rocksdb/cache/lru_cache_test.cc

1851 lines
67 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).
#include "cache/lru_cache.h"
#include <string>
#include <vector>
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 05:57:51 +00:00
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
2021-12-17 01:13:55 +00:00
#include "cache/cache_key.h"
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 01:28:28 +00:00
#include "db/db_test_util.h"
#include "file/sst_file_manager_impl.h"
#include "port/port.h"
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 01:28:28 +00:00
#include "port/stack_trace.h"
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 05:57:51 +00:00
#include "rocksdb/cache.h"
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 01:28:28 +00:00
#include "rocksdb/io_status.h"
#include "rocksdb/sst_file_manager.h"
#include "rocksdb/utilities/cache_dump_load.h"
#include "test_util/testharness.h"
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 05:57:51 +00:00
#include "util/coding.h"
#include "util/random.h"
#include "utilities/cache_dump_load_impl.h"
#include "utilities/fault_injection_fs.h"
namespace ROCKSDB_NAMESPACE {
class LRUCacheTest : public testing::Test {
public:
LRUCacheTest() {}
~LRUCacheTest() override { DeleteCache(); }
void DeleteCache() {
if (cache_ != nullptr) {
cache_->~LRUCacheShard();
port::cacheline_aligned_free(cache_);
cache_ = nullptr;
}
}
void NewCache(size_t capacity, double high_pri_pool_ratio = 0.0,
bool use_adaptive_mutex = kDefaultToAdaptiveMutex) {
DeleteCache();
cache_ = reinterpret_cast<LRUCacheShard*>(
port::cacheline_aligned_alloc(sizeof(LRUCacheShard)));
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 05:57:51 +00:00
new (cache_) LRUCacheShard(
capacity, false /*strict_capcity_limit*/, high_pri_pool_ratio,
use_adaptive_mutex, kDontChargeCacheMetadata,
24 /*max_upper_hash_bits*/, nullptr /*secondary_cache*/);
}
void Insert(const std::string& key,
Cache::Priority priority = Cache::Priority::LOW) {
EXPECT_OK(cache_->Insert(key, 0 /*hash*/, nullptr /*value*/, 1 /*charge*/,
nullptr /*deleter*/, nullptr /*handle*/,
priority));
}
void Insert(char key, Cache::Priority priority = Cache::Priority::LOW) {
Insert(std::string(1, key), priority);
}
bool Lookup(const std::string& key) {
auto handle = cache_->Lookup(key, 0 /*hash*/);
if (handle) {
cache_->Release(handle);
return true;
}
return false;
}
bool Lookup(char key) { return Lookup(std::string(1, key)); }
void Erase(const std::string& key) { cache_->Erase(key, 0 /*hash*/); }
void ValidateLRUList(std::vector<std::string> keys,
size_t num_high_pri_pool_keys = 0) {
LRUHandle* lru;
LRUHandle* lru_low_pri;
cache_->TEST_GetLRUList(&lru, &lru_low_pri);
LRUHandle* iter = lru;
bool in_high_pri_pool = false;
size_t high_pri_pool_keys = 0;
if (iter == lru_low_pri) {
in_high_pri_pool = true;
}
for (const auto& key : keys) {
iter = iter->next;
ASSERT_NE(lru, iter);
ASSERT_EQ(key, iter->key().ToString());
ASSERT_EQ(in_high_pri_pool, iter->InHighPriPool());
if (in_high_pri_pool) {
high_pri_pool_keys++;
}
if (iter == lru_low_pri) {
ASSERT_FALSE(in_high_pri_pool);
in_high_pri_pool = true;
}
}
ASSERT_EQ(lru, iter->next);
ASSERT_TRUE(in_high_pri_pool);
ASSERT_EQ(num_high_pri_pool_keys, high_pri_pool_keys);
}
private:
LRUCacheShard* cache_ = nullptr;
};
TEST_F(LRUCacheTest, BasicLRU) {
NewCache(5);
for (char ch = 'a'; ch <= 'e'; ch++) {
Insert(ch);
}
ValidateLRUList({"a", "b", "c", "d", "e"});
for (char ch = 'x'; ch <= 'z'; ch++) {
Insert(ch);
}
ValidateLRUList({"d", "e", "x", "y", "z"});
ASSERT_FALSE(Lookup("b"));
ValidateLRUList({"d", "e", "x", "y", "z"});
ASSERT_TRUE(Lookup("e"));
ValidateLRUList({"d", "x", "y", "z", "e"});
ASSERT_TRUE(Lookup("z"));
ValidateLRUList({"d", "x", "y", "e", "z"});
Erase("x");
ValidateLRUList({"d", "y", "e", "z"});
ASSERT_TRUE(Lookup("d"));
ValidateLRUList({"y", "e", "z", "d"});
Insert("u");
ValidateLRUList({"y", "e", "z", "d", "u"});
Insert("v");
ValidateLRUList({"e", "z", "d", "u", "v"});
}
TEST_F(LRUCacheTest, MidpointInsertion) {
// Allocate 2 cache entries to high-pri pool.
NewCache(5, 0.45);
Insert("a", Cache::Priority::LOW);
Insert("b", Cache::Priority::LOW);
Insert("c", Cache::Priority::LOW);
Insert("x", Cache::Priority::HIGH);
Insert("y", Cache::Priority::HIGH);
ValidateLRUList({"a", "b", "c", "x", "y"}, 2);
// Low-pri entries inserted to the tail of low-pri list (the midpoint).
// After lookup, it will move to the tail of the full list.
Insert("d", Cache::Priority::LOW);
ValidateLRUList({"b", "c", "d", "x", "y"}, 2);
ASSERT_TRUE(Lookup("d"));
ValidateLRUList({"b", "c", "x", "y", "d"}, 2);
// High-pri entries will be inserted to the tail of full list.
Insert("z", Cache::Priority::HIGH);
ValidateLRUList({"c", "x", "y", "d", "z"}, 2);
}
TEST_F(LRUCacheTest, EntriesWithPriority) {
// Allocate 2 cache entries to high-pri pool.
NewCache(5, 0.45);
Insert("a", Cache::Priority::LOW);
Insert("b", Cache::Priority::LOW);
Insert("c", Cache::Priority::LOW);
ValidateLRUList({"a", "b", "c"}, 0);
// Low-pri entries can take high-pri pool capacity if available
Insert("u", Cache::Priority::LOW);
Insert("v", Cache::Priority::LOW);
ValidateLRUList({"a", "b", "c", "u", "v"}, 0);
Insert("X", Cache::Priority::HIGH);
Insert("Y", Cache::Priority::HIGH);
ValidateLRUList({"c", "u", "v", "X", "Y"}, 2);
// High-pri entries can overflow to low-pri pool.
Insert("Z", Cache::Priority::HIGH);
ValidateLRUList({"u", "v", "X", "Y", "Z"}, 2);
// Low-pri entries will be inserted to head of low-pri pool.
Insert("a", Cache::Priority::LOW);
ValidateLRUList({"v", "X", "a", "Y", "Z"}, 2);
// Low-pri entries will be inserted to head of high-pri pool after lookup.
ASSERT_TRUE(Lookup("v"));
ValidateLRUList({"X", "a", "Y", "Z", "v"}, 2);
// High-pri entries will be inserted to the head of the list after lookup.
ASSERT_TRUE(Lookup("X"));
ValidateLRUList({"a", "Y", "Z", "v", "X"}, 2);
ASSERT_TRUE(Lookup("Z"));
ValidateLRUList({"a", "Y", "v", "X", "Z"}, 2);
Erase("Y");
ValidateLRUList({"a", "v", "X", "Z"}, 2);
Erase("X");
ValidateLRUList({"a", "v", "Z"}, 1);
Insert("d", Cache::Priority::LOW);
Insert("e", Cache::Priority::LOW);
ValidateLRUList({"a", "v", "d", "e", "Z"}, 1);
Insert("f", Cache::Priority::LOW);
Insert("g", Cache::Priority::LOW);
ValidateLRUList({"d", "e", "f", "g", "Z"}, 1);
ASSERT_TRUE(Lookup("d"));
ValidateLRUList({"e", "f", "g", "Z", "d"}, 2);
}
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 05:57:51 +00:00
class TestSecondaryCache : public SecondaryCache {
public:
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
// Specifies what action to take on a lookup for a particular key
enum ResultType {
SUCCESS,
// Fail lookup immediately
FAIL,
// Defer the result. It will returned after Wait/WaitAll is called
DEFER,
// Defer the result and eventually return failure
DEFER_AND_FAIL
};
using ResultMap = std::unordered_map<std::string, ResultType>;
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 05:57:51 +00:00
explicit TestSecondaryCache(size_t capacity)
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 01:28:28 +00:00
: num_inserts_(0), num_lookups_(0), inject_failure_(false) {
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 05:57:51 +00:00
cache_ = NewLRUCache(capacity, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
}
~TestSecondaryCache() override { cache_.reset(); }
const char* Name() const override { return "TestSecondaryCache"; }
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 05:57:51 +00:00
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 01:28:28 +00:00
void InjectFailure() { inject_failure_ = true; }
void ResetInjectFailure() { inject_failure_ = false; }
void SetDbSessionId(const std::string& db_session_id) {
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
2021-12-17 01:13:55 +00:00
// NOTE: we assume the file is smaller than kMaxFileSizeStandardEncoding
// for this to work, but that's safe in a test.
auto base = OffsetableCacheKey("unknown", db_session_id, 1, 1);
ckey_prefix_ = base.CommonPrefixSlice().ToString();
}
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 05:57:51 +00:00
Status Insert(const Slice& key, void* value,
const Cache::CacheItemHelper* helper) override {
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 01:28:28 +00:00
if (inject_failure_) {
return Status::Corruption("Insertion Data Corrupted");
}
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
2021-12-17 01:13:55 +00:00
EXPECT_TRUE(IsDbSessionLowerAsKeyPrefix(key));
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 05:57:51 +00:00
size_t size;
char* buf;
Status s;
num_inserts_++;
size = (*helper->size_cb)(value);
buf = new char[size + sizeof(uint64_t)];
EncodeFixed64(buf, size);
s = (*helper->saveto_cb)(value, 0, size, buf + sizeof(uint64_t));
if (!s.ok()) {
delete[] buf;
return s;
}
return cache_->Insert(key, buf, size,
[](const Slice& /*key*/, void* val) -> void {
delete[] static_cast<char*>(val);
});
}
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
std::unique_ptr<SecondaryCacheResultHandle> Lookup(
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 05:57:51 +00:00
const Slice& key, const Cache::CreateCallback& create_cb,
bool /*wait*/) override {
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
std::string key_str = key.ToString();
TEST_SYNC_POINT_CALLBACK("TestSecondaryCache::Lookup", &key_str);
std::unique_ptr<SecondaryCacheResultHandle> secondary_handle;
ResultType type = ResultType::SUCCESS;
auto iter = result_map_.find(key.ToString());
if (iter != result_map_.end()) {
type = iter->second;
}
if (type == ResultType::FAIL) {
return secondary_handle;
}
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 05:57:51 +00:00
Cache::Handle* handle = cache_->Lookup(key);
num_lookups_++;
if (handle) {
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
void* value = nullptr;
size_t charge = 0;
Status s;
if (type != ResultType::DEFER_AND_FAIL) {
char* ptr = (char*)cache_->Value(handle);
size_t size = DecodeFixed64(ptr);
ptr += sizeof(uint64_t);
s = create_cb(ptr, size, &value, &charge);
}
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 05:57:51 +00:00
if (s.ok()) {
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
secondary_handle.reset(new TestSecondaryCacheResultHandle(
cache_.get(), handle, value, charge, type));
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 05:57:51 +00:00
} else {
cache_->Release(handle);
}
}
return secondary_handle;
}
void Erase(const Slice& /*key*/) override {}
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
void WaitAll(std::vector<SecondaryCacheResultHandle*> handles) override {
for (SecondaryCacheResultHandle* handle : handles) {
TestSecondaryCacheResultHandle* sec_handle =
static_cast<TestSecondaryCacheResultHandle*>(handle);
sec_handle->SetReady();
}
}
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 05:57:51 +00:00
std::string GetPrintableOptions() const override { return ""; }
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
void SetResultMap(ResultMap&& map) { result_map_ = std::move(map); }
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 05:57:51 +00:00
uint32_t num_inserts() { return num_inserts_; }
uint32_t num_lookups() { return num_lookups_; }
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
2021-12-17 01:13:55 +00:00
bool IsDbSessionLowerAsKeyPrefix(const Slice& key) {
return key.starts_with(ckey_prefix_);
}
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 05:57:51 +00:00
private:
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
class TestSecondaryCacheResultHandle : public SecondaryCacheResultHandle {
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 05:57:51 +00:00
public:
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
TestSecondaryCacheResultHandle(Cache* cache, Cache::Handle* handle,
void* value, size_t size, ResultType type)
: cache_(cache),
handle_(handle),
value_(value),
size_(size),
is_ready_(true) {
if (type != ResultType::SUCCESS) {
is_ready_ = false;
}
}
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 05:57:51 +00:00
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
~TestSecondaryCacheResultHandle() override { cache_->Release(handle_); }
bool IsReady() override { return is_ready_; }
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 05:57:51 +00:00
void Wait() override {}
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
void* Value() override {
assert(is_ready_);
return value_;
}
size_t Size() override { return Value() ? size_ : 0; }
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 05:57:51 +00:00
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
void SetReady() { is_ready_ = true; }
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 05:57:51 +00:00
private:
Cache* cache_;
Cache::Handle* handle_;
void* value_;
size_t size_;
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
bool is_ready_;
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 05:57:51 +00:00
};
std::shared_ptr<Cache> cache_;
uint32_t num_inserts_;
uint32_t num_lookups_;
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 01:28:28 +00:00
bool inject_failure_;
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
2021-12-17 01:13:55 +00:00
std::string ckey_prefix_;
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
ResultMap result_map_;
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 01:28:28 +00:00
};
class DBSecondaryCacheTest : public DBTestBase {
public:
DBSecondaryCacheTest()
: DBTestBase("db_secondary_cache_test", /*env_do_fsync=*/true) {
fault_fs_.reset(new FaultInjectionTestFS(env_->GetFileSystem()));
fault_env_.reset(new CompositeEnvWrapper(env_, fault_fs_));
}
std::shared_ptr<FaultInjectionTestFS> fault_fs_;
std::unique_ptr<Env> fault_env_;
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 05:57:51 +00:00
};
class LRUSecondaryCacheTest : public LRUCacheTest {
public:
LRUSecondaryCacheTest() : fail_create_(false) {}
~LRUSecondaryCacheTest() {}
protected:
class TestItem {
public:
TestItem(const char* buf, size_t size) : buf_(new char[size]), size_(size) {
memcpy(buf_.get(), buf, size);
}
~TestItem() {}
char* Buf() { return buf_.get(); }
size_t Size() { return size_; }
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
std::string ToString() { return std::string(Buf(), Size()); }
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 05:57:51 +00:00
private:
std::unique_ptr<char[]> buf_;
size_t size_;
};
static size_t SizeCallback(void* obj) {
return reinterpret_cast<TestItem*>(obj)->Size();
}
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 01:28:28 +00:00
static Status SaveToCallback(void* from_obj, size_t from_offset,
size_t length, void* out) {
TestItem* item = reinterpret_cast<TestItem*>(from_obj);
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 05:57:51 +00:00
char* buf = item->Buf();
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 01:28:28 +00:00
EXPECT_EQ(length, item->Size());
EXPECT_EQ(from_offset, 0);
memcpy(out, buf, length);
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 05:57:51 +00:00
return Status::OK();
}
static void DeletionCallback(const Slice& /*key*/, void* obj) {
delete reinterpret_cast<TestItem*>(obj);
}
static Cache::CacheItemHelper helper_;
static Status SaveToCallbackFail(void* /*obj*/, size_t /*offset*/,
size_t /*size*/, void* /*out*/) {
return Status::NotSupported();
}
static Cache::CacheItemHelper helper_fail_;
Cache::CreateCallback test_item_creator =
[&](void* buf, size_t size, void** out_obj, size_t* charge) -> Status {
if (fail_create_) {
return Status::NotSupported();
}
*out_obj = reinterpret_cast<void*>(new TestItem((char*)buf, size));
*charge = size;
return Status::OK();
};
void SetFailCreate(bool fail) { fail_create_ = fail; }
private:
bool fail_create_;
};
Cache::CacheItemHelper LRUSecondaryCacheTest::helper_(
LRUSecondaryCacheTest::SizeCallback, LRUSecondaryCacheTest::SaveToCallback,
LRUSecondaryCacheTest::DeletionCallback);
Cache::CacheItemHelper LRUSecondaryCacheTest::helper_fail_(
LRUSecondaryCacheTest::SizeCallback,
LRUSecondaryCacheTest::SaveToCallbackFail,
LRUSecondaryCacheTest::DeletionCallback);
TEST_F(LRUSecondaryCacheTest, BasicTest) {
LRUCacheOptions opts(1024, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
std::shared_ptr<Statistics> stats = CreateDBStatistics();
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 05:57:51 +00:00
Random rnd(301);
std::string str1 = rnd.RandomString(1020);
TestItem* item1 = new TestItem(str1.data(), str1.length());
ASSERT_OK(cache->Insert("k1", item1, &LRUSecondaryCacheTest::helper_,
str1.length()));
std::string str2 = rnd.RandomString(1020);
TestItem* item2 = new TestItem(str2.data(), str2.length());
// k1 should be demoted to NVM
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 05:57:51 +00:00
ASSERT_OK(cache->Insert("k2", item2, &LRUSecondaryCacheTest::helper_,
str2.length()));
get_perf_context()->Reset();
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 05:57:51 +00:00
Cache::Handle* handle;
handle =
cache->Lookup("k2", &LRUSecondaryCacheTest::helper_, test_item_creator,
Cache::Priority::LOW, true, stats.get());
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 05:57:51 +00:00
ASSERT_NE(handle, nullptr);
cache->Release(handle);
// This lookup should promote k1 and demote k2
handle =
cache->Lookup("k1", &LRUSecondaryCacheTest::helper_, test_item_creator,
Cache::Priority::LOW, true, stats.get());
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 05:57:51 +00:00
ASSERT_NE(handle, nullptr);
cache->Release(handle);
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 1u);
ASSERT_EQ(stats->getTickerCount(SECONDARY_CACHE_HITS),
secondary_cache->num_lookups());
PerfContext perf_ctx = *get_perf_context();
ASSERT_EQ(perf_ctx.secondary_cache_hit_count, secondary_cache->num_lookups());
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 05:57:51 +00:00
cache.reset();
secondary_cache.reset();
}
TEST_F(LRUSecondaryCacheTest, BasicFailTest) {
LRUCacheOptions opts(1024, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
Random rnd(301);
std::string str1 = rnd.RandomString(1020);
Require C++17 (#9481) Summary: Drop support for some old compilers by requiring C++17 standard (or higher). See https://github.com/facebook/rocksdb/issues/9388 First modification based on this is to remove some conditional compilation in slice.h (also better for ODR) Also in this PR: * Fix some Makefile formatting that seems to affect ASSERT_STATUS_CHECKED config in some cases * Add c_test to NON_PARALLEL_TEST in Makefile * Fix a clang-analyze reported "potential leak" in lru_cache_test * Better "compatibility" definition of DEFINE_uint32 for old versions of gflags * Fix a linking problem with shared libraries in Makefile (`./random_test: error while loading shared libraries: librocksdb.so.6.29: cannot open shared object file: No such file or directory`) * Always set ROCKