, NUM_DATA_BLOCKS_READ_PER_LEVEL,
num_data_read);
RecordInHistogram(db_statistics_, NUM_SST_READ_PER_LEVEL, num_sst_read);
num_filter_read = 0;
num_index_read = 0;
num_data_read = 0;
num_sst_read = 0;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
RecordInHistogram(db_statistics_, SST_BATCH_SIZE, batch_size);
if (!s.ok() || file_picker_range.empty()) {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
break;
}
f = fp.GetNextFile();
}
// Process any left over keys
for (auto iter = range->begin(); s.ok() && iter != range->end(); ++iter) {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
GetContext& get_context = *iter->get_context;
Status* status = iter->s;
Slice user_key = iter->lkey->user_key();
if (db_statistics_ != nullptr) {
get_context.ReportCounters();
}
if (GetContext::kMerge == get_context.State()) {
if (!merge_operator_) {
*status = Status::InvalidArgument(
"merge_operator is not properly initialized.");
range->MarkKeyDone(iter);
continue;
}
// merge_operands are in saver and we hit the beginning of the key history
// do a final merge of nullptr and operands;
std::string* str_value =
iter->value != nullptr ? iter->value->GetSelf() : nullptr;
*status = MergeHelper::TimedFullMerge(
merge_operator_, user_key, nullptr, iter->merge_context.GetOperands(),
str_value, info_log_, db_statistics_, clock_,
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
nullptr /* result_operand */, true);
if (LIKELY(iter->value != nullptr)) {
iter->value->PinSelf();
range->AddValueSize(iter->value->size());
range->MarkKeyDone(iter);
if (range->GetValueSize() > read_options.value_size_soft_limit) {
s = Status::Aborted();
break;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
} else {
range->MarkKeyDone(iter);
*status = Status::NotFound(); // Use an empty error message for speed
}
}
for (auto iter = range->begin(); iter != range->end(); ++iter) {
range->MarkKeyDone(iter);
*(iter->s) = s;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
}
bool Version::IsFilterSkipped(int level, bool is_file_last_in_level) {
// Reaching the bottom level implies misses at all upper levels, so we'll
// skip checking the filters when we predict a hit.
return cfd_->ioptions()->optimize_filters_for_hits &&
(level > 0 || is_file_last_in_level) &&
level == storage_info_.num_non_empty_levels() - 1;
}
void VersionStorageInfo::GenerateLevelFilesBrief() {
level_files_brief_.resize(num_non_empty_levels_);
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
for (int level = 0; level < num_non_empty_levels_; level++) {
DoGenerateLevelFilesBrief(
&level_files_brief_[level], files_[level], &arena_);
create compressed_levels_ in Version, allocate its space using arena. Make Version::Get, Version::FindFile faster Summary: Define CompressedFileMetaData that just contains fd, smallest_slice, largest_slice. Create compressed_levels_ in Version, the space is allocated using arena Thus increase the file meta data locality, speed up "Get" and "FindFile" benchmark with in-memory tmpfs, could have 4% improvement under "random read" and 2% improvement under "read while writing" benchmark command: ./db_bench --db=/mnt/db/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --disable_wal=0 --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --benchmarks=readwhilewriting,readwhilewriting,readwhilewriting --use_existing_db=1 --num=52428800 --threads=1 —writes_per_second=81920 Read Random: From 1.8363 ms/op, improve to 1.7587 ms/op. Read while writing: From 2.985 ms/op, improve to 2.924 ms/op. Test Plan: make all check Reviewers: ljin, haobo, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, igor Differential Revision: https://reviews.facebook.net/D19419
2014-07-10 07:14:39 +02:00
}
}
void Version::PrepareApply(
const MutableCFOptions& mutable_cf_options,
bool update_stats) {
TEST_SYNC_POINT_CALLBACK(
"Version::PrepareApply:forced_check",
reinterpret_cast<void*>(&storage_info_.force_consistency_checks_));
UpdateAccumulatedStats(update_stats);
storage_info_.UpdateNumNonEmptyLevels();
storage_info_.CalculateBaseBytes(*cfd_->ioptions(), mutable_cf_options);
storage_info_.UpdateFilesByCompactionPri(cfd_->ioptions()->compaction_pri);
storage_info_.GenerateFileIndexer();
storage_info_.GenerateLevelFilesBrief();
Allowing L0 -> L1 trivial move on sorted data Summary: This diff updates the logic of how we do trivial move, now trivial move can run on any number of files in input level as long as they are not overlapping The conditions for trivial move have been updated Introduced conditions: - Trivial move cannot happen if we have a compaction filter (except if the compaction is not manual) - Input level files cannot be overlapping Removed conditions: - Trivial move only run when the compaction is not manual - Input level should can contain only 1 file More context on what tests failed because of Trivial move ``` DBTest.CompactionsGenerateMultipleFiles This test is expecting compaction on a file in L0 to generate multiple files in L1, this test will fail with trivial move because we end up with one file in L1 ``` ``` DBTest.NoSpaceCompactRange This test expect compaction to fail when we force environment to report running out of space, of course this is not valid in trivial move situation because trivial move does not need any extra space, and did not check for that ``` ``` DBTest.DropWrites Similar to DBTest.NoSpaceCompactRange ``` ``` DBTest.DeleteObsoleteFilesPendingOutputs This test expect that a file in L2 is deleted after it's moved to L3, this is not valid with trivial move because although the file was moved it is now used by L3 ``` ``` CuckooTableDBTest.CompactionIntoMultipleFiles Same as DBTest.CompactionsGenerateMultipleFiles ``` This diff is based on a work by @sdong https://reviews.facebook.net/D34149 Test Plan: make -j64 check Reviewers: rven, sdong, igor Reviewed By: igor Subscribers: yhchiang, ott, march, dhruba, sdong Differential Revision: https://reviews.facebook.net/D34797
2015-06-05 01:51:25 +02:00
storage_info_.GenerateLevel0NonOverlapping();
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
storage_info_.GenerateBottommostFiles();
}
bool Version::MaybeInitializeFileMetaData(FileMetaData* file_meta) {
if (file_meta->init_stats_from_file ||
file_meta->compensated_file_size > 0) {
return false;
}
std::shared_ptr<const TableProperties> tp;
Status s = GetTableProperties(&tp, file_meta);
file_meta->init_stats_from_file = true;
if (!s.ok()) {
ROCKS_LOG_ERROR(vset_->db_options_->info_log,
"Unable to load table properties for file %" PRIu64
" --- %s\n",
file_meta->fd.GetNumber(), s.ToString().c_str());
return false;
}
if (tp.get() == nullptr) return false;
file_meta->num_entries = tp->num_entries;
file_meta->num_deletions = tp->num_deletions;
file_meta->raw_value_size = tp->raw_value_size;
file_meta->raw_key_size = tp->raw_key_size;
return true;
}
void VersionStorageInfo::UpdateAccumulatedStats(FileMetaData* file_meta) {
TEST_SYNC_POINT_CALLBACK("VersionStorageInfo::UpdateAccumulatedStats",
nullptr);
assert(file_meta->init_stats_from_file);
accumulated_file_size_ += file_meta->fd.GetFileSize();
accumulated_raw_key_size_ += file_meta->raw_key_size;
accumulated_raw_value_size_ += file_meta->raw_value_size;
accumulated_num_non_deletions_ +=
file_meta->num_entries - file_meta->num_deletions;
accumulated_num_deletions_ += file_meta->num_deletions;
2015-12-07 19:51:08 +01:00
current_num_non_deletions_ +=
file_meta->num_entries - file_meta->num_deletions;
current_num_deletions_ += file_meta->num_deletions;
current_num_samples_++;
}
void VersionStorageInfo::RemoveCurrentStats(FileMetaData* file_meta) {
if (file_meta->init_stats_from_file) {
current_num_non_deletions_ -=
file_meta->num_entries - file_meta->num_deletions;
current_num_deletions_ -= file_meta->num_deletions;
current_num_samples_--;
}
}
void Version::UpdateAccumulatedStats(bool update_stats) {
if (update_stats) {
// maximum number of table properties loaded from files.
const int kMaxInitCount = 20;
int init_count = 0;
// here only the first kMaxInitCount files which haven't been
// initialized from file will be updated with num_deletions.
// The motivation here is to cap the maximum I/O per Version creation.
// The reason for choosing files from lower-level instead of higher-level
// is that such design is able to propagate the initialization from
// lower-level to higher-level: When the num_deletions of lower-level
// files are updated, it will make the lower-level files have accurate
// compensated_file_size, making lower-level to higher-level compaction
// will be triggered, which creates higher-level files whose num_deletions
// will be updated here.
for (int level = 0;
level < storage_info_.num_levels_ && init_count < kMaxInitCount;
++level) {
for (auto* file_meta : storage_info_.files_[level]) {
if (MaybeInitializeFileMetaData(file_meta)) {
// each FileMeta will be initialized only once.
storage_info_.UpdateAccumulatedStats(file_meta);
// when option "max_open_files" is -1, all the file metadata has
// already been read, so MaybeInitializeFileMetaData() won't incur
// any I/O cost. "max_open_files=-1" means that the table cache passed
// to the VersionSet and then to the ColumnFamilySet has a size of
// TableCache::kInfiniteCapacity
if (vset_->GetColumnFamilySet()->get_table_cache()->GetCapacity() ==
TableCache::kInfiniteCapacity) {
continue;
}
if (++init_count >= kMaxInitCount) {
break;
}
}
}
}
// In case all sampled-files contain only deletion entries, then we
// load the table-property of a file in higher-level to initialize
// that value.
for (int level = storage_info_.num_levels_ - 1;
storage_info_.accumulated_raw_value_size_ == 0 && level >= 0;
--level) {
for (int i = static_cast<int>(storage_info_.files_[level].size()) - 1;
storage_info_.accumulated_raw_value_size_ == 0 && i >= 0; --i) {
if (MaybeInitializeFileMetaData(storage_info_.files_[level][i])) {
storage_info_.UpdateAccumulatedStats(storage_info_.files_[level][i]);
}
}
}
}
storage_info_.ComputeCompensatedSizes();
}
void VersionStorageInfo::ComputeCompensatedSizes() {
static const int kDeletionWeightOnCompaction = 2;
uint64_t average_value_size = GetAverageValueSize();
// compute the compensated size
for (int level = 0; level < num_levels_; level++) {
for (auto* file_meta : files_[level]) {
// Here we only compute compensated_file_size for those file_meta
// which compensated_file_size is uninitialized (== 0). This is true only
// for files that have been created right now and no other thread has
// access to them. That's why we can safely mutate compensated_file_size.
if (file_meta->compensated_file_size == 0) {
file_meta->compensated_file_size = file_meta->fd.GetFileSize();
// Here we only boost the size of deletion entries of a file only
// when the number of deletion entries is greater than the number of
// non-deletion entries in the file. The motivation here is that in
// a stable workload, the number of deletion entries should be roughly
// equal to the number of non-deletion entries. If we compensate the
// size of deletion entries in a stable workload, the deletion
// compensation logic might introduce unwanted effet which changes the
// shape of LSM tree.
if (file_meta->num_deletions * 2 >= file_meta->num_entries) {
file_meta->compensated_file_size +=
(file_meta->num_deletions * 2 - file_meta->num_entries) *
average_value_size * kDeletionWeightOnCompaction;
}
}
}
}
}
int VersionStorageInfo::MaxInputLevel() const {
if (compaction_style_ == kCompactionStyleLevel) {
return num_levels() - 2;
}
return 0;
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
2017-08-04 00:36:28 +02:00
int VersionStorageInfo::MaxOutputLevel(bool allow_ingest_behind) const {
if (allow_ingest_behind) {
assert(num_levels() > 1);
return num_levels() - 2;
}
return num_levels() - 1;
}
void VersionStorageInfo::EstimateCompactionBytesNeeded(
const MutableCFOptions& mutable_cf_options) {
// Only implemented for level-based compaction
if (compaction_style_ != kCompactionStyleLevel) {
estimated_compaction_needed_bytes_ = 0;
return;
}
// Start from Level 0, if level 0 qualifies compaction to level 1,
// we estimate the size of compaction.
// Then we move on to the next level and see whether it qualifies compaction
// to the next level. The size of the level is estimated as the actual size
// on the level plus the input bytes from the previous level if there is any.
// If it exceeds, take the exceeded bytes as compaction input and add the size
// of the compaction size to tatal size.
// We keep doing it to Level 2, 3, etc, until the last level and return the
// accumulated bytes.
uint64_t bytes_compact_to_next_level = 0;
uint64_t level_size = 0;
for (auto* f : files_[0]) {
level_size += f->fd.GetFileSize();
}
// Level 0
bool level0_compact_triggered = false;
if (static_cast<int>(files_[0].size()) >=
mutable_cf_options.level0_file_num_compaction_trigger ||
level_size >= mutable_cf_options.max_bytes_for_level_base) {
level0_compact_triggered = true;
estimated_compaction_needed_bytes_ = level_size;
bytes_compact_to_next_level = level_size;
} else {
estimated_compaction_needed_bytes_ = 0;
}
// Level 1 and up.
uint64_t bytes_next_level = 0;
for (int level = base_level(); level <= MaxInputLevel(); level++) {
level_size = 0;
if (bytes_next_level > 0) {
#ifndef NDEBUG
uint64_t level_size2 = 0;
for (auto* f : files_[level]) {
level_size2 += f->fd.GetFileSize();
}
assert(level_size2 == bytes_next_level);
#endif
level_size = bytes_next_level;
bytes_next_level = 0;
} else {
for (auto* f : files_[level]) {
level_size += f->fd.GetFileSize();
}
}
if (level == base_level() && level0_compact_triggered) {
// Add base level size to compaction if level0 compaction triggered.
estimated_compaction_needed_bytes_ += level_size;
}
// Add size added by previous compaction
level_size += bytes_compact_to_next_level;
bytes_compact_to_next_level = 0;
uint64_t level_target = MaxBytesForLevel(level);
if (level_size > level_target) {
bytes_compact_to_next_level = level_size - level_target;
// Estimate the actual compaction fan-out ratio as size ratio between
// the two levels.
assert(bytes_next_level == 0);
if (level + 1 < num_levels_) {
for (auto* f : files_[level + 1]) {
bytes_next_level += f->fd.GetFileSize();
}
}
if (bytes_next_level > 0) {
assert(level_size > 0);
estimated_compaction_needed_bytes_ += static_cast<uint64_t>(
static_cast<double>(bytes_compact_to_next_level) *
(static_cast<double>(bytes_next_level) /
static_cast<double>(level_size) +
1));
}
}
}
}
FIFO Compaction with TTL Summary: Introducing FIFO compactions with TTL. FIFO compaction is based on size only which makes it tricky to enable in production as use cases can have organic growth. A user requested an option to drop files based on the time of their creation instead of the total size. To address that request: - Added a new TTL option to FIFO compaction options. - Updated FIFO compaction score to take TTL into consideration. - Added a new table property, creation_time, to keep track of when the SST file is created. - Creation_time is set as below: - On Flush: Set to the time of flush. - On Compaction: Set to the max creation_time of all the files involved in the compaction. - On Repair and Recovery: Set to the time of repair/recovery. - Old files created prior to this code change will have a creation_time of 0. - FIFO compaction with TTL is enabled when ttl > 0. All files older than ttl will be deleted during compaction. i.e. `if (file.creation_time < (current_time - ttl)) then delete(file)`. This will enable cases where you might want to delete all files older than, say, 1 day. - FIFO compaction will fall back to the prior way of deleting files based on size if: - the creation_time of all files involved in compaction is 0. - the total size (of all SST files combined) does not drop below `compaction_options_fifo.max_table_files_size` even if the files older than ttl are deleted. This feature is not supported if max_open_files != -1 or with table formats other than Block-based. **Test Plan:** Added tests. **Benchmark results:** Base: FIFO with max size: 100MB :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 readwhilewriting : 1.924 micros/op 519858 ops/sec; 13.6 MB/s (1176277 of 5000000 found) ``` With TTL (a low one for testing) :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 --fifo_compaction_ttl=20 readwhilewriting : 1.902 micros/op 525817 ops/sec; 13.7 MB/s (1185057 of 5000000 found) ``` Example Log lines: ``` 2017/06/26-15:17:24.609249 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609177) [db/compaction_picker.cc:1471] [default] FIFO compaction: picking file 40 with creation time 1498515423 for deletion 2017/06/26-15:17:24.609255 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609234) [db/db_impl_compaction_flush.cc:1541] [default] Deleted 1 files ... 2017/06/26-15:17:25.553185 7fd5a61a5800 [DEBUG] [db/db_impl_files.cc:309] [JOB 0] Delete /dev/shm/dbbench/000040.sst type=2 #40 -- OK 2017/06/26-15:17:25.553205 7fd5a61a5800 EVENT_LOG_v1 {"time_micros": 1498515445553199, "job": 0, "event": "table_file_deletion", "file_number": 40} ``` SST Files remaining in the dbbench dir, after db_bench execution completed: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ ls -l /dev/shm//dbbench/*.sst -rw-r--r--. 1 svemuri users 30749887 Jun 26 15:17 /dev/shm//dbbench/000042.sst -rw-r--r--. 1 svemuri users 30768779 Jun 26 15:17 /dev/shm//dbbench/000044.sst -rw-r--r--. 1 svemuri users 30757481 Jun 26 15:17 /dev/shm//dbbench/000046.sst ``` Closes https://github.com/facebook/rocksdb/pull/2480 Differential Revision: D5305116 Pulled By: sagar0 fbshipit-source-id: 3e5cfcf5dd07ed2211b5b37492eb235b45139174
2017-06-28 02:02:20 +02:00
namespace {
uint32_t GetExpiredTtlFilesCount(const ImmutableOptions& ioptions,
const MutableCFOptions& mutable_cf_options,
FIFO Compaction with TTL Summary: Introducing FIFO compactions with TTL. FIFO compaction is based on size only which makes it tricky to enable in production as use cases can have organic growth. A user requested an option to drop files based on the time of their creation instead of the total size. To address that request: - Added a new TTL option to FIFO compaction options. - Updated FIFO compaction score to take TTL into consideration. - Added a new table property, creation_time, to keep track of when the SST file is created. - Creation_time is set as below: - On Flush: Set to the time of flush. - On Compaction: Set to the max creation_time of all the files involved in the compaction. - On Repair and Recovery: Set to the time of repair/recovery. - Old files created prior to this code change will have a creation_time of 0. - FIFO compaction with TTL is enabled when ttl > 0. All files older than ttl will be deleted during compaction. i.e. `if (file.creation_time < (current_time - ttl)) then delete(file)`. This will enable cases where you might want to delete all files older than, say, 1 day. - FIFO compaction will fall back to the prior way of deleting files based on size if: - the creation_time of all files involved in compaction is 0. - the total size (of all SST files combined) does not drop below `compaction_options_fifo.max_table_files_size` even if the files older than ttl are deleted. This feature is not supported if max_open_files != -1 or with table formats other than Block-based. **Test Plan:** Added tests. **Benchmark results:** Base: FIFO with max size: 100MB :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 readwhilewriting : 1.924 micros/op 519858 ops/sec; 13.6 MB/s (1176277 of 5000000 found) ``` With TTL (a low one for testing) :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 --fifo_compaction_ttl=20 readwhilewriting : 1.902 micros/op 525817 ops/sec; 13.7 MB/s (1185057 of 5000000 found) ``` Example Log lines: ``` 2017/06/26-15:17:24.609249 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609177) [db/compaction_picker.cc:1471] [default] FIFO compaction: picking file 40 with creation time 1498515423 for deletion 2017/06/26-15:17:24.609255 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609234) [db/db_impl_compaction_flush.cc:1541] [default] Deleted 1 files ... 2017/06/26-15:17:25.553185 7fd5a61a5800 [DEBUG] [db/db_impl_files.cc:309] [JOB 0] Delete /dev/shm/dbbench/000040.sst type=2 #40 -- OK 2017/06/26-15:17:25.553205 7fd5a61a5800 EVENT_LOG_v1 {"time_micros": 1498515445553199, "job": 0, "event": "table_file_deletion", "file_number": 40} ``` SST Files remaining in the dbbench dir, after db_bench execution completed: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ ls -l /dev/shm//dbbench/*.sst -rw-r--r--. 1 svemuri users 30749887 Jun 26 15:17 /dev/shm//dbbench/000042.sst -rw-r--r--. 1 svemuri users 30768779 Jun 26 15:17 /dev/shm//dbbench/000044.sst -rw-r--r--. 1 svemuri users 30757481 Jun 26 15:17 /dev/shm//dbbench/000046.sst ``` Closes https://github.com/facebook/rocksdb/pull/2480 Differential Revision: D5305116 Pulled By: sagar0 fbshipit-source-id: 3e5cfcf5dd07ed2211b5b37492eb235b45139174
2017-06-28 02:02:20 +02:00
const std::vector<FileMetaData*>& files) {
uint32_t ttl_expired_files_count = 0;
int64_t _current_time;
auto status = ioptions.clock->GetCurrentTime(&_current_time);
FIFO Compaction with TTL Summary: Introducing FIFO compactions with TTL. FIFO compaction is based on size only which makes it tricky to enable in production as use cases can have organic growth. A user requested an option to drop files based on the time of their creation instead of the total size. To address that request: - Added a new TTL option to FIFO compaction options. - Updated FIFO compaction score to take TTL into consideration. - Added a new table property, creation_time, to keep track of when the SST file is created. - Creation_time is set as below: - On Flush: Set to the time of flush. - On Compaction: Set to the max creation_time of all the files involved in the compaction. - On Repair and Recovery: Set to the time of repair/recovery. - Old files created prior to this code change will have a creation_time of 0. - FIFO compaction with TTL is enabled when ttl > 0. All files older than ttl will be deleted during compaction. i.e. `if (file.creation_time < (current_time - ttl)) then delete(file)`. This will enable cases where you might want to delete all files older than, say, 1 day. - FIFO compaction will fall back to the prior way of deleting files based on size if: - the creation_time of all files involved in compaction is 0. - the total size (of all SST files combined) does not drop below `compaction_options_fifo.max_table_files_size` even if the files older than ttl are deleted. This feature is not supported if max_open_files != -1 or with table formats other than Block-based. **Test Plan:** Added tests. **Benchmark results:** Base: FIFO with max size: 100MB :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 readwhilewriting : 1.924 micros/op 519858 ops/sec; 13.6 MB/s (1176277 of 5000000 found) ``` With TTL (a low one for testing) :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 --fifo_compaction_ttl=20 readwhilewriting : 1.902 micros/op 525817 ops/sec; 13.7 MB/s (1185057 of 5000000 found) ``` Example Log lines: ``` 2017/06/26-15:17:24.609249 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609177) [db/compaction_picker.cc:1471] [default] FIFO compaction: picking file 40 with creation time 1498515423 for deletion 2017/06/26-15:17:24.609255 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609234) [db/db_impl_compaction_flush.cc:1541] [default] Deleted 1 files ... 2017/06/26-15:17:25.553185 7fd5a61a5800 [DEBUG] [db/db_impl_files.cc:309] [JOB 0] Delete /dev/shm/dbbench/000040.sst type=2 #40 -- OK 2017/06/26-15:17:25.553205 7fd5a61a5800 EVENT_LOG_v1 {"time_micros": 1498515445553199, "job": 0, "event": "table_file_deletion", "file_number": 40} ``` SST Files remaining in the dbbench dir, after db_bench execution completed: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ ls -l /dev/shm//dbbench/*.sst -rw-r--r--. 1 svemuri users 30749887 Jun 26 15:17 /dev/shm//dbbench/000042.sst -rw-r--r--. 1 svemuri users 30768779 Jun 26 15:17 /dev/shm//dbbench/000044.sst -rw-r--r--. 1 svemuri users 30757481 Jun 26 15:17 /dev/shm//dbbench/000046.sst ``` Closes https://github.com/facebook/rocksdb/pull/2480 Differential Revision: D5305116 Pulled By: sagar0 fbshipit-source-id: 3e5cfcf5dd07ed2211b5b37492eb235b45139174
2017-06-28 02:02:20 +02:00
if (status.ok()) {
const uint64_t current_time = static_cast<uint64_t>(_current_time);
for (FileMetaData* f : files) {
if (!f->being_compacted) {
uint64_t oldest_ancester_time = f->TryGetOldestAncesterTime();
if (oldest_ancester_time != 0 &&
oldest_ancester_time < (current_time - mutable_cf_options.ttl)) {
FIFO Compaction with TTL Summary: Introducing FIFO compactions with TTL. FIFO compaction is based on size only which makes it tricky to enable in production as use cases can have organic growth. A user requested an option to drop files based on the time of their creation instead of the total size. To address that request: - Added a new TTL option to FIFO compaction options. - Updated FIFO compaction score to take TTL into consideration. - Added a new table property, creation_time, to keep track of when the SST file is created. - Creation_time is set as below: - On Flush: Set to the time of flush. - On Compaction: Set to the max creation_time of all the files involved in the compaction. - On Repair and Recovery: Set to the time of repair/recovery. - Old files created prior to this code change will have a creation_time of 0. - FIFO compaction with TTL is enabled when ttl > 0. All files older than ttl will be deleted during compaction. i.e. `if (file.creation_time < (current_time - ttl)) then delete(file)`. This will enable cases where you might want to delete all files older than, say, 1 day. - FIFO compaction will fall back to the prior way of deleting files based on size if: - the creation_time of all files involved in compaction is 0. - the total size (of all SST files combined) does not drop below `compaction_options_fifo.max_table_files_size` even if the files older than ttl are deleted. This feature is not supported if max_open_files != -1 or with table formats other than Block-based. **Test Plan:** Added tests. **Benchmark results:** Base: FIFO with max size: 100MB :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 readwhilewriting : 1.924 micros/op 519858 ops/sec; 13.6 MB/s (1176277 of 5000000 found) ``` With TTL (a low one for testing) :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 --fifo_compaction_ttl=20 readwhilewriting : 1.902 micros/op 525817 ops/sec; 13.7 MB/s (1185057 of 5000000 found) ``` Example Log lines: ``` 2017/06/26-15:17:24.609249 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609177) [db/compaction_picker.cc:1471] [default] FIFO compaction: picking file 40 with creation time 1498515423 for deletion 2017/06/26-15:17:24.609255 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609234) [db/db_impl_compaction_flush.cc:1541] [default] Deleted 1 files ... 2017/06/26-15:17:25.553185 7fd5a61a5800 [DEBUG] [db/db_impl_files.cc:309] [JOB 0] Delete /dev/shm/dbbench/000040.sst type=2 #40 -- OK 2017/06/26-15:17:25.553205 7fd5a61a5800 EVENT_LOG_v1 {"time_micros": 1498515445553199, "job": 0, "event": "table_file_deletion", "file_number": 40} ``` SST Files remaining in the dbbench dir, after db_bench execution completed: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ ls -l /dev/shm//dbbench/*.sst -rw-r--r--. 1 svemuri users 30749887 Jun 26 15:17 /dev/shm//dbbench/000042.sst -rw-r--r--. 1 svemuri users 30768779 Jun 26 15:17 /dev/shm//dbbench/000044.sst -rw-r--r--. 1 svemuri users 30757481 Jun 26 15:17 /dev/shm//dbbench/000046.sst ``` Closes https://github.com/facebook/rocksdb/pull/2480 Differential Revision: D5305116 Pulled By: sagar0 fbshipit-source-id: 3e5cfcf5dd07ed2211b5b37492eb235b45139174
2017-06-28 02:02:20 +02:00
ttl_expired_files_count++;
}
}
}
}
return ttl_expired_files_count;
}
} // anonymous namespace
void VersionStorageInfo::ComputeCompactionScore(
const ImmutableOptions& immutable_options,
const MutableCFOptions& mutable_cf_options) {
for (int level = 0; level <= MaxInputLevel(); level++) {
double score;
if (level == 0) {
// We treat level-0 specially by bounding the number of files
// instead of number of bytes for two reasons:
//
// (1) With larger write-buffer sizes, it is nice not to do too
// many level-0 compactions.
