rocksdb/db/db_block_cache_test.cc

1697 lines
64 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 <cstdlib>
#include <functional>
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
#include <memory>
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
#include "cache/cache_entry_roles.h"
#include "cache/lru_cache.h"
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
#include "db/column_family.h"
#include "db/db_test_util.h"
#include "port/stack_trace.h"
#include "rocksdb/statistics.h"
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
#include "rocksdb/table.h"
#include "util/compression.h"
#include "util/defer.h"
#include "util/random.h"
#include "utilities/fault_injection_fs.h"
namespace ROCKSDB_NAMESPACE {
class DBBlockCacheTest : public DBTestBase {
private:
size_t miss_count_ = 0;
size_t hit_count_ = 0;
size_t insert_count_ = 0;
size_t failure_count_ = 0;
size_t compression_dict_miss_count_ = 0;
size_t compression_dict_hit_count_ = 0;
size_t compression_dict_insert_count_ = 0;
size_t compressed_miss_count_ = 0;
size_t compressed_hit_count_ = 0;
size_t compressed_insert_count_ = 0;
size_t compressed_failure_count_ = 0;
public:
const size_t kNumBlocks = 10;
const size_t kValueSize = 100;
DBBlockCacheTest()
: DBTestBase("db_block_cache_test", /*env_do_fsync=*/true) {}
BlockBasedTableOptions GetTableOptions() {
BlockBasedTableOptions table_options;
// Set a small enough block size so that each key-value get its own block.
table_options.block_size = 1;
return table_options;
}
Options GetOptions(const BlockBasedTableOptions& table_options) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.avoid_flush_during_recovery = false;
// options.compression = kNoCompression;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
return options;
}
void InitTable(const Options& /*options*/) {
std::string value(kValueSize, 'a');
for (size_t i = 0; i < kNumBlocks; i++) {
ASSERT_OK(Put(ToString(i), value.c_str()));
}
}
void RecordCacheCounters(const Options& options) {
miss_count_ = TestGetTickerCount(options, BLOCK_CACHE_MISS);
hit_count_ = TestGetTickerCount(options, BLOCK_CACHE_HIT);
insert_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD);
failure_count_ = TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
compressed_miss_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
compressed_hit_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
compressed_insert_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
compressed_failure_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
}
void RecordCacheCountersForCompressionDict(const Options& options) {
compression_dict_miss_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS);
compression_dict_hit_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_HIT);
compression_dict_insert_count_ =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD);
}
void CheckCacheCounters(const Options& options, size_t expected_misses,
size_t expected_hits, size_t expected_inserts,
size_t expected_failures) {
size_t new_miss_count = TestGetTickerCount(options, BLOCK_CACHE_MISS);
size_t new_hit_count = TestGetTickerCount(options, BLOCK_CACHE_HIT);
size_t new_insert_count = TestGetTickerCount(options, BLOCK_CACHE_ADD);
size_t new_failure_count =
TestGetTickerCount(options, BLOCK_CACHE_ADD_FAILURES);
ASSERT_EQ(miss_count_ + expected_misses, new_miss_count);
ASSERT_EQ(hit_count_ + expected_hits, new_hit_count);
ASSERT_EQ(insert_count_ + expected_inserts, new_insert_count);
ASSERT_EQ(failure_count_ + expected_failures, new_failure_count);
miss_count_ = new_miss_count;
hit_count_ = new_hit_count;
insert_count_ = new_insert_count;
failure_count_ = new_failure_count;
}
void CheckCacheCountersForCompressionDict(
const Options& options, size_t expected_compression_dict_misses,
size_t expected_compression_dict_hits,
size_t expected_compression_dict_inserts) {
size_t new_compression_dict_miss_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS);
size_t new_compression_dict_hit_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_HIT);
size_t new_compression_dict_insert_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_ADD);
ASSERT_EQ(compression_dict_miss_count_ + expected_compression_dict_misses,
new_compression_dict_miss_count);
ASSERT_EQ(compression_dict_hit_count_ + expected_compression_dict_hits,
new_compression_dict_hit_count);
ASSERT_EQ(
compression_dict_insert_count_ + expected_compression_dict_inserts,
new_compression_dict_insert_count);
compression_dict_miss_count_ = new_compression_dict_miss_count;
compression_dict_hit_count_ = new_compression_dict_hit_count;
compression_dict_insert_count_ = new_compression_dict_insert_count;
}
void CheckCompressedCacheCounters(const Options& options,
size_t expected_misses,
size_t expected_hits,
size_t expected_inserts,
size_t expected_failures) {
size_t new_miss_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS);
size_t new_hit_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT);
size_t new_insert_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD);
size_t new_failure_count =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_ADD_FAILURES);
ASSERT_EQ(compressed_miss_count_ + expected_misses, new_miss_count);
ASSERT_EQ(compressed_hit_count_ + expected_hits, new_hit_count);
ASSERT_EQ(compressed_insert_count_ + expected_inserts, new_insert_count);
ASSERT_EQ(compressed_failure_count_ + expected_failures, new_failure_count);
compressed_miss_count_ = new_miss_count;
compressed_hit_count_ = new_hit_count;
compressed_insert_count_ = new_insert_count;
compressed_failure_count_ = new_failure_count;
}
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
#ifndef ROCKSDB_LITE
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
const std::array<size_t, kNumCacheEntryRoles> GetCacheEntryRoleCountsBg() {
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
// Verify in cache entry role stats
ColumnFamilyHandleImpl* cfh =
static_cast<ColumnFamilyHandleImpl*>(dbfull()->DefaultColumnFamily());
InternalStats* internal_stats_ptr = cfh->cfd()->internal_stats();
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
InternalStats::CacheEntryRoleStats stats;
internal_stats_ptr->TEST_GetCacheEntryRoleStats(&stats,
/*foreground=*/false);
return stats.entry_counts;
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
}
#endif // ROCKSDB_LITE
};
TEST_F(DBBlockCacheTest, IteratorBlockCacheUsage) {
ReadOptions read_options;
read_options.fill_cache = false;
auto table_options = GetTableOptions();
auto options = GetOptions(table_options);
InitTable(options);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
LRUCacheOptions co;
co.capacity = 0;
co.num_shard_bits = 0;
co.strict_capacity_limit = false;
// Needed not to count entry stats collector
co.metadata_charge_policy = kDontChargeCacheMetadata;
std::shared_ptr<Cache> cache = NewLRUCache(co);
table_options.block_cache = cache;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
RecordCacheCounters(options);
std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
Iterator* iter = nullptr;
ASSERT_EQ(0, cache->GetUsage());
iter = db_->NewIterator(read_options);
iter->Seek(ToString(0));
ASSERT_LT(0, cache->GetUsage());
delete iter;
iter = nullptr;
ASSERT_EQ(0, cache->GetUsage());
}
TEST_F(DBBlockCacheTest, TestWithoutCompressedBlockCache) {
ReadOptions read_options;
auto table_options = GetTableOptions();
auto options = GetOptions(table_options);
InitTable(options);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
LRUCacheOptions co;
co.capacity = 0;
co.num_shard_bits = 0;
co.strict_capacity_limit = false;
// Needed not to count entry stats collector
co.metadata_charge_policy = kDontChargeCacheMetadata;
std::shared_ptr<Cache> cache = NewLRUCache(co);
table_options.block_cache = cache;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
RecordCacheCounters(options);
std::vector<std::unique_ptr<Iterator>> iterators(kNumBlocks - 1);
Iterator* iter = nullptr;
// Load blocks into cache.
for (size_t i = 0; i + 1 < kNumBlocks; i++) {
iter = db_->NewIterator(read_options);
iter->Seek(ToString(i));
ASSERT_OK(iter->status());
CheckCacheCounters(options, 1, 0, 1, 0);
iterators[i].reset(iter);
}
size_t usage = cache->GetUsage();
ASSERT_LT(0, usage);
cache->SetCapacity(usage);
ASSERT_EQ(usage, cache->GetPinnedUsage());
// Test with strict capacity limit.
cache->SetStrictCapacityLimit(true);
iter = db_->NewIterator(read_options);
iter->Seek(ToString(kNumBlocks - 1));
ASSERT_TRUE(iter->status().IsIncomplete());
CheckCacheCounters(options, 1, 0, 0, 1);
delete iter;
iter = nullptr;
// Release iterators and access cache again.
