rocksdb/table/block_based/block_based_table_factory.cc

819 lines
35 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 "table/block_based/block_based_table_factory.h"
#include <stdint.h>
#include <cinttypes>
#include <memory>
#include <string>
#include "options/options_helper.h"
#include "options/options_parser.h"
#include "port/port.h"
#include "rocksdb/cache.h"
#include "rocksdb/convenience.h"
#include "rocksdb/flush_block_policy.h"
#include "table/block_based/block_based_table_builder.h"
#include "table/block_based/block_based_table_reader.h"
#include "table/format.h"
#include "util/mutexlock.h"
#include "util/string_util.h"
namespace ROCKSDB_NAMESPACE {
void TailPrefetchStats::RecordEffectiveSize(size_t len) {
MutexLock l(&mutex_);
if (num_records_ < kNumTracked) {
num_records_++;
}
records_[next_++] = len;
if (next_ == kNumTracked) {
next_ = 0;
}
}
size_t TailPrefetchStats::GetSuggestedPrefetchSize() {
std::vector<size_t> sorted;
{
MutexLock l(&mutex_);
if (num_records_ == 0) {
return 0;
}
sorted.assign(records_, records_ + num_records_);
}
// Of the historic size, we find the maximum one that satisifis the condtiion
// that if prefetching all, less than 1/8 will be wasted.
std::sort(sorted.begin(), sorted.end());
// Assuming we have 5 data points, and after sorting it looks like this:
//
// +---+
// +---+ | |
// | | | |
// | | | |
// | | | |
// | | | |
// +---+ | | | |
// | | | | | |
// +---+ | | | | | |
// | | | | | | | |
// +---+ | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// +---+ +---+ +---+ +---+ +---+
//
// and we use every of the value as a candidate, and estimate how much we
// wasted, compared to read. For example, when we use the 3rd record
// as candiate. This area is what we read:
// +---+
// +---+ | |
// | | | |
// | | | |
// | | | |
// | | | |
// *** *** *** ***+ *** *** *** *** **
// * | | | | | |
// +---+ | | | | | *
// * | | | | | | | |
// +---+ | | | | | | | *
// * | | | | X | | | | |
// | | | | | | | | | *
// * | | | | | | | | |
// | | | | | | | | | *
// * | | | | | | | | |
// *** *** ***-*** ***--*** ***--*** +****
// which is (size of the record) X (number of records).
//
// While wasted is this area:
// +---+
// +---+ | |
// | | | |
// | | | |
// | | | |
// | | | |
// *** *** *** ****---+ | | | |
// * * | | | | |
// * *-*** *** | | | | |
// * * | | | | | | |
// *--** *** | | | | | | |
// | | | | | X | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// +---+ +---+ +---+ +---+ +---+
//
// Which can be calculated iteratively.
// The difference between wasted using 4st and 3rd record, will
// be following area:
// +---+
// +--+ +-+ ++ +-+ +-+ +---+ | |
// + xxxxxxxxxxxxxxxxxxxxxxxx | | | |
// xxxxxxxxxxxxxxxxxxxxxxxx | | | |
// + xxxxxxxxxxxxxxxxxxxxxxxx | | | |
// | xxxxxxxxxxxxxxxxxxxxxxxx | | | |
// +-+ +-+ +-+ ++ +---+ +--+ | | |
// | | | | | | |
// +---+ ++ | | | | | |
// | | | | | | X | | |
// +---+ ++ | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// | | | | | | | | | |
// +---+ +---+ +---+ +---+ +---+
//
// which will be the size difference between 4st and 3rd record,
// times 3, which is number of records before the 4st.
// Here we assume that all data within the prefetch range will be useful. In
// reality, it may not be the case when a partial block is inside the range,
// or there are data in the middle that is not read. We ignore those cases
// for simplicity.
assert(!sorted.empty());
size_t prev_size = sorted[0];
size_t max_qualified_size = sorted[0];
size_t wasted = 0;
for (size_t i = 1; i < sorted.size(); i++) {
size_t read = sorted[i] * sorted.size();
wasted += (sorted[i] - prev_size) * i;
if (wasted <= read / 8) {
max_qualified_size = sorted[i];
}
prev_size = sorted[i];
}
const size_t kMaxPrefetchSize = 512 * 1024; // Never exceed 512KB
return std::min(kMaxPrefetchSize, max_qualified_size);
}
#ifndef ROCKSDB_LITE
static std::unordered_map<std::string, BlockBasedTableOptions::IndexType>
block_base_table_index_type_string_map = {
{"kBinarySearch", BlockBasedTableOptions::IndexType::kBinarySearch},
{"kHashSearch", BlockBasedTableOptions::IndexType::kHashSearch},
{"kTwoLevelIndexSearch",
BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch},
{"kBinarySearchWithFirstKey",
BlockBasedTableOptions::IndexType::kBinarySearchWithFirstKey}};
static std::unordered_map<std::string,
BlockBasedTableOptions::DataBlockIndexType>
block_base_table_data_block_index_type_string_map = {
{"kDataBlockBinarySearch",
BlockBasedTableOptions::DataBlockIndexType::kDataBlockBinarySearch},
{"kDataBlockBinaryAndHash",
BlockBasedTableOptions::DataBlockIndexType::kDataBlockBinaryAndHash}};
static std::unordered_map<std::string,
BlockBasedTableOptions::IndexShorteningMode>
block_base_table_index_shortening_mode_string_map = {
{"kNoShortening",
BlockBasedTableOptions::IndexShorteningMode::kNoShortening},
{"kShortenSeparators",
BlockBasedTableOptions::IndexShorteningMode::kShortenSeparators},
{"kShortenSeparatorsAndSuccessor",
BlockBasedTableOptions::IndexShorteningMode::
kShortenSeparatorsAndSuccessor}};
static std::unordered_map<std::string, OptionTypeInfo>
block_based_table_type_info = {
/* currently not supported
std::shared_ptr<Cache> block_cache = nullptr;
std::shared_ptr<Cache> block_cache_compressed = nullptr;
*/
{"flush_block_policy_factory",
{offsetof(struct BlockBasedTableOptions, flush_block_policy_factory),
OptionType::kFlushBlockPolicyFactory, OptionVerificationType::kByName,
OptionTypeFlags::kCompareNever, 0}},
{"cache_index_and_filter_blocks",
{offsetof(struct BlockBasedTableOptions,
cache_index_and_filter_blocks),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"cache_index_and_filter_blocks_with_high_priority",
{offsetof(struct BlockBasedTableOptions,
cache_index_and_filter_blocks_with_high_priority),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"pin_l0_filter_and_index_blocks_in_cache",
{offsetof(struct BlockBasedTableOptions,
pin_l0_filter_and_index_blocks_in_cache),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"index_type", OptionTypeInfo::Enum<BlockBasedTableOptions::IndexType>(
offsetof(struct BlockBasedTableOptions, index_type),
&block_base_table_index_type_string_map)},
{"hash_index_allow_collision",
{offsetof(struct BlockBasedTableOptions, hash_index_allow_collision),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"data_block_index_type",
OptionTypeInfo::Enum<BlockBasedTableOptions::DataBlockIndexType>(
offsetof(struct BlockBasedTableOptions, data_block_index_type),
&block_base_table_data_block_index_type_string_map)},
{"index_shortening",
OptionTypeInfo::Enum<BlockBasedTableOptions::IndexShorteningMode>(
offsetof(struct BlockBasedTableOptions, index_shortening),
&block_base_table_index_shortening_mode_string_map)},
{"data_block_hash_table_util_ratio",
{offsetof(struct BlockBasedTableOptions,
data_block_hash_table_util_ratio),
OptionType::kDouble, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"checksum",
{offsetof(struct BlockBasedTableOptions, checksum),
OptionType::kChecksumType, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"no_block_cache",
{offsetof(struct BlockBasedTableOptions, no_block_cache),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"block_size",
{offsetof(struct BlockBasedTableOptions, block_size),
OptionType::kSizeT, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"block_size_deviation",
{offsetof(struct BlockBasedTableOptions, block_size_deviation),
OptionType::kInt, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"block_restart_interval",
{offsetof(struct BlockBasedTableOptions, block_restart_interval),
OptionType::kInt, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"index_block_restart_interval",
{offsetof(struct BlockBasedTableOptions, index_block_restart_interval),
OptionType::kInt, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"index_per_partition",
{0, OptionType::kUInt64T, OptionVerificationType::kDeprecated,
OptionTypeFlags::kNone, 0}},
{"metadata_block_size",
{offsetof(struct BlockBasedTableOptions, metadata_block_size),
OptionType::kUInt64T, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"partition_filters",
{offsetof(struct BlockBasedTableOptions, partition_filters),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
Minimize memory internal fragmentation for Bloom filters (#6427) Summary: New experimental option BBTO::optimize_filters_for_memory builds filters that maximize their use of "usable size" from malloc_usable_size, which is also used to compute block cache charges. Rather than always "rounding up," we track state in the BloomFilterPolicy object to mix essentially "rounding down" and "rounding up" so that the average FP rate of all generated filters is the same as without the option. (YMMV as heavily accessed filters might be unluckily lower accuracy.) Thus, the option near-minimizes what the block cache considers as "memory used" for a given target Bloom filter false positive rate and Bloom filter implementation. There are no forward or backward compatibility issues with this change, though it only works on the format_version=5 Bloom filter. With Jemalloc, we see about 10% reduction in memory footprint (and block cache charge) for Bloom filters, but 1-2% increase in storage footprint, due to encoding efficiency losses (FP rate is non-linear with bits/key). Why not weighted random round up/down rather than state tracking? By only requiring malloc_usable_size, we don't actually know what the next larger and next smaller usable sizes for the allocator are. We pick a requested size, accept and use whatever usable size it has, and use the difference to inform our next choice. This allows us to narrow in on the right balance without tracking/predicting usable sizes. Why not weight history of generated filter false positive rates by number of keys? This could lead to excess skew in small filters after generating a large filter. Results from filter_bench with jemalloc (irrelevant details omitted): (normal keys/filter, but high variance) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=30000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.6278 Number of filters: 5516 Total size (MB): 200.046 Reported total allocated memory (MB): 220.597 Reported internal fragmentation: 10.2732% Bits/key stored: 10.0097 Average FP rate %: 0.965228 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=30000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 30.5104 Number of filters: 5464 Total size (MB): 200.015 Reported total allocated memory (MB): 200.322 Reported internal fragmentation: 0.153709% Bits/key stored: 10.1011 Average FP rate %: 0.966313 (very few keys / filter, optimization not as effective due to ~59 byte internal fragmentation in blocked Bloom filter representation) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.5649 Number of filters: 162950 Total size (MB): 200.001 Reported total allocated memory (MB): 224.624 Reported internal fragmentation: 12.3117% Bits/key stored: 10.2951 Average FP rate %: 0.821534 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 31.8057 Number of filters: 159849 Total size (MB): 200 Reported total allocated memory (MB): 208.846 Reported internal fragmentation: 4.42297% Bits/key stored: 10.4948 Average FP rate %: 0.811006 (high keys/filter) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.7017 Number of filters: 164 Total size (MB): 200.352 Reported total allocated memory (MB): 221.5 Reported internal fragmentation: 10.5552% Bits/key stored: 10.0003 Average FP rate %: 0.969358 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 30.7131 Number of filters: 160 Total size (MB): 200.928 Reported total allocated memory (MB): 200.938 Reported internal fragmentation: 0.00448054% Bits/key stored: 10.1852 Average FP rate %: 0.963387 And from db_bench (block cache) with jemalloc: $ ./db_bench -db=/dev/shm/dbbench.no_optimize -benchmarks=fillrandom -format_version=5 -value_size=90 -bloom_bits=10 -num=2000000 -threads=8 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=false $ ./db_bench -db=/dev/shm/dbbench -benchmarks=fillrandom -format_version=5 -value_size=90 -bloom_bits=10 -num=2000000 -threads=8 -optimize_filters_for_memory -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=false $ (for FILE in /dev/shm/dbbench.no_optimize/*.sst; do ./sst_dump --file=$FILE --show_properties | grep 'filter block' ; done) | awk '{ t += $4; } END { print t; }' 17063835 $ (for FILE in /dev/shm/dbbench/*.sst; do ./sst_dump --file=$FILE --show_properties | grep 'filter block' ; done) | awk '{ t += $4; } END { print t; }' 17430747 $ #^ 2.1% additional filter storage $ ./db_bench -db=/dev/shm/dbbench.no_optimize -use_existing_db -benchmarks=readrandom,stats -statistics -bloom_bits=10 -num=2000000 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=false -duration=10 -cache_index_and_filter_blocks -cache_size=1000000000 rocksdb.block.cache.index.add COUNT : 33 rocksdb.block.cache.index.bytes.insert COUNT : 8440400 rocksdb.block.cache.filter.add COUNT : 33 rocksdb.block.cache.filter.bytes.insert COUNT : 21087528 rocksdb.bloom.filter.useful COUNT : 4963889 rocksdb.bloom.filter.full.positive COUNT : 1214081 rocksdb.bloom.filter.full.true.positive COUNT : 1161999 $ #^ 1.04 % observed FP rate $ ./db_bench -db=/dev/shm/dbbench -use_existing_db -benchmarks=readrandom,stats -statistics -bloom_bits=10 -num=2000000 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=false -optimize_filters_for_memory -duration=10 -cache_index_and_filter_blocks -cache_size=1000000000 rocksdb.block.cache.index.add COUNT : 33 rocksdb.block.cache.index.bytes.insert COUNT : 8448592 rocksdb.block.cache.filter.add COUNT : 33 rocksdb.block.cache.filter.bytes.insert COUNT : 18220328 rocksdb.bloom.filter.useful COUNT : 5360933 rocksdb.bloom.filter.full.positive COUNT : 1321315 rocksdb.bloom.filter.full.true.positive COUNT : 1262999 $ #^ 1.08 % observed FP rate, 13.6% less memory usage for filters (Due to specific key density, this example tends to generate filters that are "worse than average" for internal fragmentation. "Better than average" cases can show little or no improvement.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/6427 Test Plan: unit test added, 'make check' with gcc, clang and valgrind Reviewed By: siying Differential Revision: D22124374 Pulled By: pdillinger fbshipit-source-id: f3e3aa152f9043ddf4fae25799e76341d0d8714e
2020-06-22 22:30:57 +02:00
{"optimize_filters_for_memory",
{offsetof(struct BlockBasedTableOptions, optimize_filters_for_memory),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"filter_policy",
{offsetof(struct BlockBasedTableOptions, filter_policy),
OptionType::kUnknown, OptionVerificationType::kByNameAllowFromNull,
OptionTypeFlags::kNone, 0,
// Parses the Filter policy
[](const ConfigOptions& opts, const std::string&,
const std::string& value, char* addr) {
auto* policy =
reinterpret_cast<std::shared_ptr<const FilterPolicy>*>(addr);
return FilterPolicy::CreateFromString(opts, value, policy);
},
// Converts the FilterPolicy to its string representation
[](const ConfigOptions&, const std::string&, const char* addr,
std::string* value) {
const auto* policy =
reinterpret_cast<const std::shared_ptr<const FilterPolicy>*>(
addr);
if (policy->get()) {
*value = (*policy)->Name();
} else {
*value = kNullptrString;
}
return Status::OK();
},
// Compares two FilterPolicy objects for equality
[](const ConfigOptions&, const std::string&, const char* addr1,
const char* addr2, std::string*) {
const auto* policy1 =
reinterpret_cast<const std::shared_ptr<const FilterPolicy>*>(
addr1)
->get();
const auto* policy2 =
reinterpret_cast<const std::shared_ptr<FilterPolicy>*>(addr2)
->get();
if (policy1 == policy2) {
return true;
} else if (policy1 != nullptr && policy2 != nullptr) {
return (strcmp(policy1->Name(), policy2->Name()) == 0);
} else {
return false;
}
}}},
{"whole_key_filtering",
{offsetof(struct BlockBasedTableOptions, whole_key_filtering),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"skip_table_builder_flush",
{0, OptionType::kBoolean, OptionVerificationType::kDeprecated,
OptionTypeFlags::kNone, 0}},
{"format_version",
{offsetof(struct BlockBasedTableOptions, format_version),
OptionType::kUInt32T, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"verify_compression",
{offsetof(struct BlockBasedTableOptions, verify_compression),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"read_amp_bytes_per_bit",
{offsetof(struct BlockBasedTableOptions, read_amp_bytes_per_bit),
OptionType::kUInt32T, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0,
[](const ConfigOptions& /*opts*/, const std::string& /*name*/,
const std::string& value, char* addr) {
// A workaround to fix a bug in 6.10, 6.11, 6.12, 6.13
// and 6.14. The bug will write out 8 bytes to OPTIONS file from the
// starting address of BlockBasedTableOptions.read_amp_bytes_per_bit
// which is actually a uint32. Consequently, the value of
// read_amp_bytes_per_bit written in the OPTIONS file is wrong.
// From 6.15, RocksDB will try to parse the read_amp_bytes_per_bit
// from OPTIONS file as a uint32. To be able to load OPTIONS file
// generated by affected releases before the fix, we need to
// manually parse read_amp_bytes_per_bit with this special hack.
uint64_t read_amp_bytes_per_bit = ParseUint64(value);
EncodeFixed32(addr, static_cast<uint32_t>(read_amp_bytes_per_bit));
return Status::OK();
}}},
{"enable_index_compression",
{offsetof(struct BlockBasedTableOptions, enable_index_compression),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"block_align",
{offsetof(struct BlockBasedTableOptions, block_align),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"pin_top_level_index_and_filter",
{offsetof(struct BlockBasedTableOptions,
pin_top_level_index_and_filter),
OptionType::kBoolean, OptionVerificationType::kNormal,
OptionTypeFlags::kNone, 0}},
{"block_cache",
{offsetof(struct BlockBasedTableOptions, block_cache),
OptionType::kUnknown, OptionVerificationType::kNormal,
(OptionTypeFlags::kCompareNever | OptionTypeFlags::kDontSerialize), 0,
// Parses the input vsalue as a Cache
[](const ConfigOptions& opts, const std::string&,
const std::string& value, char* addr) {
auto* cache = reinterpret_cast<std::shared_ptr<Cache>*>(addr);
return Cache::CreateFromString(opts, value, cache);
}}},
{"block_cache_compressed",
{offsetof(struct BlockBasedTableOptions, block_cache_compressed),
OptionType::kUnknown, OptionVerificationType::kNormal,
(OptionTypeFlags::kCompareNever | OptionTypeFlags::kDontSerialize), 0,
// Parses the input vsalue as a Cache
[](const ConfigOptions& opts, const std::string&,
const std::string& value, char* addr) {
auto* cache = reinterpret_cast<std::shared_ptr<Cache>*>(addr);
return Cache::CreateFromString(opts, value, cache);
}}},
};
#endif // ROCKSDB_LITE
// TODO(myabandeh): We should return an error instead of silently changing the
// options
BlockBasedTableFactory::BlockBasedTableFactory(
const BlockBasedTableOptions& _table_options)
: table_options_(_table_options) {
if (table_options_.flush_block_policy_factory == nullptr) {
table_options_.flush_block_policy_factory.reset(
new FlushBlockBySizePolicyFactory());
}
if (table_options_.no_block_cache) {
table_options_.block_cache.reset();
} else if (table_options_.block_cache == nullptr) {
LRUCacheOptions co;
co.capacity = 8 << 20;
// It makes little sense to pay overhead for mid-point insertion while the
// block size is only 8MB.
co.high_pri_pool_ratio = 0.0;
table_options_.block_cache = NewLRUCache(co);
}
if (table_options_.block_size_deviation < 0 ||
table_options_.block_size_deviation > 100) {
table_options_.block_size_deviation = 0;
}
if (table_options_.block_restart_interval < 1) {
table_options_.block_restart_interval = 1;
}
if (table_options_.index_block_restart_interval < 1) {
table_options_.index_block_restart_interval = 1;
}
if (table_options_.index_type == BlockBasedTableOptions::kHashSearch &&
table_options_.index_block_restart_interval != 1) {
// Currently kHashSearch is incompatible with index_block_restart_interval > 1
table_options_.index_block_restart_interval = 1;
}
if (table_options_.partition_filters &&
table_options_.index_type !=
BlockBasedTableOptions::kTwoLevelIndexSearch) {
// We do not support partitioned filters without partitioning indexes
table_options_.partition_filters = false;
}
}
Status BlockBasedTableFactory::NewTableReader(
const ReadOptions& ro, const TableReaderOptions& table_reader_options,
std::unique_ptr<RandomAccessFileReader>&& file, uint64_t file_size,
std::unique_ptr<TableReader>* table_reader,
bool prefetch_index_and_filter_in_cache) const {
return BlockBasedTable::Open(
ro, table_reader_options.ioptions, table_reader_options.env_options,
table_options_, table_reader_options.internal_comparator, std::move(file),
file_size, table_reader, table_reader_options.prefix_extractor,
prefetch_index_and_filter_in_cache, table_reader_options.skip_filters,
table_reader_options.level, table_reader_options.immortal,
table_reader_options.largest_seqno,
table_reader_options.force_direct_prefetch, &tail_prefetch_stats_,
table_reader_options.block_cache_tracer,
table_reader_options.max_file_size_for_l0_meta_pin);
}
TableBuilder* BlockBasedTableFactory::NewTableBuilder(
const TableBuilderOptions& table_builder_options, uint32_t column_family_id,
WritableFileWriter* file) const {
auto table_builder = new BlockBasedTableBuilder(
table_builder_options.ioptions, table_builder_options.moptions,
table_options_, table_builder_options.internal_comparator,
table_builder_options.int_tbl_prop_collector_factories, column_family_id,
file, table_builder_options.compression_type,
table_builder_options.sample_for_compression,
table_builder_options.compression_opts,
table_builder_options.skip_filters,
table_builder_options.column_family_name, table_builder_options.level,
table_builder_options.creation_time,
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
2019-02-12 04:42:25 +01:00
table_builder_options.oldest_key_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
table_builder_options.target_file_size,
table_builder_options.file_creation_time, table_builder_options.db_id,
table_builder_options.db_session_id);
return table_builder;
}
Status BlockBasedTableFactory::SanitizeOptions(
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
const DBOptions& db_opts, const ColumnFamilyOptions& cf_opts) const {
if (table_options_.index_type == BlockBasedTableOptions::kHashSearch &&
cf_opts.prefix_extractor == nullptr) {
return Status::InvalidArgument(
"Hash index is specified for block-based "
"table, but prefix_extractor is not given");
}
if (table_options_.cache_index_and_filter_blocks &&
table_options_.no_block_cache) {
return Status::InvalidArgument(
"Enable cache_index_and_filter_blocks, "
", but block cache is disabled");
}
if (table_options_.pin_l0_filter_and_index_blocks_in_cache &&
table_options_.no_block_cache) {
return Status::InvalidArgument(
"Enable pin_l0_filter_and_index_blocks_in_cache, "
", but block cache is disabled");
}
if (!BlockBasedTableSupportedVersion(table_options_.format_version)) {
return Status::InvalidArgument(
"Unsupported BlockBasedTable format_version. Please check "
"include/rocksdb/table.h for more info");
}
if (table_options_.block_align && (cf_opts.compression != kNoCompression)) {
return Status::InvalidArgument(
"Enable block_align, but compression "
"enabled");
}
if (table_options_.block_align &&
(table_options_.block_size & (table_options_.block_size - 1))) {
return Status::InvalidArgument(
"Block alignment requested but block size is not a power of 2");
}
if (table_options_.block_size > port::kMaxUint32) {
return Status::InvalidArgument(
"block size exceeds maximum number (4GiB) allowed");
}
if (table_options_.data_block_index_type ==
BlockBasedTableOptions::kDataBlockBinaryAndHash &&
table_options_.data_block_hash_table_util_ratio <= 0) {
return Status::InvalidArgument(
"data_block_hash_table_util_ratio should be greater than 0 when "
"data_block_index_type is set to kDataBlockBinaryAndHash");
}
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
if (db_opts.unordered_write && cf_opts.max_successive_merges > 0) {
// TODO(myabandeh): support it
return Status::InvalidArgument(
"max_successive_merges larger than 0 is currently inconsistent with "
"unordered_write");
}
return Status::OK();
}
std::string BlockBasedTableFactory::GetPrintableTableOptions() const {
std::string ret;
ret.reserve(20000);
const int kBufferSize = 200;
char buffer[kBufferSize];
snprintf(buffer, kBufferSize, " flush_block_policy_factory: %s (%p)\n",
table_options_.flush_block_policy_factory->Name(),
static_cast<void*>(table_options_.flush_block_policy_factory.get()));
ret.append(buffer);
snprintf(buffer, kBufferSize, " cache_index_and_filter_blocks: %d\n",
table_options_.cache_index_and_filter_blocks);
ret.append(buffer);
snprintf(buffer, kBufferSize,
" cache_index_and_filter_blocks_with_high_priority: %d\n",
table_options_.cache_index_and_filter_blocks_with_high_priority);
ret.append(buffer);
snprintf(buffer, kBufferSize,
" pin_l0_filter_and_index_blocks_in_cache: %d\n",
table_options_.pin_l0_filter_and_index_blocks_in_cache);
ret.append(buffer);
snprintf(buffer, kBufferSize, " pin_top_level_index_and_filter: %d\n",
table_options_.pin_top_level_index_and_filter);
ret.append(buffer);
snprintf(buffer, kBufferSize, " index_type: %d\n",
table_options_.index_type);
ret.append(buffer);
snprintf(buffer, kBufferSize, " data_block_index_type: %d\n",
table_options_.data_block_index_type);
ret.append(buffer);
snprintf(buffer, kBufferSize, " index_shortening: %d\n",
static_cast<int>(table_options_.index_shortening));
ret.append(buffer);
snprintf(buffer, kBufferSize, " data_block_hash_table_util_ratio: %lf\n",
table_options_.data_block_hash_table_util_ratio);
ret.append(buffer);
snprintf(buffer, kBufferSize, " hash_index_allow_collision: %d\n",
table_options_.hash_index_allow_collision);
ret.append(buffer);
snprintf(buffer, kBufferSize, " checksum: %d\n", table_options_.checksum);
ret.append(buffer);
snprintf(buffer, kBufferSize, " no_block_cache: %d\n",
table_options_.no_block_cache);
ret.append(buffer);
snprintf(buffer, kBufferSize, " block_cache: %p\n",
static_cast<void*>(table_options_.block_cache.get()));
ret.append(buffer);
if (table_options_.block_cache) {
const char* block_cache_name = table_options_.block_cache->Name();
if (block_cache_name != nullptr) {
snprintf(buffer, kBufferSize, " block_cache_name: %s\n",
block_cache_name);
ret.append(buffer);
}
ret.append(" block_cache_options:\n");
ret.append(table_options_.block_cache->GetPrintableOptions());
}
snprintf(buffer, kBufferSize, " block_cache_compressed: %p\n",
static_cast<void*>(table_options_.block_cache_compressed.get()));
ret.append(buffer);
if (table_options_.block_cache_compressed) {
const char* block_cache_compressed_name =
table_options_.block_cache_compressed->Name();
if (block_cache_compressed_name != nullptr) {
snprintf(buffer, kBufferSize, " block_cache_name: %s\n",
block_cache_compressed_name);
ret.append(buffer);
}
ret.append(" block_cache_compressed_options:\n");
ret.append(table_options_.block_cache_compressed->GetPrintableOptions());
}
snprintf(buffer, kBufferSize, " persistent_cache: %p\n",
static_cast<void*>(table_options_.persistent_cache.get()));
ret.append(buffer);
if (table_options_.persistent_cache) {
snprintf(buffer, kBufferSize, " persistent_cache_options:\n");
ret.append(buffer);
ret.append(table_options_.persistent_cache->GetPrintableOptions());
}
snprintf(buffer, kBufferSize, " block_size: %" ROCKSDB_PRIszt "\n",
table_options_.block_size);
ret.append(buffer);
snprintf(buffer, kBufferSize, " block_size_deviation: %d\n",
table_options_.block_size_deviation);
ret.append(buffer);
snprintf(buffer, kBufferSize, " block_restart_interval: %d\n",
table_options_.block_restart_interval);
ret.append(buffer);
snprintf(buffer, kBufferSize, " index_block_restart_interval: %d\n",
table_options_.index_block_restart_interval);
ret.append(buffer);
snprintf(buffer, kBufferSize, " metadata_block_size: %" PRIu64 "\n",
table_options_.metadata_block_size);
ret.append(buffer);
snprintf(buffer, kBufferSize, " partition_filters: %d\n",
table_options_.partition_filters);
ret.append(buffer);
snprintf(buffer, kBufferSize, " use_delta_encoding: %d\n",
table_options_.use_delta_encoding);
ret.append(buffer);
snprintf(buffer, kBufferSize, " filter_policy: %s\n",
table_options_.filter_policy == nullptr
? "nullptr"
: table_options_.filter_policy->Name());
ret.append(buffer);
snprintf(buffer, kBufferSize, " whole_key_filtering: %d\n",
table_options_.whole_key_filtering);
ret.append(buffer);
snprintf(buffer, kBufferSize, " verify_compression: %d\n",
table_options_.verify_compression);
ret.append(buffer);
snprintf(buffer, kBufferSize, " read_amp_bytes_per_bit: %d\n",
table_options_.read_amp_bytes_per_bit);
ret.append(buffer);
snprintf(buffer, kBufferSize, " format_version: %d\n",
table_options_.format_version);
ret.append(buffer);
snprintf(buffer, kBufferSize, " enable_index_compression: %d\n",
table_options_.enable_index_compression);
ret.append(buffer);
snprintf(buffer, kBufferSize, " block_align: %d\n",
table_options_.block_align);
ret.append(buffer);
return ret;
}
#ifndef ROCKSDB_LITE
Status BlockBasedTableFactory::GetOptionString(
const ConfigOptions& config_options, std::string* opt_string) const {
assert(opt_string);
opt_string->clear();
return GetStringFromStruct(config_options, &table_options_,
block_based_table_type_info, opt_string);
}
#else
Status BlockBasedTableFactory::GetOptionString(
const ConfigOptions& /*opts*/, std::string* /*opt_string*/) const {
return Status::OK();
}
#endif // !ROCKSDB_LITE
2015-10-30 23:58:46 +01:00
const BlockBasedTableOptions& BlockBasedTableFactory::table_options() const {
return table_options_;
}
#ifndef ROCKSDB_LITE
namespace {
std::string ParseBlockBasedTableOption(const ConfigOptions& config_options,
const std::string& name,
const std::string& org_value,
BlockBasedTableOptions* new_options) {
const std::string& value = config_options.input_strings_escaped
? UnescapeOptionString(org_value)
: org_value;
const auto iter = block_based_table_type_info.find(name);
if (iter == block_based_table_type_info.end()) {
if (config_options.ignore_unknown_options) {
return "";
} else {
return "Unrecognized option";
}
}
const auto& opt_info = iter->second;
Status s =
opt_info.Parse(config_options, iter->first, value,
reinterpret_cast<char*>(new_options) + opt_info.offset_);
if (s.ok()) {
return "";
} else {
return s.ToString();
}
}
} // namespace
Status GetBlockBasedTableOptionsFromString(
const BlockBasedTableOptions& table_options, const std::string& opts_str,
BlockBasedTableOptions* new_table_options) {
ConfigOptions config_options;
config_options.input_strings_escaped = false;
config_options.ignore_unknown_options = false;
return GetBlockBasedTableOptionsFromString(config_options, table_options,
opts_str, new_table_options);
}
Status GetBlockBasedTableOptionsFromString(
const ConfigOptions& config_options,
const BlockBasedTableOptions& table_options, const std::string& opts_str,
BlockBasedTableOptions* new_table_options) {
std::unordered_map<std::string, std::string> opts_map;
Status s = StringToMap(opts_str, &opts_map);
if (!s.ok()) {
return s;
}
return GetBlockBasedTableOptionsFromMap(config_options, table_options,
opts_map, new_table_options);
}
Status GetBlockBasedTableOptionsFromMap(
const BlockBasedTableOptions& table_options,
const std::unordered_map<std::string, std::string>& opts_map,
BlockBasedTableOptions* new_table_options, bool input_strings_escaped,
bool ignore_unknown_options) {
ConfigOptions config_options;
config_options.input_strings_escaped = input_strings_escaped;
config_options.ignore_unknown_options = ignore_unknown_options;
return GetBlockBasedTableOptionsFromMap(config_options, table_options,
opts_map, new_table_options);
}
Status GetBlockBasedTableOptionsFromMap(
const ConfigOptions& config_options,
const BlockBasedTableOptions& table_options,
const std::unordered_map<std::string, std::string>& opts_map,
BlockBasedTableOptions* new_table_options) {
assert(new_table_options);
*new_table_options = table_options;
for (const auto& o : opts_map) {
auto error_message = ParseBlockBasedTableOption(
config_options, o.first, o.second, new_table_options);
if (error_message != "") {
const auto iter = block_based_table_type_info.find(o.first);
if (iter == block_based_table_type_info.end() ||
!config_options
.input_strings_escaped || // !input_strings_escaped indicates
// the old API, where everything is
// parsable.
(!iter->second.IsByName() && !iter->second.IsDeprecated())) {
// Restore "new_options" to the default "base_options".
*new_table_options = table_options;
return Status::InvalidArgument("Can't parse BlockBasedTableOptions:",
o.first + " " + error_message);
}
}
}
return Status::OK();
}
Status VerifyBlockBasedTableFactory(const ConfigOptions& config_options,
const BlockBasedTableFactory* base_tf,
const BlockBasedTableFactory* file_tf) {
if ((base_tf != nullptr) != (file_tf != nullptr) &&
config_options.sanity_level > ConfigOptions::kSanityLevelNone) {
return Status::Corruption(
"[RocksDBOptionsParser]: Inconsistent TableFactory class type");
}
if (base_tf == nullptr) {
return Status::OK();
}
assert(file_tf != nullptr);
const auto& base_opt = base_tf->table_options();
const auto& file_opt = file_tf->table_options();
std::string mismatch;
for (auto& pair : block_based_table_type_info) {
// We skip checking deprecated variables as they might
// contain random values since they might not be initialized
if (config_options.IsCheckEnabled(pair.second.GetSanityLevel())) {
const char* base_addr =
reinterpret_cast<const char*>(&base_opt) + pair.second.offset_;
const char* file_addr =
reinterpret_cast<const char*>(&file_opt) + pair.second.offset_;
if (!pair.second.AreEqual(config_options, pair.first, base_addr,
file_addr, &mismatch) &&
!pair.second.AreEqualByName(config_options, pair.first, base_addr,
file_addr)) {
return Status::Corruption(
"[RocksDBOptionsParser]: "
"failed the verification on BlockBasedTableOptions::",
pair.first);
}
}
}
return Status::OK();
}
#endif // !ROCKSDB_LITE
TableFactory* NewBlockBasedTableFactory(
const BlockBasedTableOptions& _table_options) {
return new BlockBasedTableFactory(_table_options);
}
const std::string BlockBasedTableFactory::kName = "BlockBasedTable";
const std::string BlockBasedTablePropertyNames::kIndexType =
"rocksdb.block.based.table.index.type";
const std::string BlockBasedTablePropertyNames::kWholeKeyFiltering =
"rocksdb.block.based.table.whole.key.filtering";
const std::string BlockBasedTablePropertyNames::kPrefixFiltering =
"rocksdb.block.based.table.prefix.filtering";
const std::string kHashIndexPrefixesBlock = "rocksdb.hashindex.prefixes";
const std::string kHashIndexPrefixesMetadataBlock =
"rocksdb.hashindex.metadata";
const std::string kPropTrue = "1";
const std::string kPropFalse = "0";
} // namespace ROCKSDB_NAMESPACE