//
// (2) The files in level-0 are merged on every read and
// therefore we wish to avoid too many files when the individual
// file size is small (perhaps because of a small write-buffer
// setting, or very high compression ratios, or lots of
// overwrites/deletions).
int num_sorted_runs = 0;
uint64_t total_size = 0;
for (auto* f : files_[level]) {
if (!f->being_compacted) {
total_size += f->compensated_file_size;
num_sorted_runs++;
}
}
if (compaction_style_ == kCompactionStyleUniversal) {
// For universal compaction, we use level0 score to indicate
// compaction score for the whole DB. Adding other levels as if
// they are L0 files.
for (int i = 1; i < num_levels(); i++) {
// Its possible that a subset of the files in a level may be in a
// compaction, due to delete triggered compaction or trivial move.
// In that case, the below check may not catch a level being
// compacted as it only checks the first file. The worst that can
// happen is a scheduled compaction thread will find nothing to do.
if (!files_[i].empty() && !files_[i][0]->being_compacted) {
num_sorted_runs++;
}
}
}
if (compaction_style_ == kCompactionStyleFIFO) {
score = static_cast<double>(total_size) /
mutable_cf_options.compaction_options_fifo.max_table_files_size;
if (mutable_cf_options.compaction_options_fifo.allow_compaction ||
mutable_cf_options.compaction_options_fifo.age_for_warm > 0) {
// Warm tier move can happen at any time. It's too expensive to
// check very file's timestamp now. For now, just trigger it
// slightly more frequently than FIFO compaction so that this
// happens first.
score = std::max(
static_cast<double>(num_sorted_runs) /
mutable_cf_options.level0_file_num_compaction_trigger,
score);
}
if (mutable_cf_options.ttl > 0) {
score = std::max(
static_cast<double>(GetExpiredTtlFilesCount(
immutable_options, mutable_cf_options, files_[level])),
score);
FIFO Compaction with TTL Summary: Introducing FIFO compactions with TTL. FIFO compaction is based on size only which makes it tricky to enable in production as use cases can have organic growth. A user requested an option to drop files based on the time of their creation instead of the total size. To address that request: - Added a new TTL option to FIFO compaction options. - Updated FIFO compaction score to take TTL into consideration. - Added a new table property, creation_time, to keep track of when the SST file is created. - Creation_time is set as below: - On Flush: Set to the time of flush. - On Compaction: Set to the max creation_time of all the files involved in the compaction. - On Repair and Recovery: Set to the time of repair/recovery. - Old files created prior to this code change will have a creation_time of 0. - FIFO compaction with TTL is enabled when ttl > 0. All files older than ttl will be deleted during compaction. i.e. `if (file.creation_time < (current_time - ttl)) then delete(file)`. This will enable cases where you might want to delete all files older than, say, 1 day. - FIFO compaction will fall back to the prior way of deleting files based on size if: - the creation_time of all files involved in compaction is 0. - the total size (of all SST files combined) does not drop below `compaction_options_fifo.max_table_files_size` even if the files older than ttl are deleted. This feature is not supported if max_open_files != -1 or with table formats other than Block-based. **Test Plan:** Added tests. **Benchmark results:** Base: FIFO with max size: 100MB :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 readwhilewriting : 1.924 micros/op 519858 ops/sec; 13.6 MB/s (1176277 of 5000000 found) ``` With TTL (a low one for testing) :: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=readwhilewriting --num=5000000 --threads=16 --compaction_style=2 --fifo_compaction_max_table_files_size_mb=100 --fifo_compaction_ttl=20 readwhilewriting : 1.902 micros/op 525817 ops/sec; 13.7 MB/s (1185057 of 5000000 found) ``` Example Log lines: ``` 2017/06/26-15:17:24.609249 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609177) [db/compaction_picker.cc:1471] [default] FIFO compaction: picking file 40 with creation time 1498515423 for deletion 2017/06/26-15:17:24.609255 7fd5a45ff700 (Original Log Time 2017/06/26-15:17:24.609234) [db/db_impl_compaction_flush.cc:1541] [default] Deleted 1 files ... 2017/06/26-15:17:25.553185 7fd5a61a5800 [DEBUG] [db/db_impl_files.cc:309] [JOB 0] Delete /dev/shm/dbbench/000040.sst type=2 #40 -- OK 2017/06/26-15:17:25.553205 7fd5a61a5800 EVENT_LOG_v1 {"time_micros": 1498515445553199, "job": 0, "event": "table_file_deletion", "file_number": 40} ``` SST Files remaining in the dbbench dir, after db_bench execution completed: ``` svemuri@dev15905 ~/rocksdb (fifo-compaction) $ ls -l /dev/shm//dbbench/*.sst -rw-r--r--. 1 svemuri users 30749887 Jun 26 15:17 /dev/shm//dbbench/000042.sst -rw-r--r--. 1 svemuri users 30768779 Jun 26 15:17 /dev/shm//dbbench/000044.sst -rw-r--r--. 1 svemuri users 30757481 Jun 26 15:17 /dev/shm//dbbench/000046.sst ``` Closes https://github.com/facebook/rocksdb/pull/2480 Differential Revision: D5305116 Pulled By: sagar0 fbshipit-source-id: 3e5cfcf5dd07ed2211b5b37492eb235b45139174
2017-06-28 02:02:20 +02:00
}
} else {
score = static_cast<double>(num_sorted_runs) /
mutable_cf_options.level0_file_num_compaction_trigger;
if (compaction_style_ == kCompactionStyleLevel && num_levels() > 1) {
// Level-based involves L0->L0 compactions that can lead to oversized
// L0 files. Take into account size as well to avoid later giant
// compactions to the base level.
Bound L0->Lbase fanout in dynamic leveled compaction (#7325) Summary: L0 score is based on size target and number of files. The size target used is `max_bytes_for_level_base`. However, the base level's size can dynamically expand in write burst mode. In fact, it can expand so much that L0->Lbase becomes the highest fanout in target sizes. This doesn't make sense from an efficiency perspective, so this PR bounds the L0->Lbase fanout to the smoothed level multiplier. The L0 scoring based on file count remains unchanged. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7325 Test Plan: contrived benchmark that exhibits the problem: ``` $ TEST_TMPDIR=/data/users/andrewkr/ ./db_bench -benchmarks=filluniquerandom,readrandom -write_buffer_size=1048576 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -level0_file_num_compaction_trigger=4 -level_compaction_dynamic_level_bytes=true -compression_type=none -max_background_jobs=12 -rate_limiter_bytes_per_sec=104857600 -benchmark_write_rate_limit=10485760 -num=100000000 ``` Results: - "Burst W-Amp" is the write-amp near the end of the fillrandom benchmark - "Total W-Amp" is the write-amp after readrandom has run a while and all levels no longer need compaction Branch | Burst W-Amp | Total W-Amp | fillrandom (MB/s) -- | -- | -- | -- master | 20.2 | 21.5 | 4.7 dynamic-l0-score | 12.6 | 14.1 | 7.2 Reviewed By: siying Differential Revision: D23412935 Pulled By: ajkr fbshipit-source-id: f91f2067188e432dd39deab02f1c56f195057a0e
2020-09-02 04:32:59 +02:00
uint64_t l0_target_size = mutable_cf_options.max_bytes_for_level_base;
if (immutable_options.level_compaction_dynamic_level_bytes &&
Bound L0->Lbase fanout in dynamic leveled compaction (#7325) Summary: L0 score is based on size target and number of files. The size target used is `max_bytes_for_level_base`. However, the base level's size can dynamically expand in write burst mode. In fact, it can expand so much that L0->Lbase becomes the highest fanout in target sizes. This doesn't make sense from an efficiency perspective, so this PR bounds the L0->Lbase fanout to the smoothed level multiplier. The L0 scoring based on file count remains unchanged. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7325 Test Plan: contrived benchmark that exhibits the problem: ``` $ TEST_TMPDIR=/data/users/andrewkr/ ./db_bench -benchmarks=filluniquerandom,readrandom -write_buffer_size=1048576 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -level0_file_num_compaction_trigger=4 -level_compaction_dynamic_level_bytes=true -compression_type=none -max_background_jobs=12 -rate_limiter_bytes_per_sec=104857600 -benchmark_write_rate_limit=10485760 -num=100000000 ``` Results: - "Burst W-Amp" is the write-amp near the end of the fillrandom benchmark - "Total W-Amp" is the write-amp after readrandom has run a while and all levels no longer need compaction Branch | Burst W-Amp | Total W-Amp | fillrandom (MB/s) -- | -- | -- | -- master | 20.2 | 21.5 | 4.7 dynamic-l0-score | 12.6 | 14.1 | 7.2 Reviewed By: siying Differential Revision: D23412935 Pulled By: ajkr fbshipit-source-id: f91f2067188e432dd39deab02f1c56f195057a0e
2020-09-02 04:32:59 +02:00
level_multiplier_ != 0.0) {
// Prevent L0 to Lbase fanout from growing larger than
// `level_multiplier_`. This prevents us from getting stuck picking
// L0 forever even when it is hurting write-amp. That could happen
// in dynamic level compaction's write-burst mode where the base
// level's target size can grow to be enormous.
l0_target_size =
std::max(l0_target_size,
static_cast<uint64_t>(level_max_bytes_[base_level_] /
level_multiplier_));
}
score =
std::max(score, static_cast<double>(total_size) / l0_target_size);
}
}
} else {
// Compute the ratio of current size to size limit.
uint64_t level_bytes_no_compacting = 0;
for (auto f : files_[level]) {
if (!f->being_compacted) {
level_bytes_no_compacting += f->compensated_file_size;
}
}
score = static_cast<double>(level_bytes_no_compacting) /
MaxBytesForLevel(level);
}
compaction_level_[level] = level;
compaction_score_[level] = score;
}
// sort all the levels based on their score. Higher scores get listed
// first. Use bubble sort because the number of entries are small.
for (int i = 0; i < num_levels() - 2; i++) {
for (int j = i + 1; j < num_levels() - 1; j++) {
if (compaction_score_[i] < compaction_score_[j]) {
double score = compaction_score_[i];
int level = compaction_level_[i];
compaction_score_[i] = compaction_score_[j];
compaction_level_[i] = compaction_level_[j];
compaction_score_[j] = score;
compaction_level_[j] = level;
}
}
}
ComputeFilesMarkedForCompaction();
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
ComputeBottommostFilesMarkedForCompaction();
if (mutable_cf_options.ttl > 0) {
ComputeExpiredTtlFiles(immutable_options, mutable_cf_options.ttl);
}
if (mutable_cf_options.periodic_compaction_seconds > 0) {
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
ComputeFilesMarkedForPeriodicCompaction(
immutable_options, mutable_cf_options.periodic_compaction_seconds);
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
}
EstimateCompactionBytesNeeded(mutable_cf_options);
}
void VersionStorageInfo::ComputeFilesMarkedForCompaction() {
files_marked_for_compaction_.clear();
int last_qualify_level = 0;
// Do not include files from the last level with data
// If table properties collector suggests a file on the last level,
// we should not move it to a new level.
for (int level = num_levels() - 1; level >= 1; level--) {
if (!files_[level].empty()) {
last_qualify_level = level - 1;
break;
}
}
for (int level = 0; level <= last_qualify_level; level++) {
for (auto* f : files_[level]) {
if (!f->being_compacted && f->marked_for_compaction) {
files_marked_for_compaction_.emplace_back(level, f);
}
}
}
}
void VersionStorageInfo::ComputeExpiredTtlFiles(
const ImmutableOptions& ioptions, const uint64_t ttl) {
assert(ttl > 0);
expired_ttl_files_.clear();
int64_t _current_time;
auto status = ioptions.clock->GetCurrentTime(&_current_time);
if (!status.ok()) {
return;
}
const uint64_t current_time = static_cast<uint64_t>(_current_time);
for (int level = 0; level < num_levels() - 1; level++) {
for (FileMetaData* f : files_[level]) {
if (!f->being_compacted) {
uint64_t oldest_ancester_time = f->TryGetOldestAncesterTime();
if (oldest_ancester_time > 0 &&
oldest_ancester_time < (current_time - ttl)) {
expired_ttl_files_.emplace_back(level, f);
}
}
}
}
}
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
void VersionStorageInfo::ComputeFilesMarkedForPeriodicCompaction(
const ImmutableOptions& ioptions,
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
const uint64_t periodic_compaction_seconds) {
assert(periodic_compaction_seconds > 0);
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
files_marked_for_periodic_compaction_.clear();
int64_t temp_current_time;
auto status = ioptions.clock->GetCurrentTime(&temp_current_time);
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
if (!status.ok()) {
return;
}
const uint64_t current_time = static_cast<uint64_t>(temp_current_time);
Auto enable Periodic Compactions if a Compaction Filter is used (#5865) Summary: - Periodic compactions are auto-enabled if a compaction filter or a compaction filter factory is set, in Level Compaction. - The default value of `periodic_compaction_seconds` is changed to UINT64_MAX, which lets RocksDB auto-tune periodic compactions as needed. An explicit value of 0 will still work as before ie. to disable periodic compactions completely. For now, on seeing a compaction filter along with a UINT64_MAX value for `periodic_compaction_seconds`, RocksDB will make SST files older than 30 days to go through periodic copmactions. Some RocksDB users make use of compaction filters to control when their data can be deleted, usually with a custom TTL logic. But it is occasionally possible that the compactions get delayed by considerable time due to factors like low writes to a key range, data reaching bottom level, etc before the TTL expiry. Periodic Compactions feature was originally built to help such cases. Now periodic compactions are auto enabled by default when compaction filters or compaction filter factories are used, as it is generally helpful to all cases to collect garbage. `periodic_compaction_seconds` is set to a large value, 30 days, in `SanitizeOptions` when RocksDB sees that a `compaction_filter` or `compaction_filter_factory` is used. This is done only for Level Compaction style. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5865 Test Plan: - Added a new test `DBCompactionTest.LevelPeriodicCompactionWithCompactionFilters` to make sure that `periodic_compaction_seconds` is set if either `compaction_filter` or `compaction_filter_factory` options are set. - `COMPILE_WITH_ASAN=1 make check` Differential Revision: D17659180 Pulled By: sagar0 fbshipit-source-id: 4887b9cf2e53cf2dc93a7b658c6b15e1181217ee
2019-10-29 23:04:26 +01:00
// If periodic_compaction_seconds is larger than current time, periodic
// compaction can't possibly be triggered.
Auto enable Periodic Compactions if a Compaction Filter is used (#5865) Summary: - Periodic compactions are auto-enabled if a compaction filter or a compaction filter factory is set, in Level Compaction. - The default value of `periodic_compaction_seconds` is changed to UINT64_MAX, which lets RocksDB auto-tune periodic compactions as needed. An explicit value of 0 will still work as before ie. to disable periodic compactions completely. For now, on seeing a compaction filter along with a UINT64_MAX value for `periodic_compaction_seconds`, RocksDB will make SST files older than 30 days to go through periodic copmactions. Some RocksDB users make use of compaction filters to control when their data can be deleted, usually with a custom TTL logic. But it is occasionally possible that the compactions get delayed by considerable time due to factors like low writes to a key range, data reaching bottom level, etc before the TTL expiry. Periodic Compactions feature was originally built to help such cases. Now periodic compactions are auto enabled by default when compaction filters or compaction filter factories are used, as it is generally helpful to all cases to collect garbage. `periodic_compaction_seconds` is set to a large value, 30 days, in `SanitizeOptions` when RocksDB sees that a `compaction_filter` or `compaction_filter_factory` is used. This is done only for Level Compaction style. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5865 Test Plan: - Added a new test `DBCompactionTest.LevelPeriodicCompactionWithCompactionFilters` to make sure that `periodic_compaction_seconds` is set if either `compaction_filter` or `compaction_filter_factory` options are set. - `COMPILE_WITH_ASAN=1 make check` Differential Revision: D17659180 Pulled By: sagar0 fbshipit-source-id: 4887b9cf2e53cf2dc93a7b658c6b15e1181217ee
2019-10-29 23:04:26 +01:00
if (periodic_compaction_seconds > current_time) {
return;
}
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
const uint64_t allowed_time_limit =
current_time - periodic_compaction_seconds;
for (int level = 0; level < num_levels(); level++) {
for (auto f : files_[level]) {
if (!f->being_compacted) {
// Compute a file's modification time in the following order:
// 1. Use file_creation_time table property if it is > 0.
// 2. Use creation_time table property if it is > 0.
// 3. Use file's mtime metadata if the above two table properties are 0.
// Don't consider the file at all if the modification time cannot be
// correctly determined based on the above conditions.
uint64_t file_modification_time = f->TryGetFileCreationTime();
if (file_modification_time == kUnknownFileCreationTime) {
file_modification_time = f->TryGetOldestAncesterTime();
}
if (file_modification_time == kUnknownOldestAncesterTime) {
auto file_path = TableFileName(ioptions.cf_paths, f->fd.GetNumber(),
f->fd.GetPathId());
status = ioptions.env->GetFileModificationTime(
file_path, &file_modification_time);
if (!status.ok()) {
ROCKS_LOG_WARN(ioptions.logger,
"Can't get file modification time: %s: %s",
file_path.c_str(), status.ToString().c_str());
continue;
}
}
if (file_modification_time > 0 &&
file_modification_time < allowed_time_limit) {
Periodic Compactions (#5166) Summary: Introducing Periodic Compactions. This feature allows all the files in a CF to be periodically compacted. It could help in catching any corruptions that could creep into the DB proactively as every file is constantly getting re-compacted. And also, of course, it helps to cleanup data older than certain threshold. - Introduced a new option `periodic_compaction_time` to control how long a file can live without being compacted in a CF. - This works across all levels. - The files are put in the same level after going through the compaction. (Related files in the same level are picked up as `ExpandInputstoCleanCut` is used). - Compaction filters, if any, are invoked as usual. - A new table property, `file_creation_time`, is introduced to implement this feature. This property is set to the time at which the SST file was created (and that time is given by the underlying Env/OS). This feature can be enabled on its own, or in conjunction with `ttl`. It is possible to set a different time threshold for the bottom level when used in conjunction with ttl. Since `ttl` works only on 0 to last but one levels, you could set `ttl` to, say, 1 day, and `periodic_compaction_time` to, say, 7 days. Since `ttl < periodic_compaction_time` all files in last but one levels keep getting picked up based on ttl, and almost never based on periodic_compaction_time. The files in the bottom level get picked up for compaction based on `periodic_compaction_time`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5166 Differential Revision: D14884441 Pulled By: sagar0 fbshipit-source-id: 408426cbacb409c06386a98632dcf90bfa1bda47
2019-04-11 04:24:25 +02:00
files_marked_for_periodic_compaction_.emplace_back(level, f);
}
}
}
}
}
namespace {
// used to sort files by size
struct Fsize {
size_t index;
FileMetaData* file;
};
// Comparator that is used to sort files based on their size
// In normal mode: descending size
bool CompareCompensatedSizeDescending(const Fsize& first, const Fsize& second) {
return (first.file->compensated_file_size >
second.file->compensated_file_size);
}
} // anonymous namespace
void VersionStorageInfo::AddFile(int level, FileMetaData* f) {
auto& level_files = files_[level];
level_files.push_back(f);
f->refs++;
const uint64_t file_number = f->fd.GetNumber();
assert(file_locations_.find(file_number) == file_locations_.end());
file_locations_.emplace(file_number,
FileLocation(level, level_files.size() - 1));
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
void VersionStorageInfo::AddBlobFile(
std::shared_ptr<BlobFileMetaData> blob_file_meta) {
assert(blob_file_meta);
const uint64_t blob_file_number = blob_file_meta->GetBlobFileNumber();
auto it = blob_files_.lower_bound(blob_file_number);
assert(it == blob_files_.end() || it->first != blob_file_number);
blob_files_.insert(
it, BlobFiles::value_type(blob_file_number, std::move(blob_file_meta)));
}
// Version::PrepareApply() need to be called before calling the function, or
// following functions called:
// 1. UpdateNumNonEmptyLevels();
// 2. CalculateBaseBytes();
// 3. UpdateFilesByCompactionPri();
// 4. GenerateFileIndexer();
// 5. GenerateLevelFilesBrief();
Allowing L0 -> L1 trivial move on sorted data Summary: This diff updates the logic of how we do trivial move, now trivial move can run on any number of files in input level as long as they are not overlapping The conditions for trivial move have been updated Introduced conditions: - Trivial move cannot happen if we have a compaction filter (except if the compaction is not manual) - Input level files cannot be overlapping Removed conditions: - Trivial move only run when the compaction is not manual - Input level should can contain only 1 file More context on what tests failed because of Trivial move ``` DBTest.CompactionsGenerateMultipleFiles This test is expecting compaction on a file in L0 to generate multiple files in L1, this test will fail with trivial move because we end up with one file in L1 ``` ``` DBTest.NoSpaceCompactRange This test expect compaction to fail when we force environment to report running out of space, of course this is not valid in trivial move situation because trivial move does not need any extra space, and did not check for that ``` ``` DBTest.DropWrites Similar to DBTest.NoSpaceCompactRange ``` ``` DBTest.DeleteObsoleteFilesPendingOutputs This test expect that a file in L2 is deleted after it's moved to L3, this is not valid with trivial move because although the file was moved it is now used by L3 ``` ``` CuckooTableDBTest.CompactionIntoMultipleFiles Same as DBTest.CompactionsGenerateMultipleFiles ``` This diff is based on a work by @sdong https://reviews.facebook.net/D34149 Test Plan: make -j64 check Reviewers: rven, sdong, igor Reviewed By: igor Subscribers: yhchiang, ott, march, dhruba, sdong Differential Revision: https://reviews.facebook.net/D34797
2015-06-05 01:51:25 +02:00
// 6. GenerateLevel0NonOverlapping();
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
// 7. GenerateBottommostFiles();
void VersionStorageInfo::SetFinalized() {
finalized_ = true;
#ifndef NDEBUG
if (compaction_style_ != kCompactionStyleLevel) {
// Not level based compaction.
return;
}
assert(base_level_ < 0 || num_levels() == 1 ||
(base_level_ >= 1 && base_level_ < num_levels()));
// Verify all levels newer than base_level are empty except L0
for (int level = 1; level < base_level(); level++) {
assert(NumLevelBytes(level) == 0);
}
uint64_t max_bytes_prev_level = 0;
for (int level = base_level(); level < num_levels() - 1; level++) {
if (LevelFiles(level).size() == 0) {
continue;
}
assert(MaxBytesForLevel(level) >= max_bytes_prev_level);
max_bytes_prev_level = MaxBytesForLevel(level);
}
int num_empty_non_l0_level = 0;
for (int level = 0; level < num_levels(); level++) {
assert(LevelFiles(level).size() == 0 ||
LevelFiles(level).size() == LevelFilesBrief(level).num_files);
if (level > 0 && NumLevelBytes(level) > 0) {
num_empty_non_l0_level++;
}
if (LevelFiles(level).size() > 0) {
assert(level < num_non_empty_levels());
}
}
assert(compaction_level_.size() > 0);
assert(compaction_level_.size() == compaction_score_.size());
#endif
}
void VersionStorageInfo::UpdateNumNonEmptyLevels() {
num_non_empty_levels_ = num_levels_;
for (int i = num_levels_ - 1; i >= 0; i--) {
if (files_[i].size() != 0) {
return;
} else {
num_non_empty_levels_ = i;
}
}
}
namespace {
// Sort `temp` based on ratio of overlapping size over file size
void SortFileByOverlappingRatio(
const InternalKeyComparator& icmp, const std::vector<FileMetaData*>& files,
const std::vector<FileMetaData*>& next_level_files,
std::vector<Fsize>* temp) {
std::unordered_map<uint64_t, uint64_t> file_to_order;
auto next_level_it = next_level_files.begin();
for (auto& file : files) {
uint64_t overlapping_bytes = 0;
// Skip files in next level that is smaller than current file
while (next_level_it != next_level_files.end() &&
icmp.Compare((*next_level_it)->largest, file->smallest) < 0) {
next_level_it++;
}
while (next_level_it != next_level_files.end() &&
icmp.Compare((*next_level_it)->smallest, file->largest) < 0) {
overlapping_bytes += (*next_level_it)->fd.file_size;
if (icmp.Compare((*next_level_it)->largest, file->largest) > 0) {
// next level file cross large boundary of current file.
break;
}
next_level_it++;
}
assert(file->compensated_file_size != 0);
file_to_order[file->fd.GetNumber()] =
overlapping_bytes * 1024u / file->compensated_file_size;
}
std::sort(temp->begin(), temp->end(),
[&](const Fsize& f1, const Fsize& f2) -> bool {
return file_to_order[f1.file->fd.GetNumber()] <
file_to_order[f2.file->fd.GetNumber()];
});
}
} // namespace
void VersionStorageInfo::UpdateFilesByCompactionPri(
CompactionPri compaction_pri) {
if (compaction_style_ == kCompactionStyleNone ||
compaction_style_ == kCompactionStyleFIFO ||
compaction_style_ == kCompactionStyleUniversal) {
// don't need this
return;
}
// No need to sort the highest level because it is never compacted.
for (int level = 0; level < num_levels() - 1; level++) {
const std::vector<FileMetaData*>& files = files_[level];
auto& files_by_compaction_pri = files_by_compaction_pri_[level];
assert(files_by_compaction_pri.size() == 0);
// populate a temp vector for sorting based on size
std::vector<Fsize> temp(files.size());
for (size_t i = 0; i < files.size(); i++) {
temp[i].index = i;
temp[i].file = files[i];
}
// sort the top number_of_files_to_sort_ based on file size
size_t num = VersionStorageInfo::kNumberFilesToSort;
if (num > temp.size()) {
num = temp.size();
}
switch (compaction_pri) {
case kByCompensatedSize:
std::partial_sort(temp.begin(), temp.begin() + num, temp.end(),
CompareCompensatedSizeDescending);
break;
case kOldestLargestSeqFirst:
std::sort(temp.begin(), temp.end(),
[](const Fsize& f1, const Fsize& f2) -> bool {
return f1.file->fd.largest_seqno <
f2.file->fd.largest_seqno;
});
break;
case kOldestSmallestSeqFirst:
std::sort(temp.begin(), temp.end(),
[](const Fsize& f1, const Fsize& f2) -> bool {
return f1.file->fd.smallest_seqno <
f2.file->fd.smallest_seqno;
});
break;
case kMinOverlappingRatio:
SortFileByOverlappingRatio(*internal_comparator_, files_[level],
files_[level + 1], &temp);
break;
default:
assert(false);
}
assert(temp.size() == files.size());
// initialize files_by_compaction_pri_
for (size_t i = 0; i < temp.size(); i++) {
files_by_compaction_pri.push_back(static_cast<int>(temp[i].index));
}
next_file_to_compact_by_size_[level] = 0;
assert(files_[level].size() == files_by_compaction_pri_[level].size());
}
}
Allowing L0 -> L1 trivial move on sorted data Summary: This diff updates the logic of how we do trivial move, now trivial move can run on any number of files in input level as long as they are not overlapping The conditions for trivial move have been updated Introduced conditions: - Trivial move cannot happen if we have a compaction filter (except if the compaction is not manual) - Input level files cannot be overlapping Removed conditions: - Trivial move only run when the compaction is not manual - Input level should can contain only 1 file More context on what tests failed because of Trivial move ``` DBTest.CompactionsGenerateMultipleFiles This test is expecting compaction on a file in L0 to generate multiple files in L1, this test will fail with trivial move because we end up with one file in L1 ``` ``` DBTest.NoSpaceCompactRange This test expect compaction to fail when we force environment to report running out of space, of course this is not valid in trivial move situation because trivial move does not need any extra space, and did not check for that ``` ``` DBTest.DropWrites Similar to DBTest.NoSpaceCompactRange ``` ``` DBTest.DeleteObsoleteFilesPendingOutputs This test expect that a file in L2 is deleted after it's moved to L3, this is not valid with trivial move because although the file was moved it is now used by L3 ``` ``` CuckooTableDBTest.CompactionIntoMultipleFiles Same as DBTest.CompactionsGenerateMultipleFiles ``` This diff is based on a work by @sdong https://reviews.facebook.net/D34149 Test Plan: make -j64 check Reviewers: rven, sdong, igor Reviewed By: igor Subscribers: yhchiang, ott, march, dhruba, sdong Differential Revision: https://reviews.facebook.net/D34797
2015-06-05 01:51:25 +02:00
void VersionStorageInfo::GenerateLevel0NonOverlapping() {
assert(!finalized_);
level0_non_overlapping_ = true;
if (level_files_brief_.size() == 0) {
return;
}
// A copy of L0 files sorted by smallest key
std::vector<FdWithKeyRange> level0_sorted_file(
level_files_brief_[0].files,
level_files_brief_[0].files + level_files_brief_[0].num_files);
std::sort(level0_sorted_file.begin(), level0_sorted_file.end(),
[this](const FdWithKeyRange& f1, const FdWithKeyRange& f2) -> bool {
return (internal_comparator_->Compare(f1.smallest_key,
f2.smallest_key) < 0);
});
Allowing L0 -> L1 trivial move on sorted data Summary: This diff updates the logic of how we do trivial move, now trivial move can run on any number of files in input level as long as they are not overlapping The conditions for trivial move have been updated Introduced conditions: - Trivial move cannot happen if we have a compaction filter (except if the compaction is not manual) - Input level files cannot be overlapping Removed conditions: - Trivial move only run when the compaction is not manual - Input level should can contain only 1 file More context on what tests failed because of Trivial move ``` DBTest.CompactionsGenerateMultipleFiles This test is expecting compaction on a file in L0 to generate multiple files in L1, this test will fail with trivial move because we end up with one file in L1 ``` ``` DBTest.NoSpaceCompactRange This test expect compaction to fail when we force environment to report running out of space, of course this is not valid in trivial move situation because trivial move does not need any extra space, and did not check for that ``` ``` DBTest.DropWrites Similar to DBTest.NoSpaceCompactRange ``` ``` DBTest.DeleteObsoleteFilesPendingOutputs This test expect that a file in L2 is deleted after it's moved to L3, this is not valid with trivial move because although the file was moved it is now used by L3 ``` ``` CuckooTableDBTest.CompactionIntoMultipleFiles Same as DBTest.CompactionsGenerateMultipleFiles ``` This diff is based on a work by @sdong https://reviews.facebook.net/D34149 Test Plan: make -j64 check Reviewers: rven, sdong, igor Reviewed By: igor Subscribers: yhchiang, ott, march, dhruba, sdong Differential Revision: https://reviews.facebook.net/D34797
2015-06-05 01:51:25 +02:00
for (size_t i = 1; i < level0_sorted_file.size(); ++i) {
FdWithKeyRange& f = level0_sorted_file[i];
FdWithKeyRange& prev = level0_sorted_file[i - 1];
if (internal_comparator_->Compare(prev.largest_key, f.smallest_key) >= 0) {
level0_non_overlapping_ = false;
break;
}
}
}
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
void VersionStorageInfo::GenerateBottommostFiles() {
assert(!finalized_);
assert(bottommost_files_.empty());
for (size_t level = 0; level < level_files_brief_.size(); ++level) {
for (size_t file_idx = 0; file_idx < level_files_brief_[level].num_files;
++file_idx) {
const FdWithKeyRange& f = level_files_brief_[level].files[file_idx];
int l0_file_idx;
if (level == 0) {
l0_file_idx = static_cast<int>(file_idx);
} else {
l0_file_idx = -1;
}
Slice smallest_user_key = ExtractUserKey(f.smallest_key);
Slice largest_user_key = ExtractUserKey(f.largest_key);
if (!RangeMightExistAfterSortedRun(smallest_user_key, largest_user_key,
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
static_cast<int>(level),
l0_file_idx)) {
bottommost_files_.emplace_back(static_cast<int>(level),
f.file_metadata);
}
}
}
}
void VersionStorageInfo::UpdateOldestSnapshot(SequenceNumber seqnum) {
assert(seqnum >= oldest_snapshot_seqnum_);
oldest_snapshot_seqnum_ = seqnum;
if (oldest_snapshot_seqnum_ > bottommost_files_mark_threshold_) {
ComputeBottommostFilesMarkedForCompaction();
}
}
void VersionStorageInfo::ComputeBottommostFilesMarkedForCompaction() {
bottommost_files_marked_for_compaction_.clear();
bottommost_files_mark_threshold_ = kMaxSequenceNumber;
for (auto& level_and_file : bottommost_files_) {
if (!level_and_file.second->being_compacted &&
level_and_file.second->fd.largest_seqno != 0) {
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
// largest_seqno might be nonzero due to containing the final key in an
// earlier compaction, whose seqnum we didn't zero out. Multiple deletions
// ensures the file really contains deleted or overwritten keys.
if (level_and_file.second->fd.largest_seqno < oldest_snapshot_seqnum_) {
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
bottommost_files_marked_for_compaction_.push_back(level_and_file);
} else {
bottommost_files_mark_threshold_ =
std::min(bottommost_files_mark_threshold_,
level_and_file.second->fd.largest_seqno);
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
}
}
}
}
void Version::Ref() {
++refs_;
}
bool Version::Unref() {
assert(refs_ >= 1);
--refs_;
if (refs_ == 0) {
delete this;
return true;
}
return false;
}
bool VersionStorageInfo::OverlapInLevel(int level,
const Slice* smallest_user_key,
const Slice* largest_user_key) {
if (level >= num_non_empty_levels_) {
// empty level, no overlap
return false;
}
return SomeFileOverlapsRange(*internal_comparator_, (level > 0),
level_files_brief_[level], smallest_user_key,
largest_user_key);
}
// Store in "*inputs" all files in "level" that overlap [begin,end]
// If hint_index is specified, then it points to a file in the
// overlapping range.
// The file_index returns a pointer to any file in an overlapping range.
void VersionStorageInfo::GetOverlappingInputs(
int level, const InternalKey* begin, const InternalKey* end,
std::vector<FileMetaData*>* inputs, int hint_index, int* file_index,
bool expand_range, InternalKey** next_smallest) const {
if (level >= num_non_empty_levels_) {
// this level is empty, no overlapping inputs
return;
}
inputs->clear();
if (file_index) {
*file_index = -1;
}
const Comparator* user_cmp = user_comparator_;
if (level > 0) {
GetOverlappingInputsRangeBinarySearch(level, begin, end, inputs, hint_index,
file_index, false, next_smallest);
return;
}
if (next_smallest) {
// next_smallest key only makes sense for non-level 0, where files are
// non-overlapping
*next_smallest = nullptr;
}
Slice user_begin, user_end;
if (begin != nullptr) {
user_begin = begin->user_key();
}
if (end != nullptr) {
user_end = end->user_key();
}
// index stores the file index need to check.
std::list<size_t> index;
for (size_t i = 0; i < level_files_brief_[level].num_files; i++) {
index.emplace_back(i);
}
while (!index.empty()) {
bool found_overlapping_file = false;
auto iter = index.begin();
while (iter != index.end()) {
FdWithKeyRange* f = &(level_files_brief_[level].files[*iter]);
const Slice file_start = ExtractUserKey(f->smallest_key);
const Slice file_limit = ExtractUserKey(f->largest_key);
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
if (begin != nullptr &&
user_cmp->CompareWithoutTimestamp(file_limit, user_begin) < 0) {
// "f" is completely before specified range; skip it
iter++;
} else if (end != nullptr &&
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
user_cmp->CompareWithoutTimestamp(file_start, user_end) > 0) {
// "f" is completely after specified range; skip it
iter++;
} else {
// if overlap
inputs->emplace_back(files_[level][*iter]);
found_overlapping_file = true;
// record the first file index.
if (file_index && *file_index == -1) {
*file_index = static_cast<int>(*iter);
}
// the related file is overlap, erase to avoid checking again.
iter = index.erase(iter);
if (expand_range) {
if (begin != nullptr &&
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
user_cmp->CompareWithoutTimestamp(file_start, user_begin) < 0) {
user_begin = file_start;
}
Add support for timestamp in Get/Put (#5079) Summary: It's useful to be able to (optionally) associate key-value pairs with user-provided timestamps. This PR is an early effort towards this goal and continues the work of facebook#4942. A suite of new unit tests exist in DBBasicTestWithTimestampWithParam. Support for timestamp requires the user to provide timestamp as a slice in `ReadOptions` and `WriteOptions`. All timestamps of the same database must share the same length, format, etc. The format of the timestamp is the same throughout the same database, and the user is responsible for providing a comparator function (Comparator) to order the <key, timestamp> tuples. Once created, the format and length of the timestamp cannot change (at least for now). Test plan (on devserver): ``` $COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` All tests must pass. We also run the following db_bench tests to verify whether there is regression on Get/Put while timestamp is not enabled. ``` $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 ``` Repeat for 6 times for both versions. Results are as follows: ``` | | readrandom | fillrandom | | master | 16.77 MB/s | 47.05 MB/s | | PR5079 | 16.44 MB/s | 47.03 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5079 Differential Revision: D15132946 Pulled By: riversand963 fbshipit-source-id: 833a0d657eac21182f0f206c910a6438154c742c
2019-06-06 08:07:28 +02:00
if (end != nullptr &&
user_cmp->CompareWithoutTimestamp(file_limit, user_end) > 0) {
user_end = file_limit;
}
}
}
}
// if all the files left are not overlap, break
if (!found_overlapping_file) {
break;
}
}
}
// Store in "*inputs" files in "level" that within range [begin,end]
// Guarantee a "clean cut" boundary between the files in inputs
// and the surrounding files and the maxinum number of files.
// This will ensure that no parts of a key are lost during compaction.
// If hint_index is specified, then it points to a file in the range.
// The file_index returns a pointer to any file in an overlapping range.
void VersionStorageInfo::GetCleanInputsWithinInterval(
int level, const InternalKey* begin, const InternalKey* end,
std::vector<FileMetaData*>* inputs, int hint_index, int* file_index) const {
inputs->clear();
if (file_index) {
*file_index = -1;
}
if (level >= num_non_empty_levels_ || level == 0 ||
level_files_brief_[level].num_files == 0) {
// this level is empty, no inputs within range
// also don't support clean input interval within L0
return;
}
GetOverlappingInputsRangeBinarySearch(level, begin, end, inputs,
hint_index, file_index,
true /* within_interval */);
}
// Store in "*inputs" all files in "level" that overlap [begin,end]
// Employ binary search to find at least one file that overlaps the
// specified range. From that file, iterate backwards and
// forwards to find all overlapping files.
// if within_range is set, then only store the maximum clean inputs
// within range [begin, end]. "clean" means there is a boundary
// between the files in "*inputs" and the surrounding files
void VersionStorageInfo::GetOverlappingInputsRangeBinarySearch(
int level, const InternalKey* begin, const InternalKey* end,
std::vector<FileMetaData*>* inputs, int hint_index, int* file_index,
bool within_interval, InternalKey** next_smallest) const {
assert(level > 0);
auto user_cmp = user_comparator_;
const FdWithKeyRange* files = level_files_brief_[level].files;
const int num_files = static_cast<int>(level_files_brief_[level].num_files);
// begin to use binary search to find lower bound
// and upper bound.
int start_index = 0;
int end_index = num_files;
if (begin != nullptr) {
// if within_interval is true, with file_key would find
// not overlapping ranges in std::lower_bound.
auto cmp = [&user_cmp, &within_interval](const FdWithKeyRange& f,
const InternalKey* k) {
auto& file_key = within_interval ? f.file_metadata->smallest
: f.file_metadata->largest;
return sstableKeyCompare(user_cmp, file_key, *k) < 0;
};
start_index = static_cast<int>(
std::lower_bound(files,
files + (hint_index == -1 ? num_files : hint_index),
begin, cmp) -
files);
if (start_index > 0 && within_interval) {
bool is_overlapping = true;
while (is_overlapping && start_index < num_files) {
auto& pre_limit = files[start_index - 1].file_metadata->largest;
auto& cur_start = files[start_index].file_metadata->smallest;
is_overlapping = sstableKeyCompare(user_cmp, pre_limit, cur_start) == 0;
start_index += is_overlapping;
}
}
}
if (end != nullptr) {
// if within_interval is true, with file_key would find
// not overlapping ranges in std::upper_bound.
auto cmp = [&user_cmp, &within_interval](const InternalKey* k,
const FdWithKeyRange& f) {
auto& file_key = within_interval ? f.file_metadata->largest
: f.file_metadata->smallest;
return sstableKeyCompare(user_cmp, *k, file_key) < 0;
};
end_index = static_cast<int>(
std::upper_bound(files + start_index, files + num_files, end, cmp) -
files);
if (end_index < num_files && within_interval) {
bool is_overlapping = true;
while (is_overlapping && end_index > start_index) {
auto& next_start = files[end_index].file_metadata->smallest;
auto& cur_limit = files[end_index - 1].file_metadata->largest;
is_overlapping =
sstableKeyCompare(user_cmp, cur_limit, next_start) == 0;
end_index -= is_overlapping;
}
}
}
assert(start_index <= end_index);
// If there were no overlapping files, return immediately.
if (start_index == end_index) {
if (next_smallest) {
*next_smallest = nullptr;
}
return;
}
assert(start_index < end_index);
// returns the index where an overlap is found
if (file_index) {
*file_index = start_index;
}
// insert overlapping files into vector
for (int i = start_index; i < end_index; i++) {
inputs->push_back(files_[level][i]);
}
if (next_smallest != nullptr) {
// Provide the next key outside the range covered by inputs
if (end_index < static_cast<int>(files_[level].size())) {
**next_smallest = files_[level][end_index]->smallest;
} else {
*next_smallest = nullptr;
}
}
}
uint64_t VersionStorageInfo::NumLevelBytes(int level) const {
assert(level >= 0);
assert(level < num_levels());
return TotalFileSize(files_[level]);
}
const char* VersionStorageInfo::LevelSummary(
LevelSummaryStorage* scratch) const {
int len = 0;
if (compaction_style_ == kCompactionStyleLevel && num_levels() > 1) {
assert(base_level_ < static_cast<int>(level_max_bytes_.size()));
if (level_multiplier_ != 0.0) {
len = snprintf(
scratch->buffer, sizeof(scratch->buffer),
"base level %d level multiplier %.2f max bytes base %" PRIu64 " ",
base_level_, level_multiplier_, level_max_bytes_[base_level_]);
}
}
len +=
snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len, "files[");
for (int i = 0; i < num_levels(); i++) {
int sz = sizeof(scratch->buffer) - len;
int ret = snprintf(scratch->buffer + len, sz, "%d ", int(files_[i].size()));
if (ret < 0 || ret >= sz) break;
len += ret;
}
if (len > 0) {
// overwrite the last space
--len;
}
len += snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len,
"] max score %.2f", compaction_score_[0]);
if (!files_marked_for_compaction_.empty()) {
snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len,
" (%" ROCKSDB_PRIszt " files need compaction)",
files_marked_for_compaction_.size());
}
return scratch->buffer;
}
const char* VersionStorageInfo::LevelFileSummary(FileSummaryStorage* scratch,
int level) const {
int len = snprintf(scratch->buffer, sizeof(scratch->buffer), "files_size[");
for (const auto& f : files_[level]) {
int sz = sizeof(scratch->buffer) - len;
char sztxt[16];
AppendHumanBytes(f->fd.GetFileSize(), sztxt, sizeof(sztxt));
int ret = snprintf(scratch->buffer + len, sz,
"#%" PRIu64 "(seq=%" PRIu64 ",sz=%s,%d) ",
f->fd.GetNumber(), f->fd.smallest_seqno, sztxt,
static_cast<int>(f->being_compacted));
if (ret < 0 || ret >= sz)
break;
len += ret;
}
// overwrite the last space (only if files_[level].size() is non-zero)
if (files_[level].size() && len > 0) {
--len;
}
snprintf(scratch->buffer + len, sizeof(scratch->buffer) - len, "]");
return scratch->buffer;
}
uint64_t VersionStorageInfo::MaxNextLevelOverlappingBytes() {
uint64_t result = 0;
std::vector<FileMetaData*> overlaps;
for (int level = 1; level < num_levels() - 1; level++) {
for (const auto& f : files_[level]) {
GetOverlappingInputs(level + 1, &f->smallest, &f->largest, &overlaps);
const uint64_t sum = TotalFileSize(overlaps);
if (sum > result) {
result = sum;
}
}
}
return result;
}
uint64_t VersionStorageInfo::MaxBytesForLevel(int level) const {
// Note: the result for level zero is not really used since we set
// the level-0 compaction threshold based on number of files.
assert(level >= 0);
assert(level < static_cast<int>(level_max_bytes_.size()));
return level_max_bytes_[level];
}
void VersionStorageInfo::CalculateBaseBytes(const ImmutableOptions& ioptions,
const MutableCFOptions& options) {
// Special logic to set number of sorted runs.
// It is to match the previous behavior when all files are in L0.
int num_l0_count = static_cast<int>(files_[0].size());
if (compaction_style_ == kCompactionStyleUniversal) {
// For universal compaction, we use level0 score to indicate
// compaction score for the whole DB. Adding other levels as if
// they are L0 files.
for (int i = 1; i < num_levels(); i++) {
if (!files_[i].empty()) {
num_l0_count++;
}
}
}
set_l0_delay_trigger_count(num_l0_count);
level_max_bytes_.resize(ioptions.num_levels);
if (!ioptions.level_compaction_dynamic_level_bytes) {
base_level_ = (ioptions.compaction_style == kCompactionStyleLevel) ? 1 : -1;
// Calculate for static bytes base case
for (int i = 0; i < ioptions.num_levels; ++i) {
if (i == 0 && ioptions.compaction_style == kCompactionStyleUniversal) {
level_max_bytes_[i] = options.max_bytes_for_level_base;
} else if (i > 1) {
level_max_bytes_[i] = MultiplyCheckOverflow(
MultiplyCheckOverflow(level_max_bytes_[i - 1],
options.max_bytes_for_level_multiplier),
options.MaxBytesMultiplerAdditional(i - 1));
} else {
level_max_bytes_[i] = options.max_bytes_for_level_base;
}
}
} else {
uint64_t max_level_size = 0;
int first_non_empty_level = -1;
// Find size of non-L0 level of most data.
// Cannot use the size of the last level because it can be empty or less
// than previous levels after compaction.
for (int i = 1; i < num_levels_; i++) {
uint64_t total_size = 0;
for (const auto& f : files_[i]) {
total_size += f->fd.GetFileSize();
}
if (total_size > 0 && first_non_empty_level == -1) {
first_non_empty_level = i;
}
if (total_size > max_level_size) {
max_level_size = total_size;
}
}
// Prefill every level's max bytes to disallow compaction from there.
for (int i = 0; i < num_levels_; i++) {
level_max_bytes_[i] = std::numeric_limits<uint64_t>::max();
}
if (max_level_size == 0) {
// No data for L1 and up. L0 compacts to last level directly.
// No compaction from L1+ needs to be scheduled.
base_level_ = num_levels_ - 1;
} else {
uint64_t l0_size = 0;
for (const auto& f : files_[0]) {
l0_size += f->fd.GetFileSize();
}
uint64_t base_bytes_max =
std::max(options.max_bytes_for_level_base, l0_size);
uint64_t base_bytes_min = static_cast<uint64_t>(
base_bytes_max / options.max_bytes_for_level_multiplier);
// Try whether we can make last level's target size to be max_level_size
uint64_t cur_level_size = max_level_size;
for (int i = num_levels_ - 2; i >= first_non_empty_level; i--) {
// Round up after dividing
cur_level_size = static_cast<uint64_t>(
cur_level_size / options.max_bytes_for_level_multiplier);
}
// Calculate base level and its size.
uint64_t base_level_size;
if (cur_level_size <= base_bytes_min) {
// Case 1. If we make target size of last level to be max_level_size,
// target size of the first non-empty level would be smaller than
// base_bytes_min. We set it be base_bytes_min.
base_level_size = base_bytes_min + 1U;
base_level_ = first_non_empty_level;
ROCKS_LOG_INFO(ioptions.logger,
"More existing levels in DB than needed. "
"max_bytes_for_level_multiplier may not be guaranteed.");
} else {
// Find base level (where L0 data is compacted to).
base_level_ = first_non_empty_level;
while (base_level_ > 1 && cur_level_size > base_bytes_max) {
--base_level_;
cur_level_size = static_cast<uint64_t>(
cur_level_size / options.max_bytes_for_level_multiplier);
}
if (cur_level_size > base_bytes_max) {
// Even L1 will be too large
assert(base_level_ == 1);
base_level_size = base_bytes_max;
} else {
base_level_size = cur_level_size;
}
}
level_multiplier_ = options.max_bytes_for_level_multiplier;
assert(base_level_size > 0);
if (l0_size > base_level_size &&
(l0_size > options.max_bytes_for_level_base ||
static_cast<int>(files_[0].size() / 2) >=
options.level0_file_num_compaction_trigger)) {
// We adjust the base level according to actual L0 size, and adjust
// the level multiplier accordingly, when:
// 1. the L0 size is larger than level size base, or
// 2. number of L0 files reaches twice the L0->L1 compaction trigger
// We don't do this otherwise to keep the LSM-tree structure stable
// unless the L0 compaction is backlogged.
base_level_size = l0_size;
if (base_level_ == num_levels_ - 1) {
level_multiplier_ = 1.0;
} else {
level_multiplier_ = std::pow(
static_cast<double>(max_level_size) /
static_cast<double>(base_level_size),
1.0 / static_cast<double>(num_levels_ - base_level_ - 1));
}
}
uint64_t level_size = base_level_size;
for (int i = base_level_; i < num_levels_; i++) {
if (i > base_level_) {
level_size = MultiplyCheckOverflow(level_size, level_multiplier_);
}
// Don't set any level below base_bytes_max. Otherwise, the LSM can
// assume an hourglass shape where L1+ sizes are smaller than L0. This
// causes compaction scoring, which depends on level sizes, to favor L1+
// at the expense of L0, which may fill up and stall.
level_max_bytes_[i] = std::max(level_size, base_bytes_max);
}
}
}
}
uint64_t VersionStorageInfo::EstimateLiveDataSize() const {
// Estimate the live data size by adding up the size of a maximal set of
// sst files with no range overlap in same or higher level. The less
// compacted, the more optimistic (smaller) this estimate is. Also,
// for multiple sorted runs within a level, file order will matter.
uint64_t size = 0;
auto ikey_lt = [this](InternalKey* x, InternalKey* y) {
return internal_comparator_->Compare(*x, *y) < 0;
};
// (Ordered) map of largest keys in files being included in size estimate
std::map<InternalKey*, FileMetaData*, decltype(ikey_lt)> ranges(ikey_lt);
for (int l = num_levels_ - 1; l >= 0; l--) {
bool found_end = false;
for (auto file : files_[l]) {
// Find the first file already included with largest key is larger than
// the smallest key of `file`. If that file does not overlap with the
// current file, none of the files in the map does. If there is
// no potential overlap, we can safely insert the rest of this level
// (if the level is not 0) into the map without checking again because
// the elements in the level are sorted and non-overlapping.
auto lb = (found_end && l != 0) ?
ranges.end() : ranges.lower_bound(&file->smallest);
found_end = (lb == ranges.end());
if (found_end || internal_comparator_->Compare(
file->largest, (*lb).second->smallest) < 0) {
ranges.emplace_hint(lb, &file->largest, file);
size += file->fd.file_size;
}
}
}
Add DB properties for BlobDB (#8734) Summary: RocksDB exposes certain internal statistics via the DB property interface. However, there are currently no properties related to BlobDB. For starters, we would like to add the following BlobDB properties: `rocksdb.num-blob-files`: number of blob files in the current Version (kind of like `num-files-at-level` but note this is not per level, since blob files are not part of the LSM tree). `rocksdb.blob-stats`: this could return the total number and size of all blob files, and potentially also the total amount of garbage (in bytes) in the blob files in the current Version. `rocksdb.total-blob-file-size`: the total size of all blob files (as a blob counterpart for `total-sst-file-size`) of all Versions. `rocksdb.live-blob-file-size`: the total size of all blob files in the current Version. `rocksdb.estimate-live-data-size`: this is actually an existing property that we can extend so it considers blob files as well. When it comes to blobs, we actually have an exact value for live bytes. Namely, live bytes can be computed simply as total bytes minus garbage bytes, summed over the entire set of blob files in the Version. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8734 Test Plan: ``` ➜ rocksdb git:(new_feature_blobDB_properties) ./db_blob_basic_test [==========] Running 16 tests from 2 test cases. [----------] Global test environment set-up. [----------] 10 tests from DBBlobBasicTest [ RUN ] DBBlobBasicTest.GetBlob [ OK ] DBBlobBasicTest.GetBlob (12 ms) [ RUN ] DBBlobBasicTest.MultiGetBlobs [ OK ] DBBlobBasicTest.MultiGetBlobs (11 ms) [ RUN ] DBBlobBasicTest.GetBlob_CorruptIndex [ OK ] DBBlobBasicTest.GetBlob_CorruptIndex (10 ms) [ RUN ] DBBlobBasicTest.GetBlob_InlinedTTLIndex [ OK ] DBBlobBasicTest.GetBlob_InlinedTTLIndex (12 ms) [ RUN ] DBBlobBasicTest.GetBlob_IndexWithInvalidFileNumber [ OK ] DBBlobBasicTest.GetBlob_IndexWithInvalidFileNumber (9 ms) [ RUN ] DBBlobBasicTest.GenerateIOTracing [ OK ] DBBlobBasicTest.GenerateIOTracing (11 ms) [ RUN ] DBBlobBasicTest.BestEffortsRecovery_MissingNewestBlobFile [ OK ] DBBlobBasicTest.BestEffortsRecovery_MissingNewestBlobFile (13 ms) [ RUN ] DBBlobBasicTest.GetMergeBlobWithPut [ OK ] DBBlobBasicTest.GetMergeBlobWithPut (11 ms) [ RUN ] DBBlobBasicTest.MultiGetMergeBlobWithPut [ OK ] DBBlobBasicTest.MultiGetMergeBlobWithPut (14 ms) [ RUN ] DBBlobBasicTest.BlobDBProperties [ OK ] DBBlobBasicTest.BlobDBProperties (21 ms) [----------] 10 tests from DBBlobBasicTest (124 ms total) [----------] 6 tests from DBBlobBasicTest/DBBlobBasicIOErrorTest [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/0 (12 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/1 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/0 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/1 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/0 (1011 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/1 (1013 ms) [----------] 6 tests from DBBlobBasicTest/DBBlobBasicIOErrorTest (2066 ms total) [----------] Global test environment tear-down [==========] 16 tests from 2 test cases ran. (2190 ms total) [ PASSED ] 16 tests. ``` Reviewed By: ltamasi Differential Revision: D30690849 Pulled By: Zhiyi-Zhang fbshipit-source-id: a7567319487ad76bd1a2e24bf143afdbbd9e4346
2021-09-08 21:19:01 +02:00
// For BlobDB, the result also includes the exact value of live bytes in the
// blob files of the version.
const auto& blobFiles = GetBlobFiles();
for (const auto& pair : blobFiles) {
const auto& meta = pair.second;
size += meta->GetTotalBlobBytes();
size -= meta->GetGarbageBlobBytes();
}
return size;
}
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
bool VersionStorageInfo::RangeMightExistAfterSortedRun(
const Slice& smallest_user_key, const Slice& largest_user_key,
int last_level, int last_l0_idx) {
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
assert((last_l0_idx != -1) == (last_level == 0));
// TODO(ajkr): this preserves earlier behavior where we considered an L0 file
// bottommost only if it's the oldest L0 file and there are no files on older
// levels. It'd be better to consider it bottommost if there's no overlap in
// older levels/files.
if (last_level == 0 &&
last_l0_idx != static_cast<int>(LevelFiles(0).size() - 1)) {
return true;
}
// Checks whether there are files living beyond the `last_level`. If lower
// levels have files, it checks for overlap between [`smallest_key`,
// `largest_key`] and those files. Bottomlevel optimizations can be made if
// there are no files in lower levels or if there is no overlap with the files
// in the lower levels.
for (int level = last_level + 1; level < num_levels(); level++) {
// The range is not in the bottommost level if there are files in lower
// levels when the `last_level` is 0 or if there are files in lower levels
// which overlap with [`smallest_key`, `largest_key`].
if (files_[level].size() > 0 &&
(last_level == 0 ||
OverlapInLevel(level, &smallest_user_key, &largest_user_key))) {
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
2017-10-26 01:24:29 +02:00
return true;
}
}
return false;
}
void Version::AddLiveFiles(std::vector<uint64_t>* live_table_files,
std::vector<uint64_t>* live_blob_files) const {
assert(live_table_files);
assert(live_blob_files);
for (int level = 0; level < storage_info_.num_levels(); ++level) {
const auto& level_files = storage_info_.LevelFiles(level);
for (const auto& meta : level_files) {
assert(meta);
live_table_files->emplace_back(meta->fd.GetNumber());
}
}
const auto& blob_files = storage_info_.GetBlobFiles();
for (const auto& pair : blob_files) {
const auto& meta = pair.second;
assert(meta);
live_blob_files->emplace_back(meta->GetBlobFileNumber());
}
}
std::string Version::DebugString(bool hex, bool print_stats) const {
std::string r;
for (int level = 0; level < storage_info_.num_levels_; level++) {
// E.g.,
// --- level 1 ---
// 17:123[1 .. 124]['a' .. 'd']
// 20:43[124 .. 128]['e' .. 'g']
//
// if print_stats=true:
// 17:123[1 .. 124]['a' .. 'd'](4096)
r.append("--- level ");
AppendNumberTo(&r, level);
r.append(" --- version# ");
AppendNumberTo(&r, version_number_);
r.append(" ---\n");
const std::vector<FileMetaData*>& files = storage_info_.files_[level];
for (size_t i = 0; i < files.size(); i++) {
r.push_back(' ');
AppendNumberTo(&r, files[i]->fd.GetNumber());
r.push_back(':');
AppendNumberTo(&r, files[i]->fd.GetFileSize());
r.append("[");
AppendNumberTo(&r, files[i]->fd.smallest_seqno);
r.append(" .. ");
AppendNumberTo(&r, files[i]->fd.largest_seqno);
r.append("]");
r.append("[");
r.append(files[i]->smallest.DebugString(hex));
r.append(" .. ");
r.append(files[i]->largest.DebugString(hex));
r.append("]");
if (files[i]->oldest_blob_file_number != kInvalidBlobFileNumber) {
r.append(" blob_file:");
AppendNumberTo(&r, files[i]->oldest_blob_file_number);
}
if (print_stats) {
r.append("(");
r.append(ToString(
files[i]->stats.num_reads_sampled.load(std::memory_order_relaxed)));
r.append(")");
}
r.append("\n");
}
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
const auto& blob_files = storage_info_.GetBlobFiles();
if (!blob_files.empty()) {
r.append("--- blob files --- version# ");
AppendNumberTo(&r, version_number_);
r.append(" ---\n");
for (const auto& pair : blob_files) {
const auto& blob_file_meta = pair.second;
assert(blob_file_meta);
r.append(blob_file_meta->DebugString());
r.push_back('\n');
}
}
return r;
}
// this is used to batch writes to the manifest file
struct VersionSet::ManifestWriter {
Status status;
bool done;
InstrumentedCondVar cv;
ColumnFamilyData* cfd;
const MutableCFOptions mutable_cf_options;
group multiple batch of flush into one manifest file (one call to LogAndApply) Summary: Currently, if several flush outputs are committed together, we issue each manifest write per batch (1 batch = 1 flush = 1 sst file = 1+ continuous memtables). Each manifest write requires one fsync and one fsync to parent directory. In some cases, it becomes the bottleneck of write. We should batch them and write in one manifest write when possible. Test Plan: ` ./db_bench -benchmarks="fillseq" -max_write_buffer_number=16 -max_background_flushes=16 -disable_auto_compactions=true -min_write_buffer_number_to_merge=1 -write_buffer_size=65536 -level0_stop_writes_trigger=10000 -level0_slowdown_writes_trigger=10000` **Before** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:38:17 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 166.277 micros/op 6014 ops/sec; 0.7 MB/s ``` **After** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:35:05 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 52.328 micros/op 19110 ops/sec; 2.1 MB/s ``` Reviewers: andrewkr, IslamAbdelRahman, yhchiang, sdong Reviewed By: sdong Subscribers: igor, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D60075
2016-07-06 03:09:59 +02:00
const autovector<VersionEdit*>& edit_list;
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
const std::function<void(const Status&)> manifest_write_callback;
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
explicit ManifestWriter(
InstrumentedMutex* mu, ColumnFamilyData* _cfd,
const MutableCFOptions& cf_options, const autovector<VersionEdit*>& e,
const std::function<void(const Status&)>& manifest_wcb)
: done(false),
cv(mu),
cfd(_cfd),
mutable_cf_options(cf_options),
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
edit_list(e),
manifest_write_callback(manifest_wcb) {}
~ManifestWriter() { status.PermitUncheckedError(); }
bool IsAllWalEdits() const {
bool all_wal_edits = true;
for (const auto& e : edit_list) {
if (!e->IsWalManipulation()) {
all_wal_edits = false;
break;
}
}
return all_wal_edits;
}
};
Status AtomicGroupReadBuffer::AddEdit(VersionEdit* edit) {
assert(edit);
if (edit->is_in_atomic_group_) {
TEST_SYNC_POINT("AtomicGroupReadBuffer::AddEdit:AtomicGroup");
if (replay_buffer_.empty()) {
replay_buffer_.resize(edit->remaining_entries_ + 1);
TEST_SYNC_POINT_CALLBACK(
"AtomicGroupReadBuffer::AddEdit:FirstInAtomicGroup", edit);
}
read_edits_in_atomic_group_++;
if (read_edits_in_atomic_group_ + edit->remaining_entries_ !=
static_cast<uint32_t>(replay_buffer_.size())) {
TEST_SYNC_POINT_CALLBACK(
"AtomicGroupReadBuffer::AddEdit:IncorrectAtomicGroupSize", edit);
return Status::Corruption("corrupted atomic group");
}
replay_buffer_[read_edits_in_atomic_group_ - 1] = *edit;
if (read_edits_in_atomic_group_ == replay_buffer_.size()) {
TEST_SYNC_POINT_CALLBACK(
"AtomicGroupReadBuffer::AddEdit:LastInAtomicGroup", edit);
return Status::OK();
}
return Status::OK();
}
// A normal edit.
if (!replay_buffer().empty()) {
TEST_SYNC_POINT_CALLBACK(
"AtomicGroupReadBuffer::AddEdit:AtomicGroupMixedWithNormalEdits", edit);
return Status::Corruption("corrupted atomic group");
}
return Status::OK();
}
bool AtomicGroupReadBuffer::IsFull() const {
return read_edits_in_atomic_group_ == replay_buffer_.size();
}
bool AtomicGroupReadBuffer::IsEmpty() const { return replay_buffer_.empty(); }
void AtomicGroupReadBuffer::Clear() {
read_edits_in_atomic_group_ = 0;
replay_buffer_.clear();
}
VersionSet::VersionSet(const std::string& dbname,
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
const ImmutableDBOptions* _db_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& storage_options, Cache* table_cache,
WriteBufferManager* write_buffer_manager,
WriteController* write_controller,
BlockCacheTracer* const block_cache_tracer,
const std::shared_ptr<IOTracer>& io_tracer,
const std::string& db_session_id)
: column_family_set_(
new ColumnFamilySet(dbname, _db_options, storage_options, table_cache,
write_buffer_manager, write_controller,
block_cache_tracer, io_tracer, db_session_id)),
table_cache_(table_cache),
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
env_(_db_options->env),
fs_(_db_options->fs, io_tracer),
clock_(_db_options->clock),
dbname_(dbname),
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
db_options_(_db_options),
next_file_number_(2),
manifest_file_number_(0), // Filled by Recover()
options_file_number_(0),
pending_manifest_file_number_(0),
last_sequence_(0),
last_allocated_sequence_(0),
last_published_sequence_(0),
prev_log_number_(0),
current_version_number_(0),
manifest_file_size_(0),
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
file_options_(storage_options),
block_cache_tracer_(block_cache_tracer),
io_tracer_(io_tracer),
db_session_id_(db_session_id) {}
VersionSet::~VersionSet() {
// we need to delete column_family_set_ because its destructor depends on
// VersionSet
column_family_set_.reset();
for (auto& file : obsolete_files_) {
if (file.metadata->table_reader_handle) {
table_cache_->Release(file.metadata->table_reader_handle);
TableCache::Evict(table_cache_, file.metadata->fd.GetNumber());
}
file.DeleteMetadata();
}
obsolete_files_.clear();
io_status_.PermitUncheckedError();
}
void VersionSet::Reset() {
if (column_family_set_) {
WriteBufferManager* wbm = column_family_set_->write_buffer_manager();
WriteController* wc = column_family_set_->write_controller();
column_family_set_.reset(new ColumnFamilySet(
dbname_, db_options_, file_options_, table_cache_, wbm, wc,
block_cache_tracer_, io_tracer_, db_session_id_));
}
db_id_.clear();
next_file_number_.store(2);
min_log_number_to_keep_2pc_.store(0);
manifest_file_number_ = 0;
options_file_number_ = 0;
pending_manifest_file_number_ = 0;
last_sequence_.store(0);
last_allocated_sequence_.store(0);
last_published_sequence_.store(0);
prev_log_number_ = 0;
descriptor_log_.reset();
current_version_number_ = 0;
manifest_writers_.clear();
manifest_file_size_ = 0;
obsolete_files_.clear();
obsolete_manifests_.clear();
Define WAL related classes to be used in VersionEdit and VersionSet (#7164) Summary: `WalAddition`, `WalDeletion` are defined in `wal_version.h` and used in `VersionEdit`. `WalAddition` is used to represent events of creating a new WAL (no size, just log number), or closing a WAL (with size). `WalDeletion` is used to represent events of deleting or archiving a WAL, it means the WAL is no longer alive (won't be replayed during recovery). `WalSet` is the set of alive WALs kept in `VersionSet`. 1. Why use `WalDeletion` instead of relying on `MinLogNumber` to identify outdated WALs On recovery, we can compute `MinLogNumber()` based on the log numbers kept in MANIFEST, any log with number < MinLogNumber can be ignored. So it seems that we don't need to persist `WalDeletion` to MANIFEST, since we can ignore the WALs based on MinLogNumber. But the `MinLogNumber()` is actually a lower bound, it does not exactly mean that logs starting from MinLogNumber must exist. This is because in a corner case, when a column family is empty and never flushed, its log number is set to the largest log number, but not persisted in MANIFEST. So let's say there are 2 column families, when creating the DB, the first WAL has log number 1, so it's persisted to MANIFEST for both column families. Then CF 0 is empty and never flushed, CF 1 is updated and flushed, so a new WAL with log number 2 is created and persisted to MANIFEST for CF 1. But CF 0's log number in MANIFEST is still 1. So on recovery, MinLogNumber is 1, but since log 1 only contains data for CF 1, and CF 1 is flushed, log 1 might have already been deleted from disk. We can make `MinLogNumber()` be the exactly minimum log number that must exist, by persisting the most recent log number for empty column families that are not flushed. But if there are N such column families, then every time a new WAL is created, we need to add N records to MANIFEST. In current design, a record is persisted to MANIFEST only when WAL is created, closed, or deleted/archived, so the number of WAL related records are bounded to 3x number of WALs. 2. Why keep `WalSet` in `VersionSet` instead of applying the `VersionEdit`s to `VersionStorageInfo` `VersionEdit`s are originally designed to track the addition and deletion of SST files. The SST files are related to column families, each column family has a list of `Version`s, and each `Version` keeps the set of active SST files in `VersionStorageInfo`. But WALs are a concept of DB, they are not bounded to specific column families. So logically it does not make sense to store WALs in a column family's `Version`s. Also, `Version`'s purpose is to keep reference to SST / blob files, so that they are not deleted until there is no version referencing them. But a WAL is deleted regardless of version references. So we keep the WALs in `VersionSet` for the purpose of writing out the DB state's snapshot when creating new MANIFESTs. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7164 Test Plan: make version_edit_test && ./version_edit_test make wal_edit_test && ./wal_edit_test Reviewed By: ltamasi Differential Revision: D22677936 Pulled By: cheng-chang fbshipit-source-id: 5a3b6890140e572ffd79eb37e6e4c3c32361a859
2020-08-06 01:32:26 +02:00
wals_.Reset();
}
void VersionSet::AppendVersion(ColumnFamilyData* column_family_data,
Version* v) {
// compute new compaction score
v->storage_info()->ComputeCompactionScore(
*column_family_data->ioptions(),
*column_family_data->GetLatestMutableCFOptions());
// Mark v finalized
v->storage_info_.SetFinalized();
// Make "v" current
assert(v->refs_ == 0);
Version* current = column_family_data->current();
assert(v != current);
if (current != nullptr) {
assert(current->refs_ > 0);
current->Unref();
}
column_family_data->SetCurrent(v);
v->Ref();
// Append to linked list
v->prev_ = column_family_data->dummy_versions()->prev_;
v->next_ = column_family_data->dummy_versions();
v->prev_->next_ = v;
v->next_->prev_ = v;
}
Status VersionSet::ProcessManifestWrites(
std::deque<ManifestWriter>& writers, InstrumentedMutex* mu,
FSDirectory* db_directory, bool new_descriptor_log,
const ColumnFamilyOptions* new_cf_options) {
mu->AssertHeld();
assert(!writers.empty());
ManifestWriter& first_writer = writers.front();
ManifestWriter* last_writer = &first_writer;
assert(!manifest_writers_.empty());
assert(manifest_writers_.front() == &first_writer);
group multiple batch of flush into one manifest file (one call to LogAndApply) Summary: Currently, if several flush outputs are committed together, we issue each manifest write per batch (1 batch = 1 flush = 1 sst file = 1+ continuous memtables). Each manifest write requires one fsync and one fsync to parent directory. In some cases, it becomes the bottleneck of write. We should batch them and write in one manifest write when possible. Test Plan: ` ./db_bench -benchmarks="fillseq" -max_write_buffer_number=16 -max_background_flushes=16 -disable_auto_compactions=true -min_write_buffer_number_to_merge=1 -write_buffer_size=65536 -level0_stop_writes_trigger=10000 -level0_slowdown_writes_trigger=10000` **Before** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:38:17 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 166.277 micros/op 6014 ops/sec; 0.7 MB/s ``` **After** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:35:05 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 52.328 micros/op 19110 ops/sec; 2.1 MB/s ``` Reviewers: andrewkr, IslamAbdelRahman, yhchiang, sdong Reviewed By: sdong Subscribers: igor, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D60075
2016-07-06 03:09:59 +02:00
autovector<VersionEdit*> batch_edits;
autovector<Version*> versions;
autovector<const MutableCFOptions*> mutable_cf_options_ptrs;
std::vector<std::unique_ptr<BaseReferencedVersionBuilder>> builder_guards;
if (first_writer.edit_list.front()->IsColumnFamilyManipulation()) {
// No group commits for column family add or drop
LogAndApplyCFHelper(first_writer.edit_list.front());
batch_edits.push_back(first_writer.edit_list.front());
} else {
auto it = manifest_writers_.cbegin();
size_t group_start = std::numeric_limits<size_t>::max();
while (it != manifest_writers_.cend()) {
if ((*it)->edit_list.front()->IsColumnFamilyManipulation()) {
// no group commits for column family add or drop
break;
}
last_writer = *(it++);
assert(last_writer != nullptr);
assert(last_writer->cfd != nullptr);
if (last_writer->cfd->IsDropped()) {
// If we detect a dropped CF at this point, and the corresponding
// version edits belong to an atomic group, then we need to find out
// the preceding version edits in the same atomic group, and update
// their `remaining_entries_` member variable because we are NOT going
// to write the version edits' of dropped CF to the MANIFEST. If we
// don't update, then Recover can report corrupted atomic group because
// the `remaining_entries_` do not match.
if (!batch_edits.empty()) {
if (batch_edits.back()->is_in_atomic_group_ &&
batch_edits.back()->remaining_entries_ > 0) {
assert(group_start < batch_edits.size());
const auto& edit_list = last_writer->edit_list;
size_t k = 0;
while (k < edit_list.size()) {
if (!edit_list[k]->is_in_atomic_group_) {
break;
} else if (edit_list[k]->remaining_entries_ == 0) {
++k;
break;
}
++k;
}
for (auto i = group_start; i < batch_edits.size(); ++i) {
assert(static_cast<uint32_t>(k) <=
batch_edits.back()->remaining_entries_);
batch_edits[i]->remaining_entries_ -= static_cast<uint32_t>(k);
}
}
}
continue;
}
// We do a linear search on versions because versions is small.
// TODO(yanqin) maybe consider unordered_map
Version* version = nullptr;
VersionBuilder* builder = nullptr;
for (int i = 0; i != static_cast<int>(versions.size()); ++i) {
uint32_t cf_id = last_writer->cfd->GetID();
if (versions[i]->cfd()->GetID() == cf_id) {
version = versions[i];
assert(!builder_guards.empty() &&
builder_guards.size() == versions.size());
builder = builder_guards[i]->version_builder();
TEST_SYNC_POINT_CALLBACK(
"VersionSet::ProcessManifestWrites:SameColumnFamily", &cf_id);
break;
}
}
if (version == nullptr) {
// WAL manipulations do not need to be applied to versions.
if (!last_writer->IsAllWalEdits()) {
version = new Version(last_writer->cfd, this, file_options_,
last_writer->mutable_cf_options, io_tracer_,
current_version_number_++);
versions.push_back(version);
mutable_cf_options_ptrs.push_back(&last_writer->mutable_cf_options);
builder_guards.emplace_back(
new BaseReferencedVersionBuilder(last_writer->cfd));
builder = builder_guards.back()->version_builder();
}
assert(last_writer->IsAllWalEdits() || builder);
assert(last_writer->IsAllWalEdits() || version);
TEST_SYNC_POINT_CALLBACK("VersionSet::ProcessManifestWrites:NewVersion",
version);
}
for (const auto& e : last_writer->edit_list) {
if (e->is_in_atomic_group_) {
if (batch_edits.empty() || !batch_edits.back()->is_in_atomic_group_ ||
(batch_edits.back()->is_in_atomic_group_ &&
batch_edits.back()->remaining_entries_ == 0)) {
group_start = batch_edits.size();
}
} else if (group_start != std::numeric_limits<size_t>::max()) {
group_start = std::numeric_limits<size_t>::max();
}
Status s = LogAndApplyHelper(last_writer->cfd, builder, e, mu);
if (!s.ok()) {
// free up the allocated memory
for (auto v : versions) {
delete v;
}
return s;
}
batch_edits.push_back(e);
group multiple batch of flush into one manifest file (one call to LogAndApply) Summary: Currently, if several flush outputs are committed together, we issue each manifest write per batch (1 batch = 1 flush = 1 sst file = 1+ continuous memtables). Each manifest write requires one fsync and one fsync to parent directory. In some cases, it becomes the bottleneck of write. We should batch them and write in one manifest write when possible. Test Plan: ` ./db_bench -benchmarks="fillseq" -max_write_buffer_number=16 -max_background_flushes=16 -disable_auto_compactions=true -min_write_buffer_number_to_merge=1 -write_buffer_size=65536 -level0_stop_writes_trigger=10000 -level0_slowdown_writes_trigger=10000` **Before** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:38:17 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 166.277 micros/op 6014 ops/sec; 0.7 MB/s ``` **After** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:35:05 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 52.328 micros/op 19110 ops/sec; 2.1 MB/s ``` Reviewers: andrewkr, IslamAbdelRahman, yhchiang, sdong Reviewed By: sdong Subscribers: igor, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D60075
2016-07-06 03:09:59 +02:00
}
}
for (int i = 0; i < static_cast<int>(versions.size()); ++i) {
assert(!builder_guards.empty() &&
builder_guards.size() == versions.size());
auto* builder = builder_guards[i]->version_builder();
Status s = builder->SaveTo(versions[i]->storage_info());
if (!s.ok()) {
// free up the allocated memory
for (auto v : versions) {
delete v;
}
return s;
}
}
}
#ifndef NDEBUG
// Verify that version edits of atomic groups have correct
// remaining_entries_.
size_t k = 0;
while (k < batch_edits.size()) {
while (k < batch_edits.size() && !batch_edits[k]->is_in_atomic_group_) {
++k;
}
if (k == batch_edits.size()) {
break;
}
size_t i = k;
while (i < batch_edits.size()) {
if (!batch_edits[i]->is_in_atomic_group_) {
break;
}
assert(i - k + batch_edits[i]->remaining_entries_ ==
batch_edits[k]->remaining_entries_);
if (batch_edits[i]->remaining_entries_ == 0) {
++i;
break;
}
++i;
}
assert(batch_edits[i - 1]->is_in_atomic_group_);
assert(0 == batch_edits[i - 1]->remaining_entries_);
std::vector<VersionEdit*> tmp;
for (size_t j = k; j != i; ++j) {
tmp.emplace_back(batch_edits[j]);
}
TEST_SYNC_POINT_CALLBACK(
"VersionSet::ProcessManifestWrites:CheckOneAtomicGroup", &tmp);
k = i;
}
#endif // NDEBUG
assert(pending_manifest_file_number_ == 0);
if (!descriptor_log_ ||
manifest_file_size_ > db_options_->max_manifest_file_size) {
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
TEST_SYNC_POINT("VersionSet::ProcessManifestWrites:BeforeNewManifest");
new_descriptor_log = true;
} else {
pending_manifest_file_number_ = manifest_file_number_;
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
2020-01-07 05:08:24 +01:00
// Local cached copy of state variable(s). WriteCurrentStateToManifest()
// reads its content after releasing db mutex to avoid race with
// SwitchMemtable().
std::unordered_map<uint32_t, MutableCFState> curr_state;
VersionEdit wal_additions;
if (new_descriptor_log) {
pending_manifest_file_number_ = NewFileNumber();
batch_edits.back()->SetNextFile(next_file_number_.load());
// if we are writing out new snapshot make sure to persist max column
// family.
2014-03-13 02:09:03 +01:00
if (column_family_set_->GetMaxColumnFamily() > 0) {
first_writer.edit_list.front()->SetMaxColumnFamily(
group multiple batch of flush into one manifest file (one call to LogAndApply) Summary: Currently, if several flush outputs are committed together, we issue each manifest write per batch (1 batch = 1 flush = 1 sst file = 1+ continuous memtables). Each manifest write requires one fsync and one fsync to parent directory. In some cases, it becomes the bottleneck of write. We should batch them and write in one manifest write when possible. Test Plan: ` ./db_bench -benchmarks="fillseq" -max_write_buffer_number=16 -max_background_flushes=16 -disable_auto_compactions=true -min_write_buffer_number_to_merge=1 -write_buffer_size=65536 -level0_stop_writes_trigger=10000 -level0_slowdown_writes_trigger=10000` **Before** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:38:17 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 166.277 micros/op 6014 ops/sec; 0.7 MB/s ``` **After** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:35:05 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 52.328 micros/op 19110 ops/sec; 2.1 MB/s ``` Reviewers: andrewkr, IslamAbdelRahman, yhchiang, sdong Reviewed By: sdong Subscribers: igor, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D60075
2016-07-06 03:09:59 +02:00
column_family_set_->GetMaxColumnFamily());
2014-03-13 02:09:03 +01:00
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
2020-01-07 05:08:24 +01:00
for (const auto* cfd : *column_family_set_) {
assert(curr_state.find(cfd->GetID()) == curr_state.end());
curr_state.emplace(std::make_pair(
cfd->GetID(),
MutableCFState(cfd->GetLogNumber(), cfd->GetFullHistoryTsLow())));
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
2020-01-07 05:08:24 +01:00
}
for (const auto& wal : wals_.GetWals()) {
wal_additions.AddWal(wal.first, wal.second);
}
}
uint64_t new_manifest_file_size = 0;
Status s;
IOStatus io_s;
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
IOStatus manifest_io_status;
{
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
FileOptions opt_file_opts = fs_->OptimizeForManifestWrite(file_options_);
mu->Unlock();
TEST_SYNC_POINT("VersionSet::LogAndApply:WriteManifestStart");
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
TEST_SYNC_POINT_CALLBACK("VersionSet::LogAndApply:WriteManifest", nullptr);
if (!first_writer.edit_list.front()->IsColumnFamilyManipulation()) {
for (int i = 0; i < static_cast<int>(versions.size()); ++i) {
assert(!builder_guards.empty() &&
builder_guards.size() == versions.size());
assert(!mutable_cf_options_ptrs.empty() &&
builder_guards.size() == versions.size());
ColumnFamilyData* cfd = versions[i]->cfd_;
s = builder_guards[i]->version_builder()->LoadTableHandlers(
cfd->internal_stats(), 1 /* max_threads */,
true /* prefetch_index_and_filter_in_cache */,
false /* is_initial_load */,
mutable_cf_options_ptrs[i]->prefix_extractor.get(),
MaxFileSizeForL0MetaPin(*mutable_cf_options_ptrs[i]));
if (!s.ok()) {
if (db_options_->paranoid_checks) {
break;
}
s = Status::OK();
}
}
}
if (s.ok() && new_descriptor_log) {
// This is fine because everything inside of this block is serialized --
// only one thread can be here at the same time
// create new manifest file
ROCKS_LOG_INFO(db_options_->info_log, "Creating manifest %" PRIu64 "\n",
pending_manifest_file_number_);
std::string descriptor_fname =
DescriptorFileName(dbname_, pending_manifest_file_number_);
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
std::unique_ptr<FSWritableFile> descriptor_file;
io_s = NewWritableFile(fs_.get(), descriptor_fname, &descriptor_file,
opt_file_opts);
if (io_s.ok()) {
descriptor_file->SetPreallocationBlockSize(
db_options_->manifest_preallocation_size);
FileTypeSet tmp_set = db_options_->checksum_handoff_file_types;
std::unique_ptr<WritableFileWriter> file_writer(new WritableFileWriter(
std::move(descriptor_file), descriptor_fname, opt_file_opts, clock_,
io_tracer_, nullptr, db_options_->listeners, nullptr,
Using existing crc32c checksum in checksum handoff for Manifest and WAL (#8412) Summary: In PR https://github.com/facebook/rocksdb/issues/7523 , checksum handoff is introduced in RocksDB for WAL, Manifest, and SST files. When user enable checksum handoff for a certain type of file, before the data is written to the lower layer storage system, we calculate the checksum (crc32c) of each piece of data and pass the checksum down with the data, such that data verification can be down by the lower layer storage system if it has the capability. However, it cannot cover the whole lifetime of the data in the memory and also it potentially introduces extra checksum calculation overhead. In this PR, we introduce a new interface in WritableFileWriter::Append, which allows the caller be able to pass the data and the checksum (crc32c) together. In this way, WritableFileWriter can directly use the pass-in checksum (crc32c) to generate the checksum of data being passed down to the storage system. It saves the calculation overhead and achieves higher protection coverage. When a new checksum is added with the data, we use Crc32cCombine https://github.com/facebook/rocksdb/issues/8305 to combine the existing checksum and the new checksum. To avoid the segmenting of data by rate-limiter before it is stored, rate-limiter is called enough times to accumulate enough credits for a certain write. This design only support Manifest and WAL which use log_writer in the current stage. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8412 Test Plan: make check, add new testing cases. Reviewed By: anand1976 Differential Revision: D29151545 Pulled By: zhichao-cao fbshipit-source-id: 75e2278c5126cfd58393c67b1efd18dcc7a30772
2021-06-25 09:46:33 +02:00
tmp_set.Contains(FileType::kDescriptorFile),
tmp_set.Contains(FileType::kDescriptorFile)));
group multiple batch of flush into one manifest file (one call to LogAndApply) Summary: Currently, if several flush outputs are committed together, we issue each manifest write per batch (1 batch = 1 flush = 1 sst file = 1+ continuous memtables). Each manifest write requires one fsync and one fsync to parent directory. In some cases, it becomes the bottleneck of write. We should batch them and write in one manifest write when possible. Test Plan: ` ./db_bench -benchmarks="fillseq" -max_write_buffer_number=16 -max_background_flushes=16 -disable_auto_compactions=true -min_write_buffer_number_to_merge=1 -write_buffer_size=65536 -level0_stop_writes_trigger=10000 -level0_slowdown_writes_trigger=10000` **Before** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:38:17 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 166.277 micros/op 6014 ops/sec; 0.7 MB/s ``` **After** ``` Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 4.9 Date: Fri Jul 1 15:35:05 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 1 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 52.328 micros/op 19110 ops/sec; 2.1 MB/s ``` Reviewers: andrewkr, IslamAbdelRahman, yhchiang, sdong Reviewed By: sdong Subscribers: igor, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D60075
2016-07-06 03:09:59 +02:00
descriptor_log_.reset(
new log::Writer(std::move(file_writer), 0, false));
s = WriteCurrentStateToManifest(curr_state, wal_additions,
descriptor_log_.get(), io_s);
} else {
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
manifest_io_status = io_s;
s = io_s;
}
}
if (s.ok()) {
if (!first_writer.edit_list.front()->IsColumnFamilyManipulation()) {
for (int i = 0; i < static_cast<int>(versions.size()); ++i) {
versions[i]->PrepareApply(*mutable_cf_options_ptrs[i], true);
}
}
// Write new records to MANIFEST log
#ifndef NDEBUG
size_t idx = 0;
#endif
2014-02-28 21:22:45 +01:00
for (auto& e : batch_edits) {
std::string record;
if (!e->EncodeTo(&record)) {
s = Status::Corruption("Unable to encode VersionEdit:" +
e->DebugString(true));
break;
}
TEST_KILL_RANDOM_WITH_WEIGHT("VersionSet::LogAndApply:BeforeAddRecord",
REDUCE_ODDS2);
#ifndef NDEBUG
if (batch_edits.size() > 1 && batch_edits.size() - 1 == idx) {
TEST_SYNC_POINT_CALLBACK(
"VersionSet::ProcessManifestWrites:BeforeWriteLastVersionEdit:0",
nullptr);
TEST_SYNC_POINT(
"VersionSet::ProcessManifestWrites:BeforeWriteLastVersionEdit:1");
}
++idx;
#endif /* !NDEBUG */
io_s = descriptor_log_->AddRecord(record);
if (!io_s.ok()) {
s = io_s;
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
manifest_io_status = io_s;
break;
}
}
if (s.ok()) {
io_s = SyncManifest(db_options_, descriptor_log_->file());
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
manifest_io_status = io_s;
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
2020-06-25 04:05:47 +02:00
TEST_SYNC_POINT_CALLBACK(
"VersionSet::ProcessManifestWrites:AfterSyncManifest", &io_s);
}
if (!io_s.ok()) {
s = io_s;
ROCKS_LOG_ERROR(db_options_->info_log, "MANIFEST write %s\n",
s.ToString().c_str());
}
}
// If we just created a new descriptor file, install it by writing a
// new CURRENT file that points to it.
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
if (s.ok()) {
assert(manifest_io_status.ok());
}
if (s.ok() && new_descriptor_log) {
io_s = SetCurrentFile(fs_.get(), dbname_, pending_manifest_file_number_,
db_directory);
if (!io_s.ok()) {
s = io_s;
}
}
if (s.ok()) {
// find offset in manifest file where this version is stored.
new_manifest_file_size = descriptor_log_->file()->GetFileSize();
}
if (first_writer.edit_list.front()->is_column_family_drop_) {
TEST_SYNC_POINT("VersionSet::LogAndApply::ColumnFamilyDrop:0");
LogAndApply() should fail if the column family has been dropped Summary: This patch finally fixes the ColumnFamilyTest.ReadDroppedColumnFamily test. The test has been failing very sporadically and it was hard to repro. However, I managed to write a new tests that reproes the failure deterministically. Here's what happens: 1. We start the flush for the column family 2. We check if the column family was dropped here: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/flush_job.cc#L149 3. This check goes through, ends up in InstallMemtableFlushResults() and it goes into LogAndApply() 4. At about this time, we start dropping the column family. Dropping the column family process gets to LogAndApply() at about the same time as LogAndApply() from flush process 5. Drop column family goes through LogAndApply() first, marking the column family as dropped. 6. Flush process gets woken up and gets a chance to write to the MANIFEST. However, this is where it gets stuck: https://github.com/facebook/rocksdb/blob/a3fc49bfddcdb1ff29409aacd06c04df56c7a1d7/db/version_set.cc#L1975 7. We see that the column family was dropped, so there is no need to write to the MANIFEST. We return OK. 8. Flush gets OK back from LogAndApply() and it deletes the memtable, thinking that the data is now safely persisted to sst file. The fix is pretty simple. Instead of OK, we return ShutdownInProgress. This is not really true, but we have been using this status code to also mean "this operation was canceled because the column family has been dropped". The fix is only one LOC. All other code is related to tests. I added a new test that reproes the failure. I also moved SleepingBackgroundTask to util/testutil.h (because I needed it in column_family_test for my new test). There's plenty of other places where we reimplement SleepingBackgroundTask, but I'll address that in a separate commit. Test Plan: 1. new test 2. make check 3. Make sure the ColumnFamilyTest.ReadDroppedColumnFamily doesn't fail on Travis: https://travis-ci.org/facebook/rocksdb/jobs/79952386 Reviewers: yhchiang, anthony, IslamAbdelRahman, kradhakrishnan, rven, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D46773
2015-09-15 20:28:44 +02:00
TEST_SYNC_POINT("VersionSet::LogAndApply::ColumnFamilyDrop:1");
TEST_SYNC_POINT("VersionSet::LogAndApply::ColumnFamilyDrop:2");
}
LogFlush(db_options_->info_log);
Handle concurrent manifest update and backup creation Summary: Fixed two related race conditions in backup creation. (1) CreateNewBackup() uses DB::DisableFileDeletions() to prevent table files from being deleted while it is copying; however, the MANIFEST file could still rotate during this time. The fix is to stop deleting the old manifest in the rotation logic. It will be deleted safely later when PurgeObsoleteFiles() runs (can only happen when file deletions are enabled). (2) CreateNewBackup() did not account for the CURRENT file being mutable. This is significant because the files returned by GetLiveFiles() contain a particular manifest filename, but the manifest to which CURRENT refers can change at any time. This causes problems when CURRENT changes between the call to GetLiveFiles() and when it's copied to the backup directory. To workaround this, I manually forge a CURRENT file referring to the manifest filename returned in GetLiveFiles(). (2) also applies to the checkpointing code, so let me know if this approach is good and I'll make the same change there. Test Plan: new test for roll manifest during backup creation. running the test before this change: $ ./backupable_db_test --gtest_filter=BackupableDBTest.ChangeManifestDuringBackupCreation ... IO error: /tmp/rocksdbtest-9383/backupable_db/MANIFEST-000001: No such file or directory running the test after this change: $ ./backupable_db_test --gtest_filter=BackupableDBTest.ChangeManifestDuringBackupCreation ... [ RUN ] BackupableDBTest.ChangeManifestDuringBackupCreation [ OK ] BackupableDBTest.ChangeManifestDuringBackupCreation (2836 ms) Reviewers: IslamAbdelRahman, anthony, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D54711
2016-02-29 21:56:55 +01:00
TEST_SYNC_POINT("VersionSet::LogAndApply:WriteManifestDone");
mu->Lock();
}
if (s.ok()) {
// Apply WAL edits, DB mutex must be held.
for (auto& e : batch_edits) {
if (e->IsWalAddition()) {
s = wals_.AddWals(e->GetWalAdditions());
} else if (e->IsWalDeletion()) {
s = wals_.DeleteWalsBefore(e->GetWalDeletion().GetLogNumber());
}
if (!s.ok()) {
break;
}
}
}
if (!io_s.ok()) {
if (io_status_.ok()) {
io_status_ = io_s;
}
} else if (!io_status_.ok()) {
io_status_ = io_s;
}
// Append the old manifest file to the obsolete_manifest_ list to be deleted
// by PurgeObsoleteFiles later.
if (s.ok() && new_descriptor_log) {
obsolete_manifests_.emplace_back(
DescriptorFileName("", manifest_file_number_));
}
// Install the new versions
if (s.ok()) {
if (first_writer.edit_list.front()->is_column_family_add_) {
assert(batch_edits.size() == 1);
assert(new_cf_options != nullptr);
CreateColumnFamily(*new_cf_options, first_writer.edit_list.front());
} else if (first_writer.edit_list.front()->is_column_family_drop_) {
assert(batch_edits.size() == 1);
first_writer.cfd->SetDropped();
first_writer.cfd->UnrefAndTryDelete();
} else {
// Each version in versions corresponds to a column family.
// For each column family, update its log number indicating that logs
// with number smaller than this should be ignored.
uint64_t last_min_log_number_to_keep = 0;
for (const auto& e : batch_edits) {
ColumnFamilyData* cfd = nullptr;
if (!e->IsColumnFamilyManipulation()) {
cfd = column_family_set_->GetColumnFamily(e->column_family_);
// e would not have been added to batch_edits if its corresponding
// column family is dropped.
assert(cfd);
}
if (cfd) {
if (e->has_log_number_ && e->log_number_ > cfd->GetLogNumber()) {
cfd->SetLogNumber(e->log_number_);
}
if (e->HasFullHistoryTsLow()) {
cfd->SetFullHistoryTsLow(e->GetFullHistoryTsLow());
}
}
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
if (e->has_min_log_number_to_keep_) {
last_min_log_number_to_keep =
std::max(last_min_log_number_to_keep, e->min_log_number_to_keep_);
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
}
2014-03-14 21:11:41 +01:00
}
if (last_min_log_number_to_keep != 0) {
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
// Should only be set in 2PC mode.
MarkMinLogNumberToKeep2PC(last_min_log_number_to_keep);
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
}
for (int i = 0; i < static_cast<int>(versions.size()); ++i) {
ColumnFamilyData* cfd = versions[i]->cfd_;
AppendVersion(cfd, versions[i]);
}
}
manifest_file_number_ = pending_manifest_file_number_;
manifest_file_size_ = new_manifest_file_size;
prev_log_number_ = first_writer.edit_list.front()->prev_log_number_;
} else {
std::string version_edits;
for (auto& e : batch_edits) {
version_edits += ("\n" + e->DebugString(true));
}
ROCKS_LOG_ERROR(db_options_->info_log,
"Error in committing version edit to MANIFEST: %s",
version_edits.c_str());
for (auto v : versions) {
delete v;
}
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
if (manifest_io_status.ok()) {
manifest_file_number_ = pending_manifest_file_number_;
manifest_file_size_ = new_manifest_file_size;
}
// If manifest append failed for whatever reason, the file could be
// corrupted. So we need to force the next version update to start a
// new manifest file.
descriptor_log_.reset();
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
2021-04-20 03:10:23 +02:00
// If manifest operations failed, then we know the CURRENT file still
// points to the original MANIFEST. Therefore, we can safely delete the
// new MANIFEST.
// If manifest operations succeeded, and we are here, then it is possible
// that renaming tmp file to CURRENT failed.
//
// On local POSIX-compliant FS, the CURRENT must point to the original
// MANIFEST. We can delete the new MANIFEST for simplicity, but we can also
// keep it. Future recovery will ignore this MANIFEST. It's also ok for the
// process not to crash and continue using the db. Any future LogAndApply()
// call will switch to a new MANIFEST and update CURRENT, still ignoring
// this one.
//
// On non-local FS, it is
// possible that the rename operation succeeded on the server (remote)
// side, but the client somehow returns a non-ok status to RocksDB. Note
// that this does not violate atomicity. Should we delete the new MANIFEST
// successfully, a subsequent recovery attempt will likely see the CURRENT
// pointing to the new MANIFEST, thus fail. We will not be able to open the
// DB again. Therefore, if manifest operations succeed, we should keep the
// the new MANIFEST. If the process proceeds, any future LogAndApply() call
// will switch to a new MANIFEST and update CURRENT. If user tries to
// re-open the DB,
// a) CURRENT points to the new MANIFEST, and the new MANIFEST is present.
// b) CURRENT points to the original MANIFEST, and the original MANIFEST
// also exists.
if (new_descriptor_log && !manifest_io_status.ok()) {
ROCKS_LOG_INFO(db_options_->info_log,
"Deleting manifest %" PRIu64 " current manifest %" PRIu64
"\n",
pending_manifest_file_number_, manifest_file_number_);
Status manifest_del_status = env_->DeleteFile(
DescriptorFileName(dbname_, pending_manifest_file_number_));
if (!manifest_del_status.ok()) {
ROCKS_LOG_WARN(db_options_->info_log,
"Failed to delete manifest %" PRIu64 ": %s",
pending_manifest_file_number_,
manifest_del_status.ToString().c_str());
}
}
}
pending_manifest_file_number_ = 0;
// wake up all the waiting writers
while (true) {
ManifestWriter* ready = manifest_writers_.front();
manifest_writers_.pop_front();
bool need_signal = true;
for (const auto& w : writers) {
if (&w == ready) {
need_signal = false;
break;
}
}
ready->status = s;
ready->done = true;
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
if (ready->manifest_write_callback) {
(ready->manifest_write_callback)(s);
}
if (need_signal) {
ready->cv.Signal();
}
if (ready == last_writer) {
break;
}
}
if (!manifest_writers_.empty()) {
manifest_writers_.front()->cv.Signal();
}
return s;
}
void VersionSet::WakeUpWaitingManifestWriters() {
// wake up all the waiting writers
// Notify new head of manifest write queue.
if (!manifest_writers_.empty()) {
manifest_writers_.front()->cv.Signal();
}
}
// 'datas' is grammatically incorrect. We still use this notation to indicate
// that this variable represents a collection of column_family_data.
Status VersionSet::LogAndApply(
const autovector<ColumnFamilyData*>& column_family_datas,
const autovector<const MutableCFOptions*>& mutable_cf_options_list,
const autovector<autovector<VersionEdit*>>& edit_lists,
InstrumentedMutex* mu, FSDirectory* db_directory, bool new_descriptor_log,
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
const ColumnFamilyOptions* new_cf_options,
const std::vector<std::function<void(const Status&)>>& manifest_wcbs) {
mu->AssertHeld();
int num_edits = 0;
for (const auto& elist : edit_lists) {
num_edits += static_cast<int>(elist.size());
}
if (num_edits == 0) {
return Status::OK();
} else if (num_edits > 1) {
#ifndef NDEBUG
for (const auto& edit_list : edit_lists) {
for (const auto& edit : edit_list) {
assert(!edit->IsColumnFamilyManipulation());
}
}
#endif /* ! NDEBUG */
}
int num_cfds = static_cast<int>(column_family_datas.size());
if (num_cfds == 1 && column_family_datas[0] == nullptr) {
assert(edit_lists.size() == 1 && edit_lists[0].size() == 1);
assert(edit_lists[0][0]->is_column_family_add_);
assert(new_cf_options != nullptr);
}
std::deque<ManifestWriter> writers;
if (num_cfds > 0) {
assert(static_cast<size_t>(num_cfds) == mutable_cf_options_list.size());
assert(static_cast<size_t>(num_cfds) == edit_lists.size());
}
for (int i = 0; i < num_cfds; ++i) {
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
const auto wcb =
manifest_wcbs.empty() ? [](const Status&) {} : manifest_wcbs[i];
writers.emplace_back(mu, column_family_datas[i],
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
*mutable_cf_options_list[i], edit_lists[i], wcb);
manifest_writers_.push_back(&writers[i]);
}
assert(!writers.empty());
ManifestWriter& first_writer = writers.front();
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
TEST_SYNC_POINT_CALLBACK("VersionSet::LogAndApply:BeforeWriterWaiting",
nullptr);
while (!first_writer.done && &first_writer != manifest_writers_.front()) {
first_writer.cv.Wait();
}
if (first_writer.done) {
// All non-CF-manipulation operations can be grouped together and committed
// to MANIFEST. They should all have finished. The status code is stored in
// the first manifest writer.
#ifndef NDEBUG
for (const auto& writer : writers) {
assert(writer.done);
}
Perform post-flush updates of memtable list in a callback (#6069) Summary: Currently, the following interleaving of events can lead to SuperVersion containing both immutable memtables as well as the resulting L0. This can cause Get to return incorrect result if there are merge operands. This may also affect other operations such as single deletes. ``` time main_thr bg_flush_thr bg_compact_thr compact_thr set_opts_thr 0 | WriteManifest:0 1 | issue compact 2 | wait 3 | Merge(counter) 4 | issue flush 5 | wait 6 | WriteManifest:1 7 | wake up 8 | write manifest 9 | wake up 10 | Get(counter) 11 | remove imm V ``` The reason behind is that: one bg flush thread's installing new `Version` can be batched and performed by another thread that is the "leader" MANIFEST writer. This bg thread removes the memtables from current super version only after `LogAndApply` returns. After the leader MANIFEST writer signals (releasing mutex) this bg flush thread, it is possible that another thread sees this cf with both memtables (whose data have been flushed to the newest L0) and the L0 before this bg flush thread removes the memtables. To address this issue, each bg flush thread can pass a callback function to `LogAndApply`. The callback is responsible for removing the memtables. Therefore, the leader MANIFEST writer can call this callback and remove the memtables before releasing the mutex. Test plan (devserver) ``` $make merge_test $./merge_test --gtest_filter=MergeTest.MergeWithCompactionAndFlush $make check ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6069 Reviewed By: cheng-chang Differential Revision: D18790894 Pulled By: riversand963 fbshipit-source-id: e41bd600c0448b4f4b2deb3f7677f95e3076b4ed
2020-10-27 02:20:43 +01:00
TEST_SYNC_POINT_CALLBACK("VersionSet::LogAndApply:WakeUpAndDone", mu);
#endif /* !NDEBUG */
return first_writer.status;
}
int num_undropped_cfds = 0;
for (auto cfd : column_family_datas) {
// if cfd == nullptr, it is a column family add.
if (cfd == nullptr || !cfd->IsDropped()) {
++num_undropped_cfds;
}
}
if (0 == num_undropped_cfds) {
for (int i = 0; i != num_cfds; ++i) {
manifest_writers_.pop_front();
}
// Notify new head of manifest write queue.
if (!manifest_writers_.empty()) {
manifest_writers_.front()->cv.Signal();
}
return Status::ColumnFamilyDropped();
}
return ProcessManifestWrites(writers, mu, db_directory, new_descriptor_log,
new_cf_options);
}
2014-03-13 02:09:03 +01:00
void VersionSet::LogAndApplyCFHelper(VersionEdit* edit) {
assert(edit->IsColumnFamilyManipulation());
edit->SetNextFile(next_file_number_.load());
// The log might have data that is not visible to memtbale and hence have not
// updated the last_sequence_ yet. It is also possible that the log has is
// expecting some new data that is not written yet. Since LastSequence is an
// upper bound on the sequence, it is ok to record
// last_allocated_sequence_ as the last sequence.
edit->SetLastSequence(db_options_->two_write_queues ? last_allocated_sequence_
: last_sequence_);
2014-03-13 02:09:03 +01:00
if (edit->is_column_family_drop_) {
// if we drop column family, we have to make sure to save max column family,
// so that we don't reuse existing ID
edit->SetMaxColumnFamily(column_family_set_->GetMaxColumnFamily());
}
}
Status VersionSet::LogAndApplyHelper(ColumnFamilyData* cfd,
VersionBuilder* builder, VersionEdit* edit,
InstrumentedMutex* mu) {
#ifdef NDEBUG
(void)cfd;
#endif
mu->AssertHeld();
2014-03-13 02:09:03 +01:00
assert(!edit->IsColumnFamilyManipulation());
if (edit->has_log_number_) {
assert(edit->log_number_ >= cfd->GetLogNumber());
assert(edit->log_number_ < next_file_number_.load());
}
2014-03-13 02:09:03 +01:00
if (!edit->has_prev_log_number_) {
edit->SetPrevLogNumber(prev_log_number_);
}
edit->SetNextFile(next_file_number_.load());
// The log might have data that is not visible to memtbale and hence have not
// updated the last_sequence_ yet. It is also possible that the log has is
// expecting some new data that is not written yet. Since LastSequence is an
// upper bound on the sequence, it is ok to record
// last_allocated_sequence_ as the last sequence.
edit->SetLastSequence(db_options_->two_write_queues ? last_allocated_sequence_
: last_sequence_);
2014-03-13 02:09:03 +01:00
// The builder can be nullptr only if edit is WAL manipulation,
// because WAL edits do not need to be applied to versions,
// we return Status::OK() in this case.
assert(builder || edit->IsWalManipulation());
return builder ? builder->Apply(edit) : Status::OK();
}
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
Status VersionSet::GetCurrentManifestPath(const std::string& dbname,
FileSystem* fs,
std::string* manifest_path,
uint64_t* manifest_file_number) {
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
assert(fs != nullptr);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
assert(manifest_path != nullptr);
assert(manifest_file_number != nullptr);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
std::string fname;
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
Status s = ReadFileToString(fs, CurrentFileName(dbname), &fname);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
if (!s.ok()) {
return s;
}
if (fname.empty() || fname.back() != '\n') {
return Status::Corruption("CURRENT file does not end with newline");
}
// remove the trailing '\n'
fname.resize(fname.size() - 1);
FileType type;
bool parse_ok = ParseFileName(fname, manifest_file_number, &type);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
if (!parse_ok || type != kDescriptorFile) {
return Status::Corruption("CURRENT file corrupted");
}
*manifest_path = dbname;
if (dbname.back() != '/') {
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
manifest_path->push_back('/');
}
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
manifest_path->append(fname);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
return Status::OK();
}
Status VersionSet::Recover(
const std::vector<ColumnFamilyDescriptor>& column_families, bool read_only,
std::string* db_id) {
// Read "CURRENT" file, which contains a pointer to the current manifest file
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
std::string manifest_path;
Status s = GetCurrentManifestPath(dbname_, fs_.get(), &manifest_path,
&manifest_file_number_);
if (!s.ok()) {
return s;
}
ROCKS_LOG_INFO(db_options_->info_log, "Recovering from manifest file: %s\n",
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
manifest_path.c_str());
std::unique_ptr<SequentialFileReader> manifest_file_reader;
{
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
std::unique_ptr<FSSequentialFile> manifest_file;
s = fs_->NewSequentialFile(manifest_path,
fs_->OptimizeForManifestRead(file_options_),
&manifest_file, nullptr);
if (!s.ok()) {
return s;
}
manifest_file_reader.reset(
new SequentialFileReader(std::move(manifest_file), manifest_path,
db_options_->log_readahead_size, io_tracer_));
}
uint64_t current_manifest_file_size = 0;
uint64_t log_number = 0;
{
VersionSet::LogReporter reporter;
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
Status log_read_status;
reporter.status = &log_read_status;
log::Reader reader(nullptr, std::move(manifest_file_reader), &reporter,
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
true /* checksum */, 0 /* log_number */);
VersionEditHandler handler(read_only, column_families,
const_cast<VersionSet*>(this),
/*track_missing_files=*/false,
/*no_error_if_files_missing=*/false, io_tracer_);
handler.Iterate(reader, &log_read_status);
s = handler.status();
if (s.ok()) {
log_number = handler.GetVersionEditParams().log_number_;
current_manifest_file_size = reader.GetReadOffset();
assert(current_manifest_file_size != 0);
handler.GetDbId(db_id);
}
}
if (s.ok()) {
manifest_file_size_ = current_manifest_file_size;
ROCKS_LOG_INFO(
db_options_->info_log,
"Recovered from manifest file:%s succeeded,"
"manifest_file_number is %" PRIu64 ", next_file_number is %" PRIu64
", last_sequence is %" PRIu64 ", log_number is %" PRIu64
",prev_log_number is %" PRIu64 ",max_column_family is %" PRIu32
",min_log_number_to_keep is %" PRIu64 "\n",
manifest_path.c_str(), manifest_file_number_, next_file_number_.load(),
last_sequence_.load(), log_number, prev_log_number_,
column_family_set_->GetMaxColumnFamily(), min_log_number_to_keep_2pc());
for (auto cfd : *column_family_set_) {
if (cfd->IsDropped()) {
continue;
}
ROCKS_LOG_INFO(db_options_->info_log,
"Column family [%s] (ID %" PRIu32
"), log number is %" PRIu64 "\n",
cfd->GetName().c_str(), cfd->GetID(), cfd->GetLogNumber());
}
}
return s;
}
namespace {
class ManifestPicker {
public:
explicit ManifestPicker(const std::string& dbname,
const std::vector<std::string>& files_in_dbname);
// REQUIRES Valid() == true
std::string GetNextManifest(uint64_t* file_number, std::string* file_name);
bool Valid() const { return manifest_file_iter_ != manifest_files_.end(); }
private:
const std::string& dbname_;
// MANIFEST file names(s)
std::vector<std::string> manifest_files_;
std::vector<std::string>::const_iterator manifest_file_iter_;
};
ManifestPicker::ManifestPicker(const std::string& dbname,
const std::vector<std::string>& files_in_dbname)
: dbname_(dbname) {
// populate manifest files
assert(!files_in_dbname.empty());
for (const auto& fname : files_in_dbname) {
uint64_t file_num = 0;
FileType file_type;
bool parse_ok = ParseFileName(fname, &file_num, &file_type);
if (parse_ok && file_type == kDescriptorFile) {
manifest_files_.push_back(fname);
}
}
// seek to first manifest
std::sort(manifest_files_.begin(), manifest_files_.end(),
[](const std::string& lhs, const std::string& rhs) {
uint64_t num1 = 0;
uint64_t num2 = 0;
FileType type1;
FileType type2;
bool parse_ok1 = ParseFileName(lhs, &num1, &type1);
bool parse_ok2 = ParseFileName(rhs, &num2, &type2);
#ifndef NDEBUG
assert(parse_ok1);
assert(parse_ok2);
#else
(void)parse_ok1;
(void)parse_ok2;
#endif
return num1 > num2;
});
manifest_file_iter_ = manifest_files_.begin();
}
std::string ManifestPicker::GetNextManifest(uint64_t* number,
std::string* file_name) {
assert(Valid());
std::string ret;
if (manifest_file_iter_ != manifest_files_.end()) {
ret.assign(dbname_);
if (ret.back() != kFilePathSeparator) {
ret.push_back(kFilePathSeparator);
}
ret.append(*manifest_file_iter_);
if (number) {
FileType type;
bool parse = ParseFileName(*manifest_file_iter_, number, &type);
assert(type == kDescriptorFile);
#ifndef NDEBUG
assert(parse);
#else
(void)parse;
#endif
}
if (file_name) {
*file_name = *manifest_file_iter_;
}
++manifest_file_iter_;
}
return ret;
}
} // namespace
Status VersionSet::TryRecover(
const std::vector<ColumnFamilyDescriptor>& column_families, bool read_only,
const std::vector<std::string>& files_in_dbname, std::string* db_id,
bool* has_missing_table_file) {
ManifestPicker manifest_picker(dbname_, files_in_dbname);
if (!manifest_picker.Valid()) {
return Status::Corruption("Cannot locate MANIFEST file in " + dbname_);
}
Status s;
std::string manifest_path =
manifest_picker.GetNextManifest(&manifest_file_number_, nullptr);
while (!manifest_path.empty()) {
s = TryRecoverFromOneManifest(manifest_path, column_families, read_only,
db_id, has_missing_table_file);
if (s.ok() || !manifest_picker.Valid()) {
break;
}
Reset();
manifest_path =
manifest_picker.GetNextManifest(&manifest_file_number_, nullptr);
}
return s;
}
Status VersionSet::TryRecoverFromOneManifest(
const std::string& manifest_path,
const std::vector<ColumnFamilyDescriptor>& column_families, bool read_only,
std::string* db_id, bool* has_missing_table_file) {
ROCKS_LOG_INFO(db_options_->info_log, "Trying to recover from manifest: %s\n",
manifest_path.c_str());
std::unique_ptr<SequentialFileReader> manifest_file_reader;
Status s;
{
std::unique_ptr<FSSequentialFile> manifest_file;
s = fs_->NewSequentialFile(manifest_path,
fs_->OptimizeForManifestRead(file_options_),
&manifest_file, nullptr);
if (!s.ok()) {
return s;
}
manifest_file_reader.reset(
new SequentialFileReader(std::move(manifest_file), manifest_path,
db_options_->log_readahead_size, io_tracer_));
}
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
assert(s.ok());
VersionSet::LogReporter reporter;
reporter.status = &s;
log::Reader reader(nullptr, std::move(manifest_file_reader), &reporter,
/*checksum=*/true, /*log_num=*/0);
VersionEditHandlerPointInTime handler_pit(
read_only, column_families, const_cast<VersionSet*>(this), io_tracer_);
handler_pit.Iterate(reader, &s);
handler_pit.GetDbId(db_id);
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
assert(nullptr != has_missing_table_file);
*has_missing_table_file = handler_pit.HasMissingFiles();
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
return handler_pit.status();
}
Status VersionSet::ListColumnFamilies(std::vector<std::string>* column_families,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const std::string& dbname,
FileSystem* fs) {
// these are just for performance reasons, not correctness,
// so we're fine using the defaults
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
FileOptions soptions;
// Read "CURRENT" file, which contains a pointer to the current manifest file
std::string manifest_path;
uint64_t manifest_file_number;
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
Status s =
GetCurrentManifestPath(dbname, fs, &manifest_path, &manifest_file_number);
if (!s.ok()) {
return s;
}
std::unique_ptr<SequentialFileReader> file_reader;
{
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
std::unique_ptr<FSSequentialFile> file;
s = fs->NewSequentialFile(manifest_path, soptions, &file, nullptr);
if (!s.ok()) {
return s;
}
file_reader.reset(new SequentialFileReader(std::move(file), manifest_path,
nullptr /*IOTracer*/));
}
VersionSet::LogReporter reporter;
reporter.status = &s;
log::Reader reader(nullptr, std::move(file_reader), &reporter,
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
true /* checksum */, 0 /* log_number */);
ListColumnFamiliesHandler handler;
handler.Iterate(reader, &s);
assert(column_families);
column_families->clear();
if (handler.status().ok()) {
for (const auto& iter : handler.GetColumnFamilyNames()) {
column_families->push_back(iter.second);
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
}
}
return handler.status();
}
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
#ifndef ROCKSDB_LITE
Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
const Options* options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options,
int new_levels) {
if (new_levels <= 1) {
return Status::InvalidArgument(
"Number of levels needs to be bigger than 1");
}
ImmutableDBOptions db_options(*options);
ColumnFamilyOptions cf_options(*options);
std::shared_ptr<Cache> tc(NewLRUCache(options->max_open_files - 10,
options->table_cache_numshardbits));
WriteController wc(options->delayed_write_rate);
WriteBufferManager wb(options->db_write_buffer_size);
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
VersionSet versions(dbname, &db_options, file_options, tc.get(), &wb, &wc,
nullptr /*BlockCacheTracer*/, nullptr /*IOTracer*/,
/*db_session_id*/ "");
Status status;
std::vector<ColumnFamilyDescriptor> dummy;
ColumnFamilyDescriptor dummy_descriptor(kDefaultColumnFamilyName,
ColumnFamilyOptions(*options));
2014-02-01 04:44:48 +01:00
dummy.push_back(dummy_descriptor);
status = versions.Recover(dummy);
if (!status.ok()) {
return status;
}
Version* current_version =
versions.GetColumnFamilySet()->GetDefault()->current();
auto* vstorage = current_version->storage_info();
int current_levels = vstorage->num_levels();
if (current_levels <= new_levels) {
return Status::OK();
}
// Make sure there are file only on one level from
// (new_levels-1) to (current_levels-1)
int first_nonempty_level = -1;
int first_nonempty_level_filenum = 0;
for (int i = new_levels - 1; i < current_levels; i++) {
int file_num = vstorage->NumLevelFiles(i);
if (file_num != 0) {
if (first_nonempty_level < 0) {
first_nonempty_level = i;
first_nonempty_level_filenum = file_num;
} else {
char msg[255];
snprintf(msg, sizeof(msg),
"Found at least two levels containing files: "
"[%d:%d],[%d:%d].\n",
first_nonempty_level, first_nonempty_level_filenum, i,
file_num);
return Status::InvalidArgument(msg);
}
}
}
// we need to allocate an array with the old number of levels size to
// avoid SIGSEGV in WriteCurrentStatetoManifest()
// however, all levels bigger or equal to new_levels will be empty
std::vector<FileMetaData*>* new_files_list =
new std::vector<FileMetaData*>[current_levels];
for (int i = 0; i < new_levels - 1; i++) {
new_files_list[i] = vstorage->LevelFiles(i);
}
if (first_nonempty_level > 0) {
auto& new_last_level = new_files_list[new_levels - 1];
new_last_level = vstorage->LevelFiles(first_nonempty_level);
for (size_t i = 0; i < new_last_level.size(); ++i) {
const FileMetaData* const meta = new_last_level[i];
assert(meta);
const uint64_t file_number = meta->fd.GetNumber();
vstorage->file_locations_[file_number] =
VersionStorageInfo::FileLocation(new_levels - 1, i);
}
}
delete[] vstorage -> files_;
vstorage->files_ = new_files_list;
vstorage->num_levels_ = new_levels;
MutableCFOptions mutable_cf_options(*options);
VersionEdit ve;
InstrumentedMutex dummy_mutex;
InstrumentedMutexLock l(&dummy_mutex);
return versions.LogAndApply(
versions.GetColumnFamilySet()->GetDefault(),
mutable_cf_options, &ve, &dummy_mutex, nullptr, true);
}
// Get the checksum information including the checksum and checksum function
// name of all SST and blob files in VersionSet. Store the information in
// FileChecksumList which contains a map from file number to its checksum info.
// If DB is not running, make sure call VersionSet::Recover() to load the file
// metadata from Manifest to VersionSet before calling this function.
Status VersionSet::GetLiveFilesChecksumInfo(FileChecksumList* checksum_list) {
// Clean the previously stored checksum information if any.
Status s;
if (checksum_list == nullptr) {
s = Status::InvalidArgument("checksum_list is nullptr");
return s;
}
checksum_list->reset();
for (auto cfd : *column_family_set_) {
if (cfd->IsDropped() || !cfd->initialized()) {
continue;
}
/* SST files */
for (int level = 0; level < cfd->NumberLevels(); level++) {
for (const auto& file :
cfd->current()->storage_info()->LevelFiles(level)) {
s = checksum_list->InsertOneFileChecksum(file->fd.GetNumber(),
file->file_checksum,
file->file_checksum_func_name);
if (!s.ok()) {
return s;
}
}
}
/* Blob files */
const auto& blob_files = cfd->current()->storage_info()->GetBlobFiles();
for (const auto& pair : blob_files) {
const uint64_t blob_file_number = pair.first;
const auto& meta = pair.second;
assert(meta);
assert(blob_file_number == meta->GetBlobFileNumber());
std::string checksum_value = meta->GetChecksumValue();
std::string checksum_method = meta->GetChecksumMethod();
assert(checksum_value.empty() == checksum_method.empty());
if (meta->GetChecksumMethod().empty()) {
checksum_value = kUnknownFileChecksum;
checksum_method = kUnknownFileChecksumFuncName;
}
s = checksum_list->InsertOneFileChecksum(blob_file_number, checksum_value,
checksum_method);
if (!s.ok()) {
return s;
}
}
}
return s;
}
Status VersionSet::DumpManifest(Options& options, std::string& dscname,
Added JSON manifest dump option to ldb command Summary: Added a new flag --json to the ldb manifest_dump command that prints out the version edits as JSON objects for easier reading and parsing of information. Test Plan: **Sample usage: ** ``` ./ldb manifest_dump --json --path=path/to/manifest/file ``` **Sample output:** ``` {"EditNumber": 0, "Comparator": "leveldb.BytewiseComparator", "ColumnFamily": 0} {"EditNumber": 1, "LogNumber": 0, "ColumnFamily": 0} {"EditNumber": 2, "LogNumber": 4, "PrevLogNumber": 0, "NextFileNumber": 7, "LastSeq": 35356, "AddedFiles": [{"Level": 0, "FileNumber": 5, "FileSize": 1949284, "SmallestIKey": "'", "LargestIKey": "'"}], "ColumnFamily": 0} ... {"EditNumber": 13, "PrevLogNumber": 0, "NextFileNumber": 36, "LastSeq": 290994, "DeletedFiles": [{"Level": 0, "FileNumber": 17}, {"Level": 0, "FileNumber": 20}, {"Level": 0, "FileNumber": 22}, {"Level": 0, "FileNumber": 24}, {"Level": 1, "FileNumber": 13}, {"Level": 1, "FileNumber": 14}, {"Level": 1, "FileNumber": 15}, {"Level": 1, "FileNumber": 18}], "AddedFiles": [{"Level": 1, "FileNumber": 25, "FileSize": 2114340, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 26, "FileSize": 2115213, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 27, "FileSize": 2114807, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 30, "FileSize": 2115271, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 31, "FileSize": 2115165, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 32, "FileSize": 2114683, "SmallestIKey": "'", "LargestIKey": "'"}, {"Level": 1, "FileNumber": 35, "FileSize": 1757512, "SmallestIKey": "'", "LargestIKey": "'"}], "ColumnFamily": 0} ... ``` Reviewers: sdong, anthony, yhchiang, igor Reviewed By: igor Subscribers: dhruba Differential Revision: https://reviews.facebook.net/D41727
2015-07-17 19:07:40 +02:00
bool verbose, bool hex, bool json) {
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
// Open the specified manifest file.
std::unique_ptr<SequentialFileReader> file_reader;
Status s;
{
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
std::unique_ptr<FSSequentialFile> file;
Simplify migration to FileSystem API (#6552) Summary: The current Env/FileSystem API separation has a couple of issues - 1. It requires the user to specify 2 options - ```Options::env``` and ```Options::file_system``` - which means they have to make code changes to benefit from the new APIs. Furthermore, there is a risk of accessing the same APIs in two different ways, through Env in the old way and through FileSystem in the new way. The two may not always match, for example, if env is ```PosixEnv``` and FileSystem is a custom implementation. Any stray RocksDB calls to env will use the ```PosixEnv``` implementation rather than the file_system implementation. 2. There needs to be a simple way for the FileSystem developer to instantiate an Env for backward compatibility purposes. This PR solves the above issues and simplifies the migration in the following ways - 1. Embed a shared_ptr to the ```FileSystem``` in the ```Env```, and remove ```Options::file_system``` as a configurable option. This way, no code changes will be required in application code to benefit from the new API. The default Env constructor uses a ```LegacyFileSystemWrapper``` as the embedded ```FileSystem```. 1a. - This also makes it more robust by ensuring that even if RocksDB has some stray calls to Env APIs rather than FileSystem, they will go through the same object and thus there is no risk of getting out of sync. 2. Provide a ```NewCompositeEnv()``` API that can be used to construct a PosixEnv with a custom FileSystem implementation. This eliminates an indirection to call Env APIs, and relieves the FileSystem developer of the burden of having to implement wrappers for the Env APIs. 3. Add a couple of missing FileSystem APIs - ```SanitizeEnvOptions()``` and ```NewLogger()``` Tests: 1. New unit tests 2. make check and make asan_check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6552 Reviewed By: riversand963 Differential Revision: D20592038 Pulled By: anand1976 fbshipit-source-id: c3801ad4153f96d21d5a3ae26c92ba454d1bf1f7
2020-03-24 05:50:42 +01:00
const std::shared_ptr<FileSystem>& fs = options.env->GetFileSystem();
s = fs->NewSequentialFile(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
dscname,
Simplify migration to FileSystem API (#6552) Summary: The current Env/FileSystem API separation has a couple of issues - 1. It requires the user to specify 2 options - ```Options::env``` and ```Options::file_system``` - which means they have to make code changes to benefit from the new APIs. Furthermore, there is a risk of accessing the same APIs in two different ways, through Env in the old way and through FileSystem in the new way. The two may not always match, for example, if env is ```PosixEnv``` and FileSystem is a custom implementation. Any stray RocksDB calls to env will use the ```PosixEnv``` implementation rather than the file_system implementation. 2. There needs to be a simple way for the FileSystem developer to instantiate an Env for backward compatibility purposes. This PR solves the above issues and simplifies the migration in the following ways - 1. Embed a shared_ptr to the ```FileSystem``` in the ```Env```, and remove ```Options::file_system``` as a configurable option. This way, no code changes will be required in application code to benefit from the new API. The default Env constructor uses a ```LegacyFileSystemWrapper``` as the embedded ```FileSystem```. 1a. - This also makes it more robust by ensuring that even if RocksDB has some stray calls to Env APIs rather than FileSystem, they will go through the same object and thus there is no risk of getting out of sync. 2. Provide a ```NewCompositeEnv()``` API that can be used to construct a PosixEnv with a custom FileSystem implementation. This eliminates an indirection to call Env APIs, and relieves the FileSystem developer of the burden of having to implement wrappers for the Env APIs. 3. Add a couple of missing FileSystem APIs - ```SanitizeEnvOptions()``` and ```NewLogger()``` Tests: 1. New unit tests 2. make check and make asan_check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6552 Reviewed By: riversand963 Differential Revision: D20592038 Pulled By: anand1976 fbshipit-source-id: c3801ad4153f96d21d5a3ae26c92ba454d1bf1f7
2020-03-24 05:50:42 +01:00
fs->OptimizeForManifestRead(file_options_), &file,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
nullptr);
if (!s.ok()) {
return s;
}
file_reader.reset(new SequentialFileReader(
std::move(file), dscname, db_options_->log_readahead_size, io_tracer_));
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
}
std::vector<ColumnFamilyDescriptor> column_families(
1, ColumnFamilyDescriptor(kDefaultColumnFamilyName, options));
DumpManifestHandler handler(column_families, this, io_tracer_, verbose, hex,
json);
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
{
VersionSet::LogReporter reporter;
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
reporter.status = &s;
log::Reader reader(nullptr, std::move(file_reader), &reporter,
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
true /* checksum */, 0 /* log_number */);
handler.Iterate(reader, &s);
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
}
return handler.status();
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
}
#endif // ROCKSDB_LITE
Utility to dump manifest contents. Summary: ./manifest_dump --file=/tmp/dbbench/MANIFEST-000002 Output looks like manifest_file_number 30 next_file_number 31 last_sequence 388082 log_number 28 prev_log_number 0 --- level 0 --- --- level 1 --- --- level 2 --- 5:3244155['0000000000000000' @ 1 : 1 .. '0000000000028220' @ 28221 : 1] 7:3244177['0000000000028221' @ 28222 : 1 .. '0000000000056441' @ 56442 : 1] 9:3244156['0000000000056442' @ 56443 : 1 .. '0000000000084662' @ 84663 : 1] 11:3244178['0000000000084663' @ 84664 : 1 .. '0000000000112883' @ 112884 : 1] 13:3244158['0000000000112884' @ 112885 : 1 .. '0000000000141104' @ 141105 : 1] 15:3244176['0000000000141105' @ 141106 : 1 .. '0000000000169325' @ 169326 : 1] 17:3244156['0000000000169326' @ 169327 : 1 .. '0000000000197546' @ 197547 : 1] 19:3244178['0000000000197547' @ 197548 : 1 .. '0000000000225767' @ 225768 : 1] 21:3244155['0000000000225768' @ 225769 : 1 .. '0000000000253988' @ 253989 : 1] 23:3244179['0000000000253989' @ 253990 : 1 .. '0000000000282209' @ 282210 : 1] 25:3244157['0000000000282210' @ 282211 : 1 .. '0000000000310430' @ 310431 : 1] 27:3244176['0000000000310431' @ 310432 : 1 .. '0000000000338651' @ 338652 : 1] 29:3244156['0000000000338652' @ 338653 : 1 .. '0000000000366872' @ 366873 : 1] --- level 3 --- --- level 4 --- --- level 5 --- --- level 6 --- Test Plan: run on test directory created by dbbench Reviewers: heyongqiang Reviewed By: heyongqiang CC: hustliubo Differential Revision: https://reviews.facebook.net/D4743
2012-08-17 19:48:40 +02:00
void VersionSet::MarkFileNumberUsed(uint64_t number) {
// only called during recovery and repair which are single threaded, so this
// works because there can't be concurrent calls
if (next_file_number_.load(std::memory_order_relaxed) <= number) {
next_file_number_.store(number + 1, std::memory_order_relaxed);
}
}
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
// Called only either from ::LogAndApply which is protected by mutex or during
// recovery which is single-threaded.
void VersionSet::MarkMinLogNumberToKeep2PC(uint64_t number) {
if (min_log_number_to_keep_2pc_.load(std::memory_order_relaxed) < number) {
min_log_number_to_keep_2pc_.store(number, std::memory_order_relaxed);
}
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
2020-01-07 05:08:24 +01:00
Status VersionSet::WriteCurrentStateToManifest(
const std::unordered_map<uint32_t, MutableCFState>& curr_state,
const VersionEdit& wal_additions, log::Writer* log, IOStatus& io_s) {
// TODO: Break up into multiple records to reduce memory usage on recovery?
2014-03-13 02:09:03 +01:00
// WARNING: This method doesn't hold a mutex!!
// This is done without DB mutex lock held, but only within single-threaded
// LogAndApply. Column family manipulations can only happen within LogAndApply
2014-03-13 02:09:03 +01:00
// (the same single thread), so we're safe to iterate.
assert(io_s.ok());
if (db_options_->write_dbid_to_manifest) {
VersionEdit edit_for_db_id;
assert(!db_id_.empty());
edit_for_db_id.SetDBId(db_id_);
std::string db_id_record;
if (!edit_for_db_id.EncodeTo(&db_id_record)) {
return Status::Corruption("Unable to Encode VersionEdit:" +
edit_for_db_id.DebugString(true));
}
io_s = log->AddRecord(db_id_record);
if (!io_s.ok()) {
return io_s;
}
}
// Save WALs.
if (!wal_additions.GetWalAdditions().empty()) {
TEST_SYNC_POINT_CALLBACK("VersionSet::WriteCurrentStateToManifest:SaveWal",
const_cast<VersionEdit*>(&wal_additions));
std::string record;
if (!wal_additions.EncodeTo(&record)) {
return Status::Corruption("Unable to Encode VersionEdit: " +
wal_additions.DebugString(true));
}
io_s = log->AddRecord(record);
if (!io_s.ok()) {
return io_s;
}
}
for (auto cfd : *column_family_set_) {
assert(cfd);
if (cfd->IsDropped()) {
continue;
}
assert(cfd->initialized());
{
// Store column family info
VersionEdit edit;
if (cfd->GetID() != 0) {
// default column family is always there,
// no need to explicitly write it
edit.AddColumnFamily(cfd->GetName());
edit.SetColumnFamily(cfd->GetID());
}
edit.SetComparatorName(
cfd->internal_comparator().user_comparator()->Name());
std::string record;
if (!edit.EncodeTo(&record)) {
return Status::Corruption(
"Unable to Encode VersionEdit:" + edit.DebugString(true));
}
io_s = log->AddRecord(record);
if (!io_s.ok()) {
return io_s;
}
}
{
// Save files
VersionEdit edit;
edit.SetColumnFamily(cfd->GetID());
assert(cfd->current());
assert(cfd->current()->storage_info());
for (int level = 0; level < cfd->NumberLevels(); level++) {
for (const auto& f :
cfd->current()->storage_info()->LevelFiles(level)) {
edit.AddFile(level, f->fd.GetNumber(), f->fd.GetPathId(),
f->fd.GetFileSize(), f->smallest, f->largest,
f->fd.smallest_seqno, f->fd.largest_seqno,
f->marked_for_compaction, f->oldest_blob_file_number,
f->oldest_ancester_time, f->file_creation_time,
f->file_checksum, f->file_checksum_func_name);
}
}
const auto& blob_files = cfd->current()->storage_info()->GetBlobFiles();
for (const auto& pair : blob_files) {
const uint64_t blob_file_number = pair.first;
const auto& meta = pair.second;
assert(meta);
assert(blob_file_number == meta->GetBlobFileNumber());
edit.AddBlobFile(blob_file_number, meta->GetTotalBlobCount(),
meta->GetTotalBlobBytes(), meta->GetChecksumMethod(),
meta->GetChecksumValue());
if (meta->GetGarbageBlobCount() > 0) {
edit.AddBlobFileGarbage(blob_file_number, meta->GetGarbageBlobCount(),
meta->GetGarbageBlobBytes());
}
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
2020-01-07 05:08:24 +01:00
const auto iter = curr_state.find(cfd->GetID());
assert(iter != curr_state.end());
uint64_t log_number = iter->second.log_number;
edit.SetLogNumber(log_number);
if (cfd->GetID() == 0) {
// min_log_number_to_keep is for the whole db, not for specific column family.
// So it does not need to be set for every column family, just need to be set once.
// Since default CF can never be dropped, we set the min_log to the default CF here.
uint64_t min_log = min_log_number_to_keep_2pc();
if (min_log != 0) {
edit.SetMinLogNumberToKeep(min_log);
}
}
const std::string& full_history_ts_low = iter->second.full_history_ts_low;
if (!full_history_ts_low.empty()) {
edit.SetFullHistoryTsLow(full_history_ts_low);
}
std::string record;
if (!edit.EncodeTo(&record)) {
return Status::Corruption(
"Unable to Encode VersionEdit:" + edit.DebugString(true));
}
io_s = log->AddRecord(record);
if (!io_s.ok()) {
return io_s;
}
}
}
return Status::OK();
}
// TODO(aekmekji): in CompactionJob::GenSubcompactionBoundaries(), this
// function is called repeatedly with consecutive pairs of slices. For example
// if the slice list is [a, b, c, d] this function is called with arguments
// (a,b) then (b,c) then (c,d). Knowing this, an optimization is possible where
// we avoid doing binary search for the keys b and c twice and instead somehow
// maintain state of where they first appear in the files.
uint64_t VersionSet::ApproximateSize(const SizeApproximationOptions& options,
Version* v, const Slice& start,
const Slice& end, int start_level,
int end_level, TableReaderCaller caller) {
const auto& icmp = v->cfd_->internal_comparator();
// pre-condition
assert(icmp.Compare(start, end) <= 0);
uint64_t total_full_size = 0;
const auto* vstorage = v->storage_info();
const int num_non_empty_levels = vstorage->num_non_empty_levels();
end_level = (end_level == -1) ? num_non_empty_levels
: std::min(end_level, num_non_empty_levels);
assert(start_level <= end_level);
// Outline of the optimization that uses options.files_size_error_margin.
// When approximating the files total size that is used to store a keys range,
// we first sum up the sizes of the files that fully fall into the range.
// Then we sum up the sizes of all the files that may intersect with the range
// (this includes all files in L0 as well). Then, if total_intersecting_size
// is smaller than total_full_size * options.files_size_error_margin - we can
// infer that the intersecting files have a sufficiently negligible
// contribution to the total size, and we can approximate the storage required
// for the keys in range as just half of the intersecting_files_size.
// E.g., if the value of files_size_error_margin is 0.1, then the error of the
// approximation is limited to only ~10% of the total size of files that fully
// fall into the keys range. In such case, this helps to avoid a costly
// process of binary searching the intersecting files that is required only
// for a more precise calculation of the total size.
autovector<FdWithKeyRange*, 32> first_files;
autovector<FdWithKeyRange*, 16> last_files;
// scan all the levels
for (int level = start_level; level < end_level; ++level) {
const LevelFilesBrief& files_brief = vstorage->LevelFilesBrief(level);
if (files_brief.num_files == 0) {
// empty level, skip exploration
continue;
}
if (level == 0) {
// level 0 files are not in sorted order, we need to iterate through
// the list to compute the total bytes that require scanning,
// so handle the case explicitly (similarly to first_files case)
for (size_t i = 0; i < files_brief.num_files; i++) {
first_files.push_back(&files_brief.files[i]);
}
continue;
}
assert(level > 0);
assert(files_brief.num_files > 0);
// identify the file position for start key
const int idx_start =
FindFileInRange(icmp, files_brief, start, 0,
static_cast<uint32_t>(files_brief.num_files - 1));
assert(static_cast<size_t>(idx_start) < files_brief.num_files);
// identify the file position for end key
int idx_end = idx_start;
if (icmp.Compare(files_brief.files[idx_end].largest_key, end) < 0) {
idx_end =
FindFileInRange(icmp, files_brief, end, idx_start,
static_cast<uint32_t>(files_brief.num_files - 1));
}
assert(idx_end >= idx_start &&
static_cast<size_t>(idx_end) < files_brief.num_files);
// scan all files from the starting index to the ending index
// (inferred from the sorted order)
// first scan all the intermediate full files (excluding first and last)
for (int i = idx_start + 1; i < idx_end; ++i) {
uint64_t file_size = files_brief.files[i].fd.GetFileSize();
// The entire file falls into the range, so we can just take its size.
assert(file_size ==
ApproximateSize(v, files_brief.files[i], start, end, caller));
total_full_size += file_size;
}
// save the first and the last files (which may be the same file), so we
// can scan them later.
first_files.push_back(&files_brief.files[idx_start]);
if (idx_start != idx_end) {
// we need to estimate size for both files, only if they are different
last_files.push_back(&files_brief.files[idx_end]);
}
}
// The sum of all file sizes that intersect the [start, end] keys range.
uint64_t total_intersecting_size = 0;
for (const auto* file_ptr : first_files) {
total_intersecting_size += file_ptr->fd.GetFileSize();
}
for (const auto* file_ptr : last_files) {
total_intersecting_size += file_ptr->fd.GetFileSize();
}
// Now scan all the first & last files at each level, and estimate their size.
// If the total_intersecting_size is less than X% of the total_full_size - we
// want to approximate the result in order to avoid the costly binary search
// inside ApproximateSize. We use half of file size as an approximation below.
const double margin = options.files_size_error_margin;
if (margin > 0 && total_intersecting_size <
static_cast<uint64_t>(total_full_size * margin)) {
total_full_size += total_intersecting_size / 2;
} else {
For ApproximateSizes, pro-rate table metadata size over data blocks (#6784) Summary: The implementation of GetApproximateSizes was inconsistent in its treatment of the size of non-data blocks of SST files, sometimes including and sometimes now. This was at its worst with large portion of table file used by filters and querying a small range that crossed a table boundary: the size estimate would include large filter size. It's conceivable that someone might want only to know the size in terms of data blocks, but I believe that's unlikely enough to ignore for now. Similarly, there's no evidence the internal function AppoximateOffsetOf is used for anything other than a one-sided ApproximateSize, so I intend to refactor to remove redundancy in a follow-up commit. So to fix this, GetApproximateSizes (and implementation details ApproximateSize and ApproximateOffsetOf) now consistently include in their returned sizes a portion of table file metadata (incl filters and indexes) based on the size portion of the data blocks in range. In other words, if a key range covers data blocks that are X% by size of all the table's data blocks, returned approximate size is X% of the total file size. It would technically be more accurate to attribute metadata based on number of keys, but that's not computationally efficient with data available and rarely a meaningful difference. Also includes miscellaneous comment improvements / clarifications. Also included is a new approximatesizerandom benchmark for db_bench. No significant performance difference seen with this change, whether ~700 ops/sec with cache_index_and_filter_blocks and small cache or ~150k ops/sec without cache_index_and_filter_blocks. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6784 Test Plan: Test added to DBTest.ApproximateSizesFilesWithErrorMargin. Old code running new test... [ RUN ] DBTest.ApproximateSizesFilesWithErrorMargin db/db_test.cc:1562: Failure Expected: (size) <= (11 * 100), actual: 9478 vs 1100 Other tests updated to reflect consistent accounting of metadata. Reviewed By: siying Differential Revision: D21334706 Pulled By: pdillinger fbshipit-source-id: 6f86870e45213334fedbe9c73b4ebb1d8d611185
2020-06-02 21:27:59 +02:00
// Estimate for all the first files (might also be last files), at each
// level
for (const auto file_ptr : first_files) {
total_full_size += ApproximateSize(v, *file_ptr, start, end, caller);
}
// Estimate for all the last files, at each level
for (const auto file_ptr : last_files) {
// We could use ApproximateSize here, but calling ApproximateOffsetOf
// directly is just more efficient.
total_full_size += ApproximateOffsetOf(v, *file_ptr, end, caller);
}
}
return total_full_size;
}
uint64_t VersionSet::ApproximateOffsetOf(Version* v, const FdWithKeyRange& f,
const Slice& key,
TableReaderCaller caller) {
// pre-condition
assert(v);
const auto& icmp = v->cfd_->internal_comparator();
uint64_t result = 0;
if (icmp.Compare(f.largest_key, key) <= 0) {
// Entire file is before "key", so just add the file size
result = f.fd.GetFileSize();
} else if (icmp.Compare(f.smallest_key, key) > 0) {
// Entire file is after "key", so ignore
result = 0;
} else {
// "key" falls in the range for this table. Add the
// approximate offset of "key" within the table.
TableCache* table_cache = v->cfd_->table_cache();
if (table_cache != nullptr) {
result = table_cache->ApproximateOffsetOf(
key, f.file_metadata->fd, caller, icmp,
v->GetMutableCFOptions().prefix_extractor.get());
}
}
return result;
}
uint64_t VersionSet::ApproximateSize(Version* v, const FdWithKeyRange& f,
const Slice& start, const Slice& end,
TableReaderCaller caller) {
// pre-condition
assert(v);
const auto& icmp = v->cfd_->internal_comparator();
assert(icmp.Compare(start, end) <= 0);
if (icmp.Compare(f.largest_key, start) <= 0 ||
icmp.Compare(f.smallest_key, end) > 0) {
// Entire file is before or after the start/end keys range
return 0;
}
if (icmp.Compare(f.smallest_key, start) >= 0) {
// Start of the range is before the file start - approximate by end offset
return ApproximateOffsetOf(v, f, end, caller);
}
if (icmp.Compare(f.largest_key, end) < 0) {
// End of the range is after the file end - approximate by subtracting
// start offset from the file size
uint64_t start_offset = ApproximateOffsetOf(v, f, start, caller);
assert(f.fd.GetFileSize() >= start_offset);
return f.fd.GetFileSize() - start_offset;
}
// The interval falls entirely in the range for this file.
TableCache* table_cache = v->cfd_->table_cache();
if (table_cache == nullptr) {
return 0;
}
return table_cache->ApproximateSize(
start, end, f.file_metadata->fd, caller, icmp,
v->GetMutableCFOptions().prefix_extractor.get());
}
void VersionSet::AddLiveFiles(std::vector<uint64_t>* live_table_files,
std::vector<uint64_t>* live_blob_files) const {
assert(live_table_files);
assert(live_blob_files);
// pre-calculate space requirement
size_t total_table_files = 0;
size_t total_blob_files = 0;
assert(column_family_set_);
for (auto cfd : *column_family_set_) {
assert(cfd);
if (!cfd->initialized()) {
continue;
}
Version* const dummy_versions = cfd->dummy_versions();
assert(dummy_versions);
for (Version* v = dummy_versions->next_; v != dummy_versions;
v = v->next_) {
assert(v);
const auto* vstorage = v->storage_info();
assert(vstorage);
for (int level = 0; level < vstorage->num_levels(); ++level) {
total_table_files += vstorage->LevelFiles(level).size();
}
total_blob_files += vstorage->GetBlobFiles().size();
}
}
// just one time extension to the right size
live_table_files->reserve(live_table_files->size() + total_table_files);
live_blob_files->reserve(live_blob_files->size() + total_blob_files);
assert(column_family_set_);
for (auto cfd : *column_family_set_) {
assert(cfd);
if (!cfd->initialized()) {
continue;
}
auto* current = cfd->current();
bool found_current = false;
Version* const dummy_versions = cfd->dummy_versions();
assert(dummy_versions);
for (Version* v = dummy_versions->next_; v != dummy_versions;
v = v->next_) {
v->AddLiveFiles(live_table_files, live_blob_files);
if (v == current) {
found_current = true;
}
}
if (!found_current && current != nullptr) {
// Should never happen unless it is a bug.
assert(false);
current->AddLiveFiles(live_table_files, live_blob_files);
}
}
}
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
InternalIterator* VersionSet::MakeInputIterator(
const ReadOptions& read_options, const Compaction* c,
RangeDelAggregator* range_del_agg,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options_compactions) {
auto cfd = c->column_family_data();
// Level-0 files have to be merged together. For other levels,
// we will make a concatenating iterator per level.
// TODO(opt): use concatenating iterator for level-0 if there is no overlap
const size_t space = (c->level() == 0 ? c->input_levels(0)->num_files +
c->num_input_levels() - 1
: c->num_input_levels());
InternalIterator** list = new InternalIterator* [space];
size_t num = 0;
for (size_t which = 0; which < c->num_input_levels(); which++) {
if (c->input_levels(which)->num_files != 0) {
if (c->level(which) == 0) {
const LevelFilesBrief* flevel = c->input_levels(which);
for (size_t i = 0; i < flevel->num_files; i++) {
list[num++] = cfd->table_cache()->NewIterator(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
read_options, file_options_compactions,
cfd->internal_comparator(), *flevel->files[i].file_metadata,
range_del_agg, c->mutable_cf_options()->prefix_extractor.get(),
/*table_reader_ptr=*/nullptr,
/*file_read_hist=*/nullptr, TableReaderCaller::kCompaction,
/*arena=*/nullptr,
/*skip_filters=*/false,
/*level=*/static_cast<int>(c->level(which)),
MaxFileSizeForL0MetaPin(*c->mutable_cf_options()),
/*smallest_compaction_key=*/nullptr,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
/*largest_compaction_key=*/nullptr,
/*allow_unprepared_value=*/false);
}
} else {
// Create concatenating iterator for the files from this level
list[num++] = new LevelIterator(
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
cfd->table_cache(), read_options, file_options_compactions,
cfd->internal_comparator(), c->input_levels(which),
c->mutable_cf_options()->prefix_extractor.get(),
/*should_sample=*/false,
/*no per level latency histogram=*/nullptr,
TableReaderCaller::kCompaction, /*skip_filters=*/false,
/*level=*/static_cast<int>(c->level(which)), range_del_agg,
c->boundaries(which));
}
}
}
assert(num <= space);
InternalIterator* result =
NewMergingIterator(&c->column_family_data()->internal_comparator(), list,
static_cast<int>(num));
delete[] list;
return result;
}
Status VersionSet::GetMetadataForFile(uint64_t number, int* filelevel,
FileMetaData** meta,
ColumnFamilyData** cfd) {
for (auto cfd_iter : *column_family_set_) {
if (!cfd_iter->initialized()) {
continue;
}
Version* version = cfd_iter->current();
const auto* vstorage = version->storage_info();
for (int level = 0; level < vstorage->num_levels(); level++) {
for (const auto& file : vstorage->LevelFiles(level)) {
if (file->fd.GetNumber() == number) {
*meta = file;
*filelevel = level;
*cfd = cfd_iter;
return Status::OK();
}
}
}
}
return Status::NotFound("File not present in any level");
}
void VersionSet::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
for (auto cfd : *column_family_set_) {
if (cfd->IsDropped() || !cfd->initialized()) {
continue;
}
for (int level = 0; level < cfd->NumberLevels(); level++) {
for (const auto& file :
cfd->current()->storage_info()->LevelFiles(level)) {
LiveFileMetaData filemetadata;
filemetadata.column_family_name = cfd->GetName();
uint32_t path_id = file->fd.GetPathId();
if (path_id < cfd->ioptions()->cf_paths.size()) {
filemetadata.db_path = cfd->ioptions()->cf_paths[path_id].path;
} else {
assert(!cfd->ioptions()->cf_paths.empty());
filemetadata.db_path = cfd->ioptions()->cf_paths.back().path;
}
const uint64_t file_number = file->fd.GetNumber();
filemetadata.name = MakeTableFileName("", file_number);
filemetadata.file_number = file_number;
filemetadata.level = level;
filemetadata.size = static_cast<size_t>(file->fd.GetFileSize());
filemetadata.smallestkey = file->smallest.user_key().ToString();
filemetadata.largestkey = file->largest.user_key().ToString();
filemetadata.smallest_seqno = file->fd.smallest_seqno;
filemetadata.largest_seqno = file->fd.largest_seqno;
filemetadata.num_reads_sampled = file->stats.num_reads_sampled.load(
std::memory_order_relaxed);
filemetadata.being_compacted = file->being_compacted;
filemetadata.num_entries = file->num_entries;
filemetadata.num_deletions = file->num_deletions;
filemetadata.oldest_blob_file_number = file->oldest_blob_file_number;
filemetadata.file_checksum = file->file_checksum;
filemetadata.file_checksum_func_name = file->file_checksum_func_name;
filemetadata.temperature = file->temperature;
filemetadata.oldest_ancester_time = file->TryGetOldestAncesterTime();
filemetadata.file_creation_time = file->TryGetFileCreationTime();
metadata->push_back(filemetadata);
}
}
}
}
void VersionSet::GetObsoleteFiles(std::vector<ObsoleteFileInfo>* files,
std::vector<ObsoleteBlobFileInfo>* blob_files,
std::vector<std::string>* manifest_filenames,
uint64_t min_pending_output) {
assert(files);
assert(blob_files);
assert(manifest_filenames);
assert(files->empty());
assert(blob_files->empty());
assert(manifest_filenames->empty());
std::vector<ObsoleteFileInfo> pending_files;
for (auto& f : obsolete_files_) {
if (f.metadata->fd.GetNumber() < min_pending_output) {
files->emplace_back(std::move(f));
} else {
pending_files.emplace_back(std::move(f));
}
}
obsolete_files_.swap(pending_files);
std::vector<ObsoleteBlobFileInfo> pending_blob_files;
for (auto& blob_file : obsolete_blob_files_) {
if (blob_file.GetBlobFileNumber() < min_pending_output) {
blob_files->emplace_back(std::move(blob_file));
} else {
pending_blob_files.emplace_back(std::move(blob_file));
}
}
obsolete_blob_files_.swap(pending_blob_files);
obsolete_manifests_.swap(*manifest_filenames);
}
ColumnFamilyData* VersionSet::CreateColumnFamily(
const ColumnFamilyOptions& cf_options, const VersionEdit* edit) {
assert(edit->is_column_family_add_);
MutableCFOptions dummy_cf_options;
Version* dummy_versions =
new Version(nullptr, this, file_options_, dummy_cf_options, io_tracer_);
// Ref() dummy version once so that later we can call Unref() to delete it
// by avoiding calling "delete" explicitly (~Version is private)
dummy_versions->Ref();
auto new_cfd = column_family_set_->CreateColumnFamily(
edit->column_family_name_, edit->column_family_, dummy_versions,
cf_options);
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
Version* v = new Version(new_cfd, this, file_options_,
*new_cfd->GetLatestMutableCFOptions(), io_tracer_,
current_version_number_++);
2014-03-18 22:23:47 +01:00
// Fill level target base information.
v->storage_info()->CalculateBaseBytes(*new_cfd->ioptions(),
*new_cfd->GetLatestMutableCFOptions());
2014-03-18 22:23:47 +01:00
AppendVersion(new_cfd, v);
// GetLatestMutableCFOptions() is safe here without mutex since the
// cfd is not available to client
new_cfd->CreateNewMemtable(*new_cfd->GetLatestMutableCFOptions(),
LastSequence());
2014-02-28 20:08:24 +01:00
new_cfd->SetLogNumber(edit->log_number_);
return new_cfd;
}
uint64_t VersionSet::GetNumLiveVersions(Version* dummy_versions) {
uint64_t count = 0;
for (Version* v = dummy_versions->next_; v != dummy_versions; v = v->next_) {
count++;
}
return count;
}
uint64_t VersionSet::GetTotalSstFilesSize(Version* dummy_versions) {
std::unordered_set<uint64_t> unique_files;
uint64_t total_files_size = 0;
for (Version* v = dummy_versions->next_; v != dummy_versions; v = v->next_) {
VersionStorageInfo* storage_info = v->storage_info();
for (int level = 0; level < storage_info->num_levels_; level++) {
for (const auto& file_meta : storage_info->LevelFiles(level)) {
if (unique_files.find(file_meta->fd.packed_number_and_path_id) ==
unique_files.end()) {
unique_files.insert(file_meta->fd.packed_number_and_path_id);
total_files_size += file_meta->fd.GetFileSize();
}
}
}
}
return total_files_size;
}
Add DB properties for BlobDB (#8734) Summary: RocksDB exposes certain internal statistics via the DB property interface. However, there are currently no properties related to BlobDB. For starters, we would like to add the following BlobDB properties: `rocksdb.num-blob-files`: number of blob files in the current Version (kind of like `num-files-at-level` but note this is not per level, since blob files are not part of the LSM tree). `rocksdb.blob-stats`: this could return the total number and size of all blob files, and potentially also the total amount of garbage (in bytes) in the blob files in the current Version. `rocksdb.total-blob-file-size`: the total size of all blob files (as a blob counterpart for `total-sst-file-size`) of all Versions. `rocksdb.live-blob-file-size`: the total size of all blob files in the current Version. `rocksdb.estimate-live-data-size`: this is actually an existing property that we can extend so it considers blob files as well. When it comes to blobs, we actually have an exact value for live bytes. Namely, live bytes can be computed simply as total bytes minus garbage bytes, summed over the entire set of blob files in the Version. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8734 Test Plan: ``` ➜ rocksdb git:(new_feature_blobDB_properties) ./db_blob_basic_test [==========] Running 16 tests from 2 test cases. [----------] Global test environment set-up. [----------] 10 tests from DBBlobBasicTest [ RUN ] DBBlobBasicTest.GetBlob [ OK ] DBBlobBasicTest.GetBlob (12 ms) [ RUN ] DBBlobBasicTest.MultiGetBlobs [ OK ] DBBlobBasicTest.MultiGetBlobs (11 ms) [ RUN ] DBBlobBasicTest.GetBlob_CorruptIndex [ OK ] DBBlobBasicTest.GetBlob_CorruptIndex (10 ms) [ RUN ] DBBlobBasicTest.GetBlob_InlinedTTLIndex [ OK ] DBBlobBasicTest.GetBlob_InlinedTTLIndex (12 ms) [ RUN ] DBBlobBasicTest.GetBlob_IndexWithInvalidFileNumber [ OK ] DBBlobBasicTest.GetBlob_IndexWithInvalidFileNumber (9 ms) [ RUN ] DBBlobBasicTest.GenerateIOTracing [ OK ] DBBlobBasicTest.GenerateIOTracing (11 ms) [ RUN ] DBBlobBasicTest.BestEffortsRecovery_MissingNewestBlobFile [ OK ] DBBlobBasicTest.BestEffortsRecovery_MissingNewestBlobFile (13 ms) [ RUN ] DBBlobBasicTest.GetMergeBlobWithPut [ OK ] DBBlobBasicTest.GetMergeBlobWithPut (11 ms) [ RUN ] DBBlobBasicTest.MultiGetMergeBlobWithPut [ OK ] DBBlobBasicTest.MultiGetMergeBlobWithPut (14 ms) [ RUN ] DBBlobBasicTest.BlobDBProperties [ OK ] DBBlobBasicTest.BlobDBProperties (21 ms) [----------] 10 tests from DBBlobBasicTest (124 ms total) [----------] 6 tests from DBBlobBasicTest/DBBlobBasicIOErrorTest [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/0 (12 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.GetBlob_IOError/1 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/0 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.MultiGetBlobs_IOError/1 (10 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/0 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/0 (1011 ms) [ RUN ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/1 [ OK ] DBBlobBasicTest/DBBlobBasicIOErrorTest.CompactionFilterReadBlob_IOError/1 (1013 ms) [----------] 6 tests from DBBlobBasicTest/DBBlobBasicIOErrorTest (2066 ms total) [----------] Global test environment tear-down [==========] 16 tests from 2 test cases ran. (2190 ms total) [ PASSED ] 16 tests. ``` Reviewed By: ltamasi Differential Revision: D30690849 Pulled By: Zhiyi-Zhang fbshipit-source-id: a7567319487ad76bd1a2e24bf143afdbbd9e4346
2021-09-08 21:19:01 +02:00
uint64_t VersionSet::GetTotalBlobFileSize(Version* dummy_versions) {
std::unordered_set<uint64_t> unique_blob_files;
uint64_t all_v_blob_file_size = 0;
for (auto* v = dummy_versions->next_; v != dummy_versions; v = v->next_) {
// iterate all the versions
auto* vstorage = v->storage_info();
const auto& blob_files = vstorage->GetBlobFiles();
for (const auto& pair : blob_files) {
if (unique_blob_files.find(pair.first) == unique_blob_files.end()) {
// find Blob file that has not been counted
unique_blob_files.insert(pair.first);
const auto& meta = pair.second;
all_v_blob_file_size += meta->GetTotalBlobBytes();
}
}
}
return all_v_blob_file_size;
}
Status VersionSet::VerifyFileMetadata(const std::string& fpath,
const FileMetaData& meta) const {
uint64_t fsize = 0;
Status status = fs_->GetFileSize(fpath, IOOptions(), &fsize, nullptr);
if (status.ok()) {
if (fsize != meta.fd.GetFileSize()) {
status = Status::Corruption("File size mismatch: " + fpath);
}
}
return status;
}
ReactiveVersionSet::ReactiveVersionSet(
const std::string& dbname, const ImmutableDBOptions* _db_options,
const FileOptions& _file_options, Cache* table_cache,
WriteBufferManager* write_buffer_manager, WriteController* write_controller,
const std::shared_ptr<IOTracer>& io_tracer)
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
: VersionSet(dbname, _db_options, _file_options, table_cache,
write_buffer_manager, write_controller,
/*block_cache_tracer=*/nullptr, io_tracer,
/*db_session_id*/ "") {}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
ReactiveVersionSet::~ReactiveVersionSet() {}
Status ReactiveVersionSet::Recover(
const std::vector<ColumnFamilyDescriptor>& column_families,
std::unique_ptr<log::FragmentBufferedReader>* manifest_reader,
std::unique_ptr<log::Reader::Reporter>* manifest_reporter,
std::unique_ptr<Status>* manifest_reader_status) {
assert(manifest_reader != nullptr);
assert(manifest_reporter != nullptr);
assert(manifest_reader_status != nullptr);
manifest_reader_status->reset(new Status());
manifest_reporter->reset(new LogReporter());
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
static_cast_with_check<LogReporter>(manifest_reporter->get())->status =
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
manifest_reader_status->get();
Status s = MaybeSwitchManifest(manifest_reporter->get(), manifest_reader);
if (!s.ok()) {
return s;
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
log::Reader* reader = manifest_reader->get();
assert(reader);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
manifest_tailer_.reset(new ManifestTailer(
column_families, const_cast<ReactiveVersionSet*>(this), io_tracer_));
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
manifest_tailer_->Iterate(*reader, manifest_reader_status->get());
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
return manifest_tailer_->status();
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
}
Status ReactiveVersionSet::ReadAndApply(
InstrumentedMutex* mu,
std::unique_ptr<log::FragmentBufferedReader>* manifest_reader,
Status* manifest_read_status,
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
std::unordered_set<ColumnFamilyData*>* cfds_changed) {
assert(manifest_reader != nullptr);
assert(cfds_changed != nullptr);
mu->AssertHeld();
Status s;
log::Reader* reader = manifest_reader->get();
assert(reader);
s = MaybeSwitchManifest(reader->GetReporter(), manifest_reader);
if (!s.ok()) {
return s;
}
manifest_tailer_->Iterate(*(manifest_reader->get()), manifest_read_status);
s = manifest_tailer_->status();
if (s.ok()) {
*cfds_changed = std::move(manifest_tailer_->GetUpdatedColumnFamilies());
}
return s;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
}
Status ReactiveVersionSet::MaybeSwitchManifest(
log::Reader::Reporter* reporter,
std::unique_ptr<log::FragmentBufferedReader>* manifest_reader) {
assert(manifest_reader != nullptr);
Status s;
std::string manifest_path;
s = GetCurrentManifestPath(dbname_, fs_.get(), &manifest_path,
&manifest_file_number_);
if (!s.ok()) {
return s;
}
std::unique_ptr<FSSequentialFile> manifest_file;
if (manifest_reader->get() != nullptr &&
manifest_reader->get()->file()->file_name() == manifest_path) {
// CURRENT points to the same MANIFEST as before, no need to switch
// MANIFEST.
return s;
}
assert(nullptr == manifest_reader->get() ||
manifest_reader->get()->file()->file_name() != manifest_path);
s = fs_->FileExists(manifest_path, IOOptions(), nullptr);
if (s.IsNotFound()) {
return Status::TryAgain(
"The primary may have switched to a new MANIFEST and deleted the old "
"one.");
} else if (!s.ok()) {
return s;
}
TEST_SYNC_POINT(
"ReactiveVersionSet::MaybeSwitchManifest:"
"AfterGetCurrentManifestPath:0");
TEST_SYNC_POINT(
"ReactiveVersionSet::MaybeSwitchManifest:"
"AfterGetCurrentManifestPath:1");
// The primary can also delete the MANIFEST while the secondary is reading
// it. This is OK on POSIX. For other file systems, maybe create a hard link
// to MANIFEST. The hard link should be cleaned up later by the secondary.
s = fs_->NewSequentialFile(manifest_path,
fs_->OptimizeForManifestRead(file_options_),
&manifest_file, nullptr);
std::unique_ptr<SequentialFileReader> manifest_file_reader;
if (s.ok()) {
manifest_file_reader.reset(
new SequentialFileReader(std::move(manifest_file), manifest_path,
db_options_->log_readahead_size, io_tracer_));
manifest_reader->reset(new log::FragmentBufferedReader(
nullptr, std::move(manifest_file_reader), reporter, true /* checksum */,
0 /* log_number */));
ROCKS_LOG_INFO(db_options_->info_log, "Switched to new manifest: %s\n",
manifest_path.c_str());
if (manifest_tailer_) {
manifest_tailer_->PrepareToReadNewManifest();
}
} else if (s.IsPathNotFound()) {
// This can happen if the primary switches to a new MANIFEST after the
// secondary reads the CURRENT file but before the secondary actually tries
// to open the MANIFEST.
s = Status::TryAgain(
"The primary may have switched to a new MANIFEST and deleted the old "
"one.");
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
return s;
}
#ifndef NDEBUG
uint64_t ReactiveVersionSet::TEST_read_edits_in_atomic_group() const {
assert(manifest_tailer_);
return manifest_tailer_->GetReadBuffer().TEST_read_edits_in_atomic_group();
}
#endif // !NDEBUG
std::vector<VersionEdit>& ReactiveVersionSet::replay_buffer() {
assert(manifest_tailer_);
return manifest_tailer_->GetReadBuffer().replay_buffer();
}
} // namespace ROCKSDB_NAMESPACE