for (size_t i = 0; i + 1 < kNumBlocks; i++) {
iterators[i].reset();
CheckCacheCounters(options, 0, 0, 0, 0);
}
ASSERT_EQ(0, cache->GetPinnedUsage());
for (size_t i = 0; i + 1 < kNumBlocks; i++) {
iter = db_->NewIterator(read_options);
iter->Seek(ToString(i));
ASSERT_OK(iter->status());
CheckCacheCounters(options, 0, 1, 0, 0);
iterators[i].reset(iter);
}
}
#ifdef SNAPPY
TEST_F(DBBlockCacheTest, TestWithCompressedBlockCache) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.no_block_cache = true;
table_options.block_cache_compressed = nullptr;
table_options.block_size = 1;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
table_options.cache_index_and_filter_blocks = false;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.compression = CompressionType::kSnappyCompression;
DestroyAndReopen(options);
std::string value(kValueSize, 'a');
for (size_t i = 0; i < kNumBlocks; i++) {
ASSERT_OK(Put(ToString(i), value));
ASSERT_OK(Flush());
}
ReadOptions read_options;
std::shared_ptr<Cache> compressed_cache = NewLRUCache(1 << 25, 0, false);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
LRUCacheOptions co;
co.capacity = 0;
co.num_shard_bits = 0;
co.strict_capacity_limit = false;
// Needed not to count entry stats collector
co.metadata_charge_policy = kDontChargeCacheMetadata;
std::shared_ptr<Cache> cache = NewLRUCache(co);
table_options.block_cache = cache;
table_options.no_block_cache = false;
table_options.block_cache_compressed = compressed_cache;
table_options.max_auto_readahead_size = 0;
table_options.cache_index_and_filter_blocks = false;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
RecordCacheCounters(options);
// Load blocks into cache.
for (size_t i = 0; i < kNumBlocks - 1; i++) {
ASSERT_EQ(value, Get(ToString(i)));
CheckCacheCounters(options, 1, 0, 1, 0);
CheckCompressedCacheCounters(options, 1, 0, 1, 0);
}
size_t usage = cache->GetUsage();
ASSERT_EQ(0, usage);
ASSERT_EQ(usage, cache->GetPinnedUsage());
size_t compressed_usage = compressed_cache->GetUsage();
ASSERT_LT(0, compressed_usage);
// Compressed block cache cannot be pinned.
ASSERT_EQ(0, compressed_cache->GetPinnedUsage());
// Set strict capacity limit flag. Now block will only load into compressed
// block cache.
cache->SetCapacity(usage);
cache->SetStrictCapacityLimit(true);
ASSERT_EQ(usage, cache->GetPinnedUsage());
// Load last key block.
ASSERT_EQ("Result incomplete: Insert failed due to LRU cache being full.",
Get(ToString(kNumBlocks - 1)));
// Failure will also record the miss counter.
CheckCacheCounters(options, 1, 0, 0, 1);
CheckCompressedCacheCounters(options, 1, 0, 1, 0);
// Clear strict capacity limit flag. This time we shall hit compressed block
// cache and load into block cache.
cache->SetStrictCapacityLimit(false);
// Load last key block.
ASSERT_EQ(value, Get(ToString(kNumBlocks - 1)));
CheckCacheCounters(options, 1, 0, 1, 0);
CheckCompressedCacheCounters(options, 0, 1, 0, 0);
}
#endif // SNAPPY
#ifndef ROCKSDB_LITE
// Make sure that when options.block_cache is set, after a new table is
// created its index/filter blocks are added to block cache.
TEST_F(DBBlockCacheTest, IndexAndFilterBlocksOfNewTableAddedToCache) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "key", "val"));
// Create a new table.
ASSERT_OK(Flush(1));
// index/filter blocks added to block cache right after table creation.
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(2, /* only index/filter were added */
TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
uint64_t int_num;
ASSERT_TRUE(
dbfull()->GetIntProperty("rocksdb.estimate-table-readers-mem", &int_num));
ASSERT_EQ(int_num, 0U);
// Make sure filter block is in cache.
std::string value;
ReadOptions ropt;
db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
// Miss count should remain the same.
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value);
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
// Make sure index block is in cache.
auto index_block_hit = TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT);
value = Get(1, "key");
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(index_block_hit + 1,
TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
value = Get(1, "key");
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(index_block_hit + 2,
TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
// With fill_cache = false, fills up the cache, then iterates over the entire
// db, verify dummy entries inserted in `BlockBasedTable::NewDataBlockIterator`
// does not cause heap-use-after-free errors in COMPILE_WITH_ASAN=1 runs
TEST_F(DBBlockCacheTest, FillCacheAndIterateDB) {
ReadOptions read_options;
read_options.fill_cache = false;
auto table_options = GetTableOptions();
auto options = GetOptions(table_options);
InitTable(options);
std::shared_ptr<Cache> cache = NewLRUCache(10, 0, true);
table_options.block_cache = cache;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
ASSERT_OK(Put("key1", "val1"));
ASSERT_OK(Put("key2", "val2"));
ASSERT_OK(Flush());
ASSERT_OK(Put("key3", "val3"));
ASSERT_OK(Put("key4", "val4"));
ASSERT_OK(Flush());
ASSERT_OK(Put("key5", "val5"));
ASSERT_OK(Put("key6", "val6"));
ASSERT_OK(Flush());
Iterator* iter = nullptr;
iter = db_->NewIterator(read_options);
iter->Seek(ToString(0));
while (iter->Valid()) {
iter->Next();
}
delete iter;
iter = nullptr;
}
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
TEST_F(DBBlockCacheTest, IndexAndFilterBlocksStats) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
LRUCacheOptions co;
// 500 bytes are enough to hold the first two blocks
co.capacity = 500;
co.num_shard_bits = 0;
co.strict_capacity_limit = false;
co.metadata_charge_policy = kDontChargeCacheMetadata;
std::shared_ptr<Cache> cache = NewLRUCache(co);
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
table_options.block_cache = cache;
table_options.filter_policy.reset(NewBloomFilterPolicy(20, true));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "longer_key", "val"));
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
// Create a new table
ASSERT_OK(Flush(1));
size_t index_bytes_insert =
TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT);
size_t filter_bytes_insert =
TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT);
ASSERT_GT(index_bytes_insert, 0);
ASSERT_GT(filter_bytes_insert, 0);
ASSERT_EQ(cache->GetUsage(), index_bytes_insert + filter_bytes_insert);
// set the cache capacity to the current usage
cache->SetCapacity(index_bytes_insert + filter_bytes_insert);
// The index and filter eviction statistics were broken by the refactoring
// that moved the readers out of the block cache. Disabling these until we can
// bring the stats back.
// ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT), 0);
// ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT), 0);
// Note that the second key needs to be no longer than the first one.
// Otherwise the second index block may not fit in cache.
ASSERT_OK(Put(1, "key", "val"));
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
// Create a new table
ASSERT_OK(Flush(1));
// cache evicted old index and block entries
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_INSERT),
index_bytes_insert);
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_INSERT),
filter_bytes_insert);
// The index and filter eviction statistics were broken by the refactoring
// that moved the readers out of the block cache. Disabling these until we can
// bring the stats back.
// ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_INDEX_BYTES_EVICT),
// index_bytes_insert);
// ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_FILTER_BYTES_EVICT),
// filter_bytes_insert);
Add statistics field to show total size of index and filter blocks in block cache Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes. Test Plan: write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is: ``` [gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production GEN util/build_version.cc make: `db_block_cache_test' is up to date. Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBBlockCacheTest [ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats [ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms) [----------] 1 test from DBBlockCacheTest (689 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (689 ms total) [ PASSED ] 1 test. ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D58677
2016-06-03 19:47:47 +02:00
}
#if (defined OS_LINUX || defined OS_WIN)
TEST_F(DBBlockCacheTest, WarmCacheWithDataBlocksDuringFlush) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.block_cache = NewLRUCache(1 << 25, 0, false);
table_options.cache_index_and_filter_blocks = false;
table_options.prepopulate_block_cache =
BlockBasedTableOptions::PrepopulateBlockCache::kFlushOnly;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
std::string value(kValueSize, 'a');
for (size_t i = 1; i <= kNumBlocks; i++) {
ASSERT_OK(Put(ToString(i), value));
ASSERT_OK(Flush());
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(value, Get(ToString(i)));
ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_DATA_MISS));
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_HIT));
}
}
// This test cache data, index and filter blocks during flush.
TEST_F(DBBlockCacheTest, WarmCacheWithBlocksDuringFlush) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.block_cache = NewLRUCache(1 << 25, 0, false);
table_options.cache_index_and_filter_blocks = true;
table_options.prepopulate_block_cache =
BlockBasedTableOptions::PrepopulateBlockCache::kFlushOnly;
table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
std::string value(kValueSize, 'a');
for (size_t i = 1; i <= kNumBlocks; i++) {
ASSERT_OK(Put(ToString(i), value));
ASSERT_OK(Flush());
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_ADD));
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_FILTER_ADD));
ASSERT_EQ(value, Get(ToString(i)));
ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_DATA_MISS));
ASSERT_EQ(i, options.statistics->getTickerCount(BLOCK_CACHE_DATA_HIT));
ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(i * 3, options.statistics->getTickerCount(BLOCK_CACHE_INDEX_HIT));
ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(i * 2,
options.statistics->getTickerCount(BLOCK_CACHE_FILTER_HIT));
}
}
TEST_F(DBBlockCacheTest, DynamicallyWarmCacheDuringFlush) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.block_cache = NewLRUCache(1 << 25, 0, false);
table_options.cache_index_and_filter_blocks = false;
table_options.prepopulate_block_cache =
BlockBasedTableOptions::PrepopulateBlockCache::kFlushOnly;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
std::string value(kValueSize, 'a');
for (size_t i = 1; i <= 5; i++) {
ASSERT_OK(Put(ToString(i), value));
ASSERT_OK(Flush());
ASSERT_EQ(1,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(value, Get(ToString(i)));
ASSERT_EQ(0,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(
0, options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_MISS));
ASSERT_EQ(1,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_HIT));
}
ASSERT_OK(dbfull()->SetOptions(
{{"block_based_table_factory", "{prepopulate_block_cache=kDisable;}"}}));
for (size_t i = 6; i <= kNumBlocks; i++) {
ASSERT_OK(Put(ToString(i), value));
ASSERT_OK(Flush());
ASSERT_EQ(0,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(value, Get(ToString(i)));
ASSERT_EQ(1,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(
1, options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_MISS));
ASSERT_EQ(0,
options.statistics->getAndResetTickerCount(BLOCK_CACHE_DATA_HIT));
}
}
#endif
namespace {
// A mock cache wraps LRUCache, and record how many entries have been
// inserted for each priority.
class MockCache : public LRUCache {
public:
static uint32_t high_pri_insert_count;
static uint32_t low_pri_insert_count;
MockCache()
: LRUCache((size_t)1 << 25 /*capacity*/, 0 /*num_shard_bits*/,
false /*strict_capacity_limit*/, 0.0 /*high_pri_pool_ratio*/) {
}
Initial support for secondary cache in LRUCache (#8271) Summary: Defined the abstract interface for a secondary cache in include/rocksdb/secondary_cache.h, and updated LRUCacheOptions to take a std::shared_ptr<SecondaryCache>. An item is initially inserted into the LRU (primary) cache. When it ages out and evicted from memory, its inserted into the secondary cache. On a LRU cache miss and successful lookup in the secondary cache, the item is promoted to the LRU cache. Only support synchronous lookup currently. The secondary cache would be used to implement a persistent (flash cache) or compressed cache. Tests: Results from cache_bench and db_bench don't show any regression due to these changes. cache_bench results before and after this change - Command ```./cache_bench -ops_per_thread=10000000 -threads=1``` Before ```Complete in 40.688 s; QPS = 245774``` ```Complete in 40.486 s; QPS = 246996``` ```Complete in 42.019 s; QPS = 237989``` After ```Complete in 40.672 s; QPS = 245869``` ```Complete in 44.622 s; QPS = 224107``` ```Complete in 42.445 s; QPS = 235599``` db_bench results before this change, and with this change + https://github.com/facebook/rocksdb/issues/8213 and https://github.com/facebook/rocksdb/issues/8191 - Commands ```./db_bench --benchmarks="fillseq,compact" -num=30000000 -key_size=32 -value_size=256 -use_direct_io_for_flush_and_compaction=true -db=/home/anand76/nvm_cache/db -partition_index_and_filters=true``` ```./db_bench -db=/home/anand76/nvm_cache/db -use_existing_db=true -benchmarks=readrandom -num=30000000 -key_size=32 -value_size=256 -use_direct_reads=true -cache_size=1073741824 -cache_numshardbits=6 -cache_index_and_filter_blocks=true -read_random_exp_range=17 -statistics -partition_index_and_filters=true -threads=16 -duration=300``` Before ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 80.702 micros/op 198104 ops/sec; 54.4 MB/s (3708999 of 3708999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 87.124 micros/op 183625 ops/sec; 50.4 MB/s (3439999 of 3439999 found) ``` After ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 77.653 micros/op 206025 ops/sec; 56.6 MB/s (3866999 of 3866999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 84.962 micros/op 188299 ops/sec; 51.7 MB/s (3535999 of 3535999 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8271 Reviewed By: zhichao-cao Differential Revision: D28357511 Pulled By: anand1976 fbshipit-source-id: d1cfa236f00e649a18c53328be10a8062a4b6da2
2021-05-14 07:57:51 +02:00
using ShardedCache::Insert;
Use new Insert and Lookup APIs in table reader to support secondary cache (#8315) Summary: Secondary cache is implemented to achieve the secondary cache tier for block cache. New Insert and Lookup APIs are introduced in https://github.com/facebook/rocksdb/issues/8271 . To support and use the secondary cache in block based table reader, this PR introduces the corresponding callback functions that will be used in secondary cache, and update the Insert and Lookup APIs accordingly. benchmarking: ./db_bench --benchmarks="fillrandom" -num=1000000 -key_size=32 -value_size=256 -use_direct_io_for_flush_and_compaction=true -db=/tmp/rocks_t/db -partition_index_and_filters=true ./db_bench -db=/tmp/rocks_t/db -use_existing_db=true -benchmarks=readrandom -num=1000000 -key_size=32 -value_size=256 -use_direct_reads=true -cache_size=1073741824 -cache_numshardbits=5 -cache_index_and_filter_blocks=true -read_random_exp_range=17 -statistics -partition_index_and_filters=true -stats_dump_period_sec=30 -reads=50000000 master benchmarking results: readrandom : 3.923 micros/op 254881 ops/sec; 33.4 MB/s (23849796 of 50000000 found) rocksdb.db.get.micros P50 : 2.820992 P95 : 5.636716 P99 : 16.450553 P100 : 8396.000000 COUNT : 50000000 SUM : 179947064 Current PR benchmarking results readrandom : 4.083 micros/op 244925 ops/sec; 32.1 MB/s (23849796 of 50000000 found) rocksdb.db.get.micros P50 : 2.967687 P95 : 5.754916 P99 : 15.665912 P100 : 8213.000000 COUNT : 50000000 SUM : 187250053 About 3.8% throughput reduction. P50: 5.2% increasing, P95, 2.09% increasing, P99 4.77% improvement Pull Request resolved: https://github.com/facebook/rocksdb/pull/8315 Test Plan: added the testing case Reviewed By: anand1976 Differential Revision: D28599774 Pulled By: zhichao-cao fbshipit-source-id: 098c4df0d7327d3a546df7604b2f1602f13044ed
2021-05-22 03:28:28 +02:00
Status Insert(const Slice& key, void* value,
const Cache::CacheItemHelper* helper_cb, size_t charge,
Handle** handle, Priority priority) override {
DeleterFn delete_cb = helper_cb->del_cb;
if (priority == Priority::LOW) {
low_pri_insert_count++;
} else {
high_pri_insert_count++;
}
Use new Insert and Lookup APIs in table reader to support secondary cache (#8315) Summary: Secondary cache is implemented to achieve the secondary cache tier for block cache. New Insert and Lookup APIs are introduced in https://github.com/facebook/rocksdb/issues/8271 . To support and use the secondary cache in block based table reader, this PR introduces the corresponding callback functions that will be used in secondary cache, and update the Insert and Lookup APIs accordingly. benchmarking: ./db_bench --benchmarks="fillrandom" -num=1000000 -key_size=32 -value_size=256 -use_direct_io_for_flush_and_compaction=true -db=/tmp/rocks_t/db -partition_index_and_filters=true ./db_bench -db=/tmp/rocks_t/db -use_existing_db=true -benchmarks=readrandom -num=1000000 -key_size=32 -value_size=256 -use_direct_reads=true -cache_size=1073741824 -cache_numshardbits=5 -cache_index_and_filter_blocks=true -read_random_exp_range=17 -statistics -partition_index_and_filters=true -stats_dump_period_sec=30 -reads=50000000 master benchmarking results: readrandom : 3.923 micros/op 254881 ops/sec; 33.4 MB/s (23849796 of 50000000 found) rocksdb.db.get.micros P50 : 2.820992 P95 : 5.636716 P99 : 16.450553 P100 : 8396.000000 COUNT : 50000000 SUM : 179947064 Current PR benchmarking results readrandom : 4.083 micros/op 244925 ops/sec; 32.1 MB/s (23849796 of 50000000 found) rocksdb.db.get.micros P50 : 2.967687 P95 : 5.754916 P99 : 15.665912 P100 : 8213.000000 COUNT : 50000000 SUM : 187250053 About 3.8% throughput reduction. P50: 5.2% increasing, P95, 2.09% increasing, P99 4.77% improvement Pull Request resolved: https://github.com/facebook/rocksdb/pull/8315 Test Plan: added the testing case Reviewed By: anand1976 Differential Revision: D28599774 Pulled By: zhichao-cao fbshipit-source-id: 098c4df0d7327d3a546df7604b2f1602f13044ed
2021-05-22 03:28:28 +02:00
return LRUCache::Insert(key, value, charge, delete_cb, handle, priority);
}
};
uint32_t MockCache::high_pri_insert_count = 0;
uint32_t MockCache::low_pri_insert_count = 0;
} // anonymous namespace
TEST_F(DBBlockCacheTest, IndexAndFilterBlocksCachePriority) {
for (auto priority : {Cache::Priority::LOW, Cache::Priority::HIGH}) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.block_cache.reset(new MockCache());
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
table_options.cache_index_and_filter_blocks_with_high_priority =
priority == Cache::Priority::HIGH ? true : false;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
MockCache::high_pri_insert_count = 0;
MockCache::low_pri_insert_count = 0;
// Create a new table.
ASSERT_OK(Put("foo", "value"));
ASSERT_OK(Put("bar", "value"));
ASSERT_OK(Flush());
ASSERT_EQ(1, NumTableFilesAtLevel(0));
// index/filter blocks added to block cache right after table creation.
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(2, /* only index/filter were added */
TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
if (priority == Cache::Priority::LOW) {
ASSERT_EQ(0u, MockCache::high_pri_insert_count);
ASSERT_EQ(2u, MockCache::low_pri_insert_count);
} else {
ASSERT_EQ(2u, MockCache::high_pri_insert_count);
ASSERT_EQ(0u, MockCache::low_pri_insert_count);
}
// Access data block.
ASSERT_EQ("value", Get("foo"));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(3, /*adding data block*/
TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
// Data block should be inserted with low priority.
if (priority == Cache::Priority::LOW) {
ASSERT_EQ(0u, MockCache::high_pri_insert_count);
ASSERT_EQ(3u, MockCache::low_pri_insert_count);
} else {
ASSERT_EQ(2u, MockCache::high_pri_insert_count);
ASSERT_EQ(1u, MockCache::low_pri_insert_count);
}
}
}
Stats for redundant insertions into block cache (#6681) Summary: Since read threads do not coordinate on loading data into block cache, two threads between Lookup and Insert can end up loading and inserting the same data. This is particularly concerning with cache_index_and_filter_blocks since those are hot and more likely to be race targets if ejected from (or not pre-populated in) the cache. Particularly with moves toward disaggregated / network storage, the cost of redundant retrieval might be high, and we should at least have some hard statistics from which we can estimate impact. Example with full filter thrashing "cliff": $ ./db_bench --benchmarks=fillrandom --num=15000000 --cache_index_and_filter_blocks -bloom_bits=10 ... $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((130 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 14181 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 476 rocksdb.block.cache.data.add COUNT : 12749 rocksdb.block.cache.data.add.redundant COUNT : 18 rocksdb.block.cache.filter.add COUNT : 1003 rocksdb.block.cache.filter.add.redundant COUNT : 217 rocksdb.block.cache.index.add COUNT : 429 rocksdb.block.cache.index.add.redundant COUNT : 241 $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((120 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 1182223 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 302728 rocksdb.block.cache.data.add COUNT : 31425 rocksdb.block.cache.data.add.redundant COUNT : 12 rocksdb.block.cache.filter.add COUNT : 795455 rocksdb.block.cache.filter.add.redundant COUNT : 130238 rocksdb.block.cache.index.add COUNT : 355343 rocksdb.block.cache.index.add.redundant COUNT : 172478 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6681 Test Plan: Some manual testing (above) and unit test covering key metrics is included Reviewed By: ltamasi Differential Revision: D21134113 Pulled By: pdillinger fbshipit-source-id: c11497b5f00f4ffdfe919823904e52d0a1a91d87
2020-04-27 22:18:18 +02:00
namespace {
// An LRUCache wrapper that can falsely report "not found" on Lookup.
// This allows us to manipulate BlockBasedTableReader into thinking
// another thread inserted the data in between Lookup and Insert,
// while mostly preserving the LRUCache interface/behavior.
class LookupLiarCache : public CacheWrapper {
int nth_lookup_not_found_ = 0;
public:
explicit LookupLiarCache(std::shared_ptr<Cache> target)
: CacheWrapper(std::move(target)) {}
Initial support for secondary cache in LRUCache (#8271) Summary: Defined the abstract interface for a secondary cache in include/rocksdb/secondary_cache.h, and updated LRUCacheOptions to take a std::shared_ptr<SecondaryCache>. An item is initially inserted into the LRU (primary) cache. When it ages out and evicted from memory, its inserted into the secondary cache. On a LRU cache miss and successful lookup in the secondary cache, the item is promoted to the LRU cache. Only support synchronous lookup currently. The secondary cache would be used to implement a persistent (flash cache) or compressed cache. Tests: Results from cache_bench and db_bench don't show any regression due to these changes. cache_bench results before and after this change - Command ```./cache_bench -ops_per_thread=10000000 -threads=1``` Before ```Complete in 40.688 s; QPS = 245774``` ```Complete in 40.486 s; QPS = 246996``` ```Complete in 42.019 s; QPS = 237989``` After ```Complete in 40.672 s; QPS = 245869``` ```Complete in 44.622 s; QPS = 224107``` ```Complete in 42.445 s; QPS = 235599``` db_bench results before this change, and with this change + https://github.com/facebook/rocksdb/issues/8213 and https://github.com/facebook/rocksdb/issues/8191 - Commands ```./db_bench --benchmarks="fillseq,compact" -num=30000000 -key_size=32 -value_size=256 -use_direct_io_for_flush_and_compaction=true -db=/home/anand76/nvm_cache/db -partition_index_and_filters=true``` ```./db_bench -db=/home/anand76/nvm_cache/db -use_existing_db=true -benchmarks=readrandom -num=30000000 -key_size=32 -value_size=256 -use_direct_reads=true -cache_size=1073741824 -cache_numshardbits=6 -cache_index_and_filter_blocks=true -read_random_exp_range=17 -statistics -partition_index_and_filters=true -threads=16 -duration=300``` Before ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 80.702 micros/op 198104 ops/sec; 54.4 MB/s (3708999 of 3708999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 87.124 micros/op 183625 ops/sec; 50.4 MB/s (3439999 of 3439999 found) ``` After ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 77.653 micros/op 206025 ops/sec; 56.6 MB/s (3866999 of 3866999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 84.962 micros/op 188299 ops/sec; 51.7 MB/s (3535999 of 3535999 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8271 Reviewed By: zhichao-cao Differential Revision: D28357511 Pulled By: anand1976 fbshipit-source-id: d1cfa236f00e649a18c53328be10a8062a4b6da2
2021-05-14 07:57:51 +02:00
using Cache::Lookup;
Stats for redundant insertions into block cache (#6681) Summary: Since read threads do not coordinate on loading data into block cache, two threads between Lookup and Insert can end up loading and inserting the same data. This is particularly concerning with cache_index_and_filter_blocks since those are hot and more likely to be race targets if ejected from (or not pre-populated in) the cache. Particularly with moves toward disaggregated / network storage, the cost of redundant retrieval might be high, and we should at least have some hard statistics from which we can estimate impact. Example with full filter thrashing "cliff": $ ./db_bench --benchmarks=fillrandom --num=15000000 --cache_index_and_filter_blocks -bloom_bits=10 ... $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((130 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 14181 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 476 rocksdb.block.cache.data.add COUNT : 12749 rocksdb.block.cache.data.add.redundant COUNT : 18 rocksdb.block.cache.filter.add COUNT : 1003 rocksdb.block.cache.filter.add.redundant COUNT : 217 rocksdb.block.cache.index.add COUNT : 429 rocksdb.block.cache.index.add.redundant COUNT : 241 $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((120 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 1182223 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 302728 rocksdb.block.cache.data.add COUNT : 31425 rocksdb.block.cache.data.add.redundant COUNT : 12 rocksdb.block.cache.filter.add COUNT : 795455 rocksdb.block.cache.filter.add.redundant COUNT : 130238 rocksdb.block.cache.index.add COUNT : 355343 rocksdb.block.cache.index.add.redundant COUNT : 172478 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6681 Test Plan: Some manual testing (above) and unit test covering key metrics is included Reviewed By: ltamasi Differential Revision: D21134113 Pulled By: pdillinger fbshipit-source-id: c11497b5f00f4ffdfe919823904e52d0a1a91d87
2020-04-27 22:18:18 +02:00
Handle* Lookup(const Slice& key, Statistics* stats) override {
if (nth_lookup_not_found_ == 1) {
nth_lookup_not_found_ = 0;
return nullptr;
}
if (nth_lookup_not_found_ > 1) {
--nth_lookup_not_found_;
}
return CacheWrapper::Lookup(key, stats);
}
// 1 == next lookup, 2 == after next, etc.
void SetNthLookupNotFound(int n) { nth_lookup_not_found_ = n; }
};
} // anonymous namespace
TEST_F(DBBlockCacheTest, AddRedundantStats) {
const size_t capacity = size_t{1} << 25;
const int num_shard_bits = 0; // 1 shard
int iterations_tested = 0;
for (std::shared_ptr<Cache> base_cache :
{NewLRUCache(capacity, num_shard_bits),
NewClockCache(capacity, num_shard_bits)}) {
if (!base_cache) {
// Skip clock cache when not supported
continue;
}
++iterations_tested;
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
std::shared_ptr<LookupLiarCache> cache =
std::make_shared<LookupLiarCache>(base_cache);
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.block_cache = cache;
table_options.filter_policy.reset(NewBloomFilterPolicy(50));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Stats for redundant insertions into block cache (#6681) Summary: Since read threads do not coordinate on loading data into block cache, two threads between Lookup and Insert can end up loading and inserting the same data. This is particularly concerning with cache_index_and_filter_blocks since those are hot and more likely to be race targets if ejected from (or not pre-populated in) the cache. Particularly with moves toward disaggregated / network storage, the cost of redundant retrieval might be high, and we should at least have some hard statistics from which we can estimate impact. Example with full filter thrashing "cliff": $ ./db_bench --benchmarks=fillrandom --num=15000000 --cache_index_and_filter_blocks -bloom_bits=10 ... $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((130 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 14181 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 476 rocksdb.block.cache.data.add COUNT : 12749 rocksdb.block.cache.data.add.redundant COUNT : 18 rocksdb.block.cache.filter.add COUNT : 1003 rocksdb.block.cache.filter.add.redundant COUNT : 217 rocksdb.block.cache.index.add COUNT : 429 rocksdb.block.cache.index.add.redundant COUNT : 241 $ ./db_bench --db=/tmp/rocksdbtest-172704/dbbench --use_existing_db --benchmarks=readrandom,stats --num=200000 --cache_index_and_filter_blocks --cache_size=$((120 * 1024 * 1024)) --bloom_bits=10 --threads=16 -statistics 2>&1 | egrep '^rocksdb.block.cache.(.*add|.*redundant)' | grep -v compress | sort rocksdb.block.cache.add COUNT : 1182223 rocksdb.block.cache.add.failures COUNT : 0 rocksdb.block.cache.add.redundant COUNT : 302728 rocksdb.block.cache.data.add COUNT : 31425 rocksdb.block.cache.data.add.redundant COUNT : 12 rocksdb.block.cache.filter.add COUNT : 795455 rocksdb.block.cache.filter.add.redundant COUNT : 130238 rocksdb.block.cache.index.add COUNT : 355343 rocksdb.block.cache.index.add.redundant COUNT : 172478 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6681 Test Plan: Some manual testing (above) and unit test covering key metrics is included Reviewed By: ltamasi Differential Revision: D21134113 Pulled By: pdillinger fbshipit-source-id: c11497b5f00f4ffdfe919823904e52d0a1a91d87
2020-04-27 22:18:18 +02:00
DestroyAndReopen(options);
// Create a new table.
ASSERT_OK(Put("foo", "value"));
ASSERT_OK(Put("bar", "value"));
ASSERT_OK(Flush());
ASSERT_EQ(1, NumTableFilesAtLevel(0));
// Normal access filter+index+data.
ASSERT_EQ("value", Get("foo"));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
// --------
ASSERT_EQ(3, TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
// --------
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
// Againt access filter+index+data, but force redundant load+insert on index
cache->SetNthLookupNotFound(2);
ASSERT_EQ("value", Get("bar"));
ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
// --------
ASSERT_EQ(4, TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
// --------
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
// Access just filter (with high probability), and force redundant
// load+insert
cache->SetNthLookupNotFound(1);
ASSERT_EQ("NOT_FOUND", Get("this key was not added"));
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
// --------
EXPECT_EQ(5, TestGetTickerCount(options, BLOCK_CACHE_ADD));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
EXPECT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
// --------
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
// Access just data, forcing redundant load+insert
ReadOptions read_options;
std::unique_ptr<Iterator> iter{db_->NewIterator(read_options)};
cache->SetNthLookupNotFound(1);
iter->SeekToFirst();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ(iter->key(), "bar");
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD));
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD));
EXPECT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD));
// --------
EXPECT_EQ(6, TestGetTickerCount(options, BLOCK_CACHE_ADD));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_ADD_REDUNDANT));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_ADD_REDUNDANT));
EXPECT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_DATA_ADD_REDUNDANT));
// --------
EXPECT_EQ(3, TestGetTickerCount(options, BLOCK_CACHE_ADD_REDUNDANT));
}
EXPECT_GE(iterations_tested, 1);
}
TEST_F(DBBlockCacheTest, ParanoidFileChecks) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.level0_file_num_compaction_trigger = 2;
options.paranoid_file_checks = true;
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = false;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "1_key", "val"));
ASSERT_OK(Put(1, "9_key", "val"));
// Create a new table.
ASSERT_OK(Flush(1));
ASSERT_EQ(1, /* read and cache data block */
TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_OK(Put(1, "1_key2", "val2"));
ASSERT_OK(Put(1, "9_key2", "val2"));
// Create a new SST file. This will further trigger a compaction
// and generate another file.
ASSERT_OK(Flush(1));
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_EQ(3, /* Totally 3 files created up to now */
TestGetTickerCount(options, BLOCK_CACHE_ADD));
// After disabling options.paranoid_file_checks. NO further block
// is added after generating a new file.
ASSERT_OK(
dbfull()->SetOptions(handles_[1], {{"paranoid_file_checks", "false"}}));
ASSERT_OK(Put(1, "1_key3", "val3"));
ASSERT_OK(Put(1, "9_key3", "val3"));
ASSERT_OK(Flush(1));
ASSERT_OK(Put(1, "1_key4", "val4"));
ASSERT_OK(Put(1, "9_key4", "val4"));
ASSERT_OK(Flush(1));
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_EQ(3, /* Totally 3 files created up to now */
TestGetTickerCount(options, BLOCK_CACHE_ADD));
}
TEST_F(DBBlockCacheTest, CompressedCache) {
if (!Snappy_Supported()) {
return;
}
int num_iter = 80;
// Run this test three iterations.
// Iteration 1: only a uncompressed block cache
// Iteration 2: only a compressed block cache
// Iteration 3: both block cache and compressed cache
// Iteration 4: both block cache and compressed cache, but DB is not
// compressed
for (int iter = 0; iter < 4; iter++) {
Options options = CurrentOptions();
options.write_buffer_size = 64 * 1024; // small write buffer
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
switch (iter) {
case 0:
// only uncompressed block cache
table_options.block_cache = NewLRUCache(8 * 1024);
table_options.block_cache_compressed = nullptr;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
break;
case 1:
// no block cache, only compressed cache
table_options.no_block_cache = true;
table_options.block_cache = nullptr;
table_options.block_cache_compressed = NewLRUCache(8 * 1024);
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
break;
case 2:
// both compressed and uncompressed block cache
table_options.block_cache = NewLRUCache(1024);
table_options.block_cache_compressed = NewLRUCache(8 * 1024);
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
break;
case 3:
// both block cache and compressed cache, but DB is not compressed
// also, make block cache sizes bigger, to trigger block cache hits
table_options.block_cache = NewLRUCache(1024 * 1024);
table_options.block_cache_compressed = NewLRUCache(8 * 1024 * 1024);
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.compression = kNoCompression;
break;
default:
FAIL();
}
CreateAndReopenWithCF({"pikachu"}, options);
// default column family doesn't have block cache
Options no_block_cache_opts;
no_block_cache_opts.statistics = options.statistics;
no_block_cache_opts = CurrentOptions(no_block_cache_opts);
BlockBasedTableOptions table_options_no_bc;
table_options_no_bc.no_block_cache = true;
no_block_cache_opts.table_factory.reset(
NewBlockBasedTableFactory(table_options_no_bc));
ReopenWithColumnFamilies(
{"default", "pikachu"},
std::vector<Options>({no_block_cache_opts, options}));
Random rnd(301);
// Write 8MB (80 values, each 100K)
ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
std::vector<std::string> values;
std::string str;
for (int i = 0; i < num_iter; i++) {
if (i % 4 == 0) { // high compression ratio
str = rnd.RandomString(1000);
}
values.push_back(str);
ASSERT_OK(Put(1, Key(i), values[i]));
}
// flush all data from memtable so that reads are from block cache
ASSERT_OK(Flush(1));
for (int i = 0; i < num_iter; i++) {
ASSERT_EQ(Get(1, Key(i)), values[i]);
}
// check that we triggered the appropriate code paths in the cache
switch (iter) {
case 0:
// only uncompressed block cache
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
break;
case 1:
// no block cache, only compressed cache
ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
break;
case 2:
// both compressed and uncompressed block cache
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
break;
case 3:
// both compressed and uncompressed block cache
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_MISS), 0);
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_HIT), 0);
ASSERT_GT(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_MISS), 0);
// compressed doesn't have any hits since blocks are not compressed on
// storage
ASSERT_EQ(TestGetTickerCount(options, BLOCK_CACHE_COMPRESSED_HIT), 0);
break;
default:
FAIL();
}
options.create_if_missing = true;
DestroyAndReopen(options);
}
}
TEST_F(DBBlockCacheTest, CacheCompressionDict) {
const int kNumFiles = 4;
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
const int kNumEntriesPerFile = 128;
const int kNumBytesPerEntry = 1024;
// Try all the available libraries that support dictionary compression
std::vector<CompressionType> compression_types;
if (Zlib_Supported()) {
compression_types.push_back(kZlibCompression);
}
if (LZ4_Supported()) {
compression_types.push_back(kLZ4Compression);
compression_types.push_back(kLZ4HCCompression);
}
if (ZSTD_Supported()) {
compression_types.push_back(kZSTD);
} else if (ZSTDNotFinal_Supported()) {
compression_types.push_back(kZSTDNotFinalCompression);
}
Random rnd(301);
for (auto compression_type : compression_types) {
Options options = CurrentOptions();
options.bottommost_compression = compression_type;
options.bottommost_compression_opts.max_dict_bytes = 4096;
options.bottommost_compression_opts.enabled = true;
options.create_if_missing = true;
options.num_levels = 2;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.target_file_size_base = kNumEntriesPerFile * kNumBytesPerEntry;
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.block_cache.reset(new MockCache());
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
RecordCacheCountersForCompressionDict(options);
for (int i = 0; i < kNumFiles; ++i) {
ASSERT_EQ(i, NumTableFilesAtLevel(0, 0));
for (int j = 0; j < kNumEntriesPerFile; ++j) {
std::string value = rnd.RandomString(kNumBytesPerEntry);
ASSERT_OK(Put(Key(j * kNumFiles + i), value.c_str()));
}
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_EQ(0, NumTableFilesAtLevel(0));
ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(1));
// Compression dictionary blocks are preloaded.
CheckCacheCountersForCompressionDict(
options, kNumFiles /* expected_compression_dict_misses */,
0 /* expected_compression_dict_hits */,
kNumFiles /* expected_compression_dict_inserts */);
// Seek to a key in a file. It should cause the SST's dictionary meta-block
// to be read.
RecordCacheCounters(options);
RecordCacheCountersForCompressionDict(options);
ReadOptions read_options;
ASSERT_NE("NOT_FOUND", Get(Key(kNumFiles * kNumEntriesPerFile - 1)));
// Two block hits: index and dictionary since they are prefetched
// One block missed/added: data block
CheckCacheCounters(options, 1 /* expected_misses */, 2 /* expected_hits */,
1 /* expected_inserts */, 0 /* expected_failures */);
CheckCacheCountersForCompressionDict(
options, 0 /* expected_compression_dict_misses */,
1 /* expected_compression_dict_hits */,
0 /* expected_compression_dict_inserts */);
}
}
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
static void ClearCache(Cache* cache) {
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
auto roles = CopyCacheDeleterRoleMap();
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
std::deque<std::string> keys;
Cache::ApplyToAllEntriesOptions opts;
auto callback = [&](const Slice& key, void* /*value*/, size_t /*charge*/,
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
Cache::DeleterFn deleter) {
if (roles.find(deleter) == roles.end()) {
// Keep the stats collector
return;
}
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
keys.push_back(key.ToString());
};
cache->ApplyToAllEntries(callback, opts);
for (auto& k : keys) {
cache->Erase(k);
}
}
TEST_F(DBBlockCacheTest, CacheEntryRoleStats) {
const size_t capacity = size_t{1} << 25;
int iterations_tested = 0;
for (bool partition : {false, true}) {
for (std::shared_ptr<Cache> cache :
{NewLRUCache(capacity), NewClockCache(capacity)}) {
if (!cache) {
// Skip clock cache when not supported
continue;
}
++iterations_tested;
Options options = CurrentOptions();
SetTimeElapseOnlySleepOnReopen(&options);
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.max_open_files = 13;
options.table_cache_numshardbits = 0;
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
// If this wakes up, it could interfere with test
options.stats_dump_period_sec = 0;
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.cache_index_and_filter_blocks = true;
table_options.filter_policy.reset(NewBloomFilterPolicy(50));
if (partition) {
table_options.index_type = BlockBasedTableOptions::kTwoLevelIndexSearch;
table_options.partition_filters = true;
}
table_options.metadata_cache_options.top_level_index_pinning =
PinningTier::kNone;
table_options.metadata_cache_options.partition_pinning =
PinningTier::kNone;
table_options.metadata_cache_options.unpartitioned_pinning =
PinningTier::kNone;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
// Create a new table.
ASSERT_OK(Put("foo", "value"));
ASSERT_OK(Put("bar", "value"));
ASSERT_OK(Flush());
ASSERT_OK(Put("zfoo", "value"));
ASSERT_OK(Put("zbar", "value"));
ASSERT_OK(Flush());
ASSERT_EQ(2, NumTableFilesAtLevel(0));
// Fresh cache
ClearCache(cache.get());
std::array<size_t, kNumCacheEntryRoles> expected{};
// For CacheEntryStatsCollector
expected[static_cast<size_t>(CacheEntryRole::kMisc)] = 1;
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
EXPECT_EQ(expected, GetCacheEntryRoleCountsBg());
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
std::array<size_t, kNumCacheEntryRoles> prev_expected = expected;
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
// First access only filters
ASSERT_EQ("NOT_FOUND", Get("different from any key added"));
expected[static_cast<size_t>(CacheEntryRole::kFilterBlock)] += 2;
if (partition) {
expected[static_cast<size_t>(CacheEntryRole::kFilterMetaBlock)] += 2;
}
// Within some time window, we will get cached entry stats
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
EXPECT_EQ(prev_expected, GetCacheEntryRoleCountsBg());
// Not enough to force a miss
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
env_->MockSleepForSeconds(45);
EXPECT_EQ(prev_expected, GetCacheEntryRoleCountsBg());
// Enough to force a miss
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
env_->MockSleepForSeconds(601);
EXPECT_EQ(expected, GetCacheEntryRoleCountsBg());
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
// Now access index and data block
ASSERT_EQ("value", Get("foo"));
expected[static_cast<size_t>(CacheEntryRole::kIndexBlock)]++;
if (partition) {
// top-level
expected[static_cast<size_t>(CacheEntryRole::kIndexBlock)]++;
}
expected[static_cast<size_t>(CacheEntryRole::kDataBlock)]++;
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
// Enough to force a miss
env_->MockSleepForSeconds(601);
// But inject a simulated long scan so that we need a longer
// interval to force a miss next time.
SyncPoint::GetInstance()->SetCallBack(
"CacheEntryStatsCollector::GetStats:AfterApplyToAllEntries",
[this](void*) {
// To spend no more than 0.2% of time scanning, we would need
// interval of at least 10000s
env_->MockSleepForSeconds(20);
});
SyncPoint::GetInstance()->EnableProcessing();
EXPECT_EQ(expected, GetCacheEntryRoleCountsBg());
prev_expected = expected;
SyncPoint::GetInstance()->DisableProcessing();
SyncPoint::GetInstance()->ClearAllCallBacks();
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
// The same for other file
ASSERT_EQ("value", Get("zfoo"));
expected[static_cast<size_t>(CacheEntryRole::kIndexBlock)]++;
if (partition) {
// top-level
expected[static_cast<size_t>(CacheEntryRole::kIndexBlock)]++;
}
expected[static_cast<size_t>(CacheEntryRole::kDataBlock)]++;
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
// Because of the simulated long scan, this is not enough to force
// a miss
env_->MockSleepForSeconds(601);
EXPECT_EQ(prev_expected, GetCacheEntryRoleCountsBg());
// But this is enough
env_->MockSleepForSeconds(10000);
EXPECT_EQ(expected, GetCacheEntryRoleCountsBg());
prev_expected = expected;
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
// Also check the GetProperty interface
std::map<std::string, std::string> values;
ASSERT_TRUE(
db_->GetMapProperty(DB::Properties::kBlockCacheEntryStats, &values));
EXPECT_EQ(
ToString(expected[static_cast<size_t>(CacheEntryRole::kIndexBlock)]),
values["count.index-block"]);
EXPECT_EQ(
ToString(expected[static_cast<size_t>(CacheEntryRole::kDataBlock)]),
values["count.data-block"]);
EXPECT_EQ(
ToString(expected[static_cast<size_t>(CacheEntryRole::kFilterBlock)]),
values["count.filter-block"]);
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
EXPECT_EQ(
ToString(
prev_expected[static_cast<size_t>(CacheEntryRole::kWriteBuffer)]),
values["count.write-buffer"]);
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
EXPECT_EQ(ToString(expected[static_cast<size_t>(CacheEntryRole::kMisc)]),
values["count.misc"]);
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
// Add one for kWriteBuffer
{
WriteBufferManager wbm(size_t{1} << 20, cache);
wbm.ReserveMem(1024);
expected[static_cast<size_t>(CacheEntryRole::kWriteBuffer)]++;
// Now we check that the GetProperty interface is more agressive about
// re-scanning stats, but not totally aggressive.
// Within some time window, we will get cached entry stats
env_->MockSleepForSeconds(1);
EXPECT_EQ(ToString(prev_expected[static_cast<size_t>(
CacheEntryRole::kWriteBuffer)]),
values["count.write-buffer"]);
// Not enough for a "background" miss but enough for a "foreground" miss
env_->MockSleepForSeconds(45);
ASSERT_TRUE(db_->GetMapProperty(DB::Properties::kBlockCacheEntryStats,
&values));
EXPECT_EQ(
ToString(
expected[static_cast<size_t>(CacheEntryRole::kWriteBuffer)]),
values["count.write-buffer"]);
}
prev_expected = expected;
// With collector pinned in cache, we should be able to hit
// even if the cache is full
ClearCache(cache.get());
Cache::Handle* h = nullptr;
ASSERT_OK(cache->Insert("Fill-it-up", nullptr, capacity + 1,
GetNoopDeleterForRole<CacheEntryRole::kMisc>(),
&h, Cache::Priority::HIGH));
ASSERT_GT(cache->GetUsage(), cache->GetCapacity());
expected = {};
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
// For CacheEntryStatsCollector
expected[static_cast<size_t>(CacheEntryRole::kMisc)] = 1;
// For Fill-it-up
Pin CacheEntryStatsCollector to fix performance bug (#8385) Summary: If the block Cache is full with strict_capacity_limit=false, then our CacheEntryStatsCollector could be immediately evicted on release, so iterating through column families with shared block cache could trigger re-scan for each CF. This change fixes that problem by pinning the CacheEntryStatsCollector from InternalStats so that it's not evicted. I had originally thought that this object could participate in LRU like everything else, but even though a re-load+re-scan only touches memory, it can be orders of magnitude more expensive than other cache misses. One service in Facebook has scans that take ~20s over 100GB block cache that is mostly 4KB entries. (The up-side of this bug and https://github.com/facebook/rocksdb/issues/8369 is that we had a natural experiment on the effect on some service metrics even with block cache scans running continuously in the background--a kind of worst case scenario. Metrics like latency were not affected enough to trigger warnings.) Other smaller fixes: 20s is already a sizable portion of 600s stats dump period, or 180s default max age to force re-scan, so added logic to ensure that (for each block cache) we don't spend more than 0.2% of our background thread time scanning it. Nevertheless, "foreground" requests for cache entry stats (calls to `db->GetMapProperty(DB::Properties::kBlockCacheEntryStats)`) are permitted to consume more CPU. Renamed field to cache_entry_stats_ to match code style. This change is intended for patching in 6.21 release. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8385 Test Plan: unit test expanded to cover new logic (detect regression), some manual testing with db_bench Reviewed By: ajkr Differential Revision: D29042759 Pulled By: pdillinger fbshipit-source-id: 236faa902397f50038c618f50fbc8cf3f277308c
2021-06-14 17:14:21 +02:00
expected[static_cast<size_t>(CacheEntryRole::kMisc)]++;
// Still able to hit on saved stats
EXPECT_EQ(prev_expected, GetCacheEntryRoleCountsBg());
// Enough to force a miss
env_->MockSleepForSeconds(1000);
EXPECT_EQ(expected, GetCacheEntryRoleCountsBg());
cache->Release(h);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
2021-07-16 23:12:06 +02:00
// Now we test that the DB mutex is not held during scans, for the ways
// we know how to (possibly) trigger them. Without a better good way to
// check this, we simply inject an acquire & release of the DB mutex
// deep in the stat collection code. If we were already holding the
// mutex, that is UB that would at least be found by TSAN.
int scan_count = 0;
SyncPoint::GetInstance()->SetCallBack(
"CacheEntryStatsCollector::GetStats:AfterApplyToAllEntries",
[this, &scan_count](void*) {
dbfull()->TEST_LockMutex();
dbfull()->TEST_UnlockMutex();
++scan_count;
});
SyncPoint::GetInstance()->EnableProcessing();
// Different things that might trigger a scan, with mock sleeps to
// force a miss.
env_->MockSleepForSeconds(10000);
dbfull()->DumpStats();
ASSERT_EQ(scan_count, 1);
env_->MockSleepForSeconds(10000);
ASSERT_TRUE(
db_->GetMapProperty(DB::Properties::kBlockCacheEntryStats, &values));
ASSERT_EQ(scan_count, 2);
env_->MockSleepForSeconds(10000);
std::string value_str;
ASSERT_TRUE(
db_->GetProperty(DB::Properties::kBlockCacheEntryStats, &value_str));
ASSERT_EQ(scan_count, 3);
env_->MockSleepForSeconds(10000);
ASSERT_TRUE(db_->GetProperty(DB::Properties::kCFStats, &value_str));
// To match historical speed, querying this property no longer triggers
// a scan, even if results are old. But periodic dump stats should keep
// things reasonably updated.
ASSERT_EQ(scan_count, /*unchanged*/ 3);
SyncPoint::GetInstance()->DisableProcessing();
SyncPoint::GetInstance()->ClearAllCallBacks();
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
2021-05-20 01:45:51 +02:00
}
EXPECT_GE(iterations_tested, 1);
}
}
#endif // ROCKSDB_LITE
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
class DBBlockCacheKeyTest
: public DBTestBase,
public testing::WithParamInterface<std::tuple<bool, bool>> {
public:
DBBlockCacheKeyTest()
: DBTestBase("db_block_cache_test", /*env_do_fsync=*/false) {}
void SetUp() override {
use_compressed_cache_ = std::get<0>(GetParam());
exclude_file_numbers_ = std::get<1>(GetParam());
}
bool use_compressed_cache_;
bool exclude_file_numbers_;
};
// Disable LinkFile so that we can physically copy a DB using Checkpoint.
// Disable file GetUniqueId to enable stable cache keys.
class StableCacheKeyTestFS : public FaultInjectionTestFS {
public:
explicit StableCacheKeyTestFS(const std::shared_ptr<FileSystem>& base)
: FaultInjectionTestFS(base) {
SetFailGetUniqueId(true);
}
virtual ~StableCacheKeyTestFS() override {}
IOStatus LinkFile(const std::string&, const std::string&, const IOOptions&,
IODebugContext*) override {
return IOStatus::NotSupported("Disabled");
}
};
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
TEST_P(DBBlockCacheKeyTest, StableCacheKeys) {
std::shared_ptr<StableCacheKeyTestFS> test_fs{
new StableCacheKeyTestFS(env_->GetFileSystem())};
std::unique_ptr<CompositeEnvWrapper> test_env{
new CompositeEnvWrapper(env_, test_fs)};
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.env = test_env.get();
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
BlockBasedTableOptions table_options;
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
int key_count = 0;
uint64_t expected_stat = 0;
std::function<void()> verify_stats;
if (use_compressed_cache_) {
if (!Snappy_Supported()) {
ROCKSDB_GTEST_SKIP("Compressed cache test requires snappy support");
return;
}
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
options.compression = CompressionType::kSnappyCompression;
table_options.no_block_cache = true;
table_options.block_cache_compressed = NewLRUCache(1 << 25, 0, false);
verify_stats = [&options, &expected_stat] {
// One for ordinary SST file and one for external SST file
ASSERT_EQ(expected_stat,
options.statistics->getTickerCount(BLOCK_CACHE_COMPRESSED_ADD));
};
} else {
table_options.cache_index_and_filter_blocks = true;
table_options.block_cache = NewLRUCache(1 << 25, 0, false);
verify_stats = [&options, &expected_stat] {
ASSERT_EQ(expected_stat,
options.statistics->getTickerCount(BLOCK_CACHE_DATA_ADD));
ASSERT_EQ(expected_stat,
options.statistics->getTickerCount(BLOCK_CACHE_INDEX_ADD));
ASSERT_EQ(expected_stat,
options.statistics->getTickerCount(BLOCK_CACHE_FILTER_ADD));
};
}
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"koko"}, options);
if (exclude_file_numbers_) {
// Simulate something like old behavior without file numbers in properties.
// This is a "control" side of the test that also ensures safely degraded
// behavior on old files.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"PropertyBlockBuilder::AddTableProperty:Start", [&](void* arg) {
TableProperties* props = reinterpret_cast<TableProperties*>(arg);
props->orig_file_number = 0;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
}
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
std::function<void()> perform_gets = [&key_count, &expected_stat, this]() {
if (exclude_file_numbers_) {
// No cache key reuse should happen, because we can't rely on current
// file number being stable
expected_stat += key_count;
} else {
// Cache keys should be stable
expected_stat = key_count;
}
for (int i = 0; i < key_count; ++i) {
ASSERT_EQ(Get(1, Key(i)), "abc");
}
};
// Ordinary SST files with same session id
const std::string something_compressible(500U, 'x');
for (int i = 0; i < 2; ++i) {
ASSERT_OK(Put(1, Key(key_count), "abc"));
ASSERT_OK(Put(1, Key(key_count) + "a", something_compressible));
ASSERT_OK(Flush(1));
++key_count;
}
#ifndef ROCKSDB_LITE
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
// Save an export of those ordinary SST files for later
std::string export_files_dir = dbname_ + "/exported";
ExportImportFilesMetaData* metadata_ptr_ = nullptr;
Checkpoint* checkpoint;
ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
ASSERT_OK(checkpoint->ExportColumnFamily(handles_[1], export_files_dir,
&metadata_ptr_));
ASSERT_NE(metadata_ptr_, nullptr);
delete checkpoint;
checkpoint = nullptr;
// External SST files with same session id
SstFileWriter sst_file_writer(EnvOptions(), options);
std::vector<std::string> external;
for (int i = 0; i < 2; ++i) {
std::string f = dbname_ + "/external" + ToString(i) + ".sst";
external.push_back(f);
ASSERT_OK(sst_file_writer.Open(f));
ASSERT_OK(sst_file_writer.Put(Key(key_count), "abc"));
ASSERT_OK(
sst_file_writer.Put(Key(key_count) + "a", something_compressible));
++key_count;
ExternalSstFileInfo external_info;
ASSERT_OK(sst_file_writer.Finish(&external_info));
IngestExternalFileOptions ingest_opts;
ASSERT_OK(db_->IngestExternalFile(handles_[1], {f}, ingest_opts));
}
if (exclude_file_numbers_) {
// FIXME(peterd): figure out where these extra two ADDs are coming from
options.statistics->recordTick(BLOCK_CACHE_INDEX_ADD,
uint64_t{0} - uint64_t{2});
options.statistics->recordTick(BLOCK_CACHE_FILTER_ADD,
uint64_t{0} - uint64_t{2});
options.statistics->recordTick(BLOCK_CACHE_COMPRESSED_ADD,
uint64_t{0} - uint64_t{2});
}
#endif
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
perform_gets();
verify_stats();
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
// Make sure we can cache hit after re-open
ReopenWithColumnFamilies({"default", "koko"}, options);
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
perform_gets();
verify_stats();
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
// Make sure we can cache hit even on a full copy of the DB. Using
// StableCacheKeyTestFS, Checkpoint will resort to full copy not hard link.
// (Checkpoint not available in LITE mode to test this.)
#ifndef ROCKSDB_LITE
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
auto db_copy_name = dbname_ + "-copy";
ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
ASSERT_OK(checkpoint->CreateCheckpoint(db_copy_name));
delete checkpoint;
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
Close();
Destroy(options);
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
// Switch to the DB copy
SaveAndRestore<std::string> save_dbname(&dbname_, db_copy_name);
ReopenWithColumnFamilies({"default", "koko"}, options);
perform_gets();
verify_stats();
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
// And ensure that re-importing + ingesting the same files into a
// different DB uses same cache keys
DestroyAndReopen(options);
ColumnFamilyHandle* cfh = nullptr;
ASSERT_OK(db_->CreateColumnFamilyWithImport(ColumnFamilyOptions(), "yoyo",
ImportColumnFamilyOptions(),
*metadata_ptr_, &cfh));
ASSERT_NE(cfh, nullptr);
delete cfh;
cfh = nullptr;
delete metadata_ptr_;
metadata_ptr_ = nullptr;
DestroyDB(export_files_dir, options);
ReopenWithColumnFamilies({"default", "yoyo"}, options);
IngestExternalFileOptions ingest_opts;
ASSERT_OK(db_->IngestExternalFile(handles_[1], {external}, ingest_opts));
if (exclude_file_numbers_) {
// FIXME(peterd): figure out where these extra two ADDs are coming from
options.statistics->recordTick(BLOCK_CACHE_INDEX_ADD,
uint64_t{0} - uint64_t{2});
options.statistics->recordTick(BLOCK_CACHE_FILTER_ADD,
uint64_t{0} - uint64_t{2});
}
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
perform_gets();
verify_stats();
#endif // !ROCKSDB_LITE
Close();
Destroy(options);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
Embed original file number in SST table properties (#8686) Summary: I very recently realized that with https://github.com/facebook/rocksdb/issues/8669 we cannot later add file numbers to external SST files (so that more can share db session ids for better uniqueness properties), because of forward compatibility. We would have a version of RocksDB that assumes session IDs are unique on external SST files and therefore can't really break that invariant in future files. This change adds a table property for "orig_file_number" which is populated by normal SST files and also external SST files generated by SstFileWriter. SstFileWriter now keeps a db_session_id for life of the object and increments its own file numbers for embedding in table properties. (They are arguably "fake" file numbers because these numbers and not embedded in the file name.) While updating block_based_table_builder, I removed several unnecessary fields from Rep, because following the pattern would have created another unnecessary field. This change also updates block_based_table_reader to use this new property when available, which means that for newer SST files, we can determine the stable/original <db_session_id,file_number> unique identifier using just the file contents, not the file name. (It's a bit complicated; detailed comments in block_based_table_reader.) Also added DB host id to properties listing by sst_dump, which could be useful in debugging. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8686 Test Plan: majorly overhauled StableCacheKeys test for this change Reviewed By: zhichao-cao Differential Revision: D30457742 Pulled By: pdillinger fbshipit-source-id: 2e5ae7dddeb94fb9d8eac8a928486aed8b8cd445
2021-08-21 05:39:52 +02:00
INSTANTIATE_TEST_CASE_P(DBBlockCacheKeyTest, DBBlockCacheKeyTest,
::testing::Combine(::testing::Bool(),
::testing::Bool()));
class DBBlockCachePinningTest
: public DBTestBase,
public testing::WithParamInterface<
std::tuple<bool, PinningTier, PinningTier, PinningTier>> {
public:
DBBlockCachePinningTest()
: DBTestBase("db_block_cache_test", /*env_do_fsync=*/false) {}
void SetUp() override {
partition_index_and_filters_ = std::get<0>(GetParam());
top_level_index_pinning_ = std::get<1>(GetParam());
partition_pinning_ = std::get<2>(GetParam());
unpartitioned_pinning_ = std::get<3>(GetParam());
}
bool partition_index_and_filters_;
PinningTier top_level_index_pinning_;
PinningTier partition_pinning_;
PinningTier unpartitioned_pinning_;
};
TEST_P(DBBlockCachePinningTest, TwoLevelDB) {
// Creates one file in L0 and one file in L1. Both files have enough data that
// their index and filter blocks are partitioned. The L1 file will also have
// a compression dictionary (those are trained only during compaction), which
// must be unpartitioned.
const int kKeySize = 32;
const int kBlockSize = 128;
const int kNumBlocksPerFile = 128;
const int kNumKeysPerFile = kBlockSize * kNumBlocksPerFile / kKeySize;
Options options = CurrentOptions();
// `kNoCompression` makes the unit test more portable. But it relies on the
// current behavior of persisting/accessing dictionary even when there's no
// (de)compression happening, which seems fairly likely to change over time.
options.compression = kNoCompression;
options.compression_opts.max_dict_bytes = 4 << 10;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.block_cache = NewLRUCache(1 << 20 /* capacity */);
table_options.block_size = kBlockSize;
table_options.metadata_block_size = kBlockSize;
table_options.cache_index_and_filter_blocks = true;
table_options.metadata_cache_options.top_level_index_pinning =
top_level_index_pinning_;
table_options.metadata_cache_options.partition_pinning = partition_pinning_;
table_options.metadata_cache_options.unpartitioned_pinning =
unpartitioned_pinning_;
table_options.filter_policy.reset(
NewBloomFilterPolicy(10 /* bits_per_key */));
if (partition_index_and_filters_) {
table_options.index_type =
BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
table_options.partition_filters = true;
}
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
Random rnd(301);
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < kNumKeysPerFile; ++j) {
ASSERT_OK(Put(Key(i * kNumKeysPerFile + j), rnd.RandomString(kKeySize)));
}
ASSERT_OK(Flush());
if (i == 0) {
// Prevent trivial move so file will be rewritten with dictionary and
// reopened with L1's pinning settings.
CompactRangeOptions cro;
cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
}
}
// Clear all unpinned blocks so unpinned blocks will show up as cache misses
// when reading a key from a file.
table_options.block_cache->EraseUnRefEntries();
// Get base cache values
uint64_t filter_misses = TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
uint64_t index_misses = TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS);
uint64_t compression_dict_misses =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS);
// Read a key from the L0 file
Get(Key(kNumKeysPerFile));
uint64_t expected_filter_misses = filter_misses;
uint64_t expected_index_misses = index_misses;
uint64_t expected_compression_dict_misses = compression_dict_misses;
if (partition_index_and_filters_) {
if (top_level_index_pinning_ == PinningTier::kNone) {
++expected_filter_misses;
++expected_index_misses;
}
if (partition_pinning_ == PinningTier::kNone) {
++expected_filter_misses;
++expected_index_misses;
}
} else {
if (unpartitioned_pinning_ == PinningTier::kNone) {
++expected_filter_misses;
++expected_index_misses;
}
}
if (unpartitioned_pinning_ == PinningTier::kNone) {
++expected_compression_dict_misses;
}
ASSERT_EQ(expected_filter_misses,
TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(expected_index_misses,
TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(expected_compression_dict_misses,
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
// Clear all unpinned blocks so unpinned blocks will show up as cache misses
// when reading a key from a file.
table_options.block_cache->EraseUnRefEntries();
// Read a key from the L1 file
Get(Key(0));
if (partition_index_and_filters_) {
if (top_level_index_pinning_ == PinningTier::kNone ||
top_level_index_pinning_ == PinningTier::kFlushedAndSimilar) {
++expected_filter_misses;
++expected_index_misses;
}
if (partition_pinning_ == PinningTier::kNone ||
partition_pinning_ == PinningTier::kFlushedAndSimilar) {
++expected_filter_misses;
++expected_index_misses;
}
} else {
if (unpartitioned_pinning_ == PinningTier::kNone ||
unpartitioned_pinning_ == PinningTier::kFlushedAndSimilar) {
++expected_filter_misses;
++expected_index_misses;
}
}
if (unpartitioned_pinning_ == PinningTier::kNone ||
unpartitioned_pinning_ == PinningTier::kFlushedAndSimilar) {
++expected_compression_dict_misses;
}
ASSERT_EQ(expected_filter_misses,
TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(expected_index_misses,
TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(expected_compression_dict_misses,
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_MISS));
}
INSTANTIATE_TEST_CASE_P(
DBBlockCachePinningTest, DBBlockCachePinningTest,
::testing::Combine(
::testing::Bool(),
::testing::Values(PinningTier::kNone, PinningTier::kFlushedAndSimilar,
PinningTier::kAll),
::testing::Values(PinningTier::kNone, PinningTier::kFlushedAndSimilar,
PinningTier::kAll),
::testing::Values(PinningTier::kNone, PinningTier::kFlushedAndSimilar,
PinningTier::kAll)));
} // namespace ROCKSDB_NAMESPACE
int main(int argc, char** argv) {
ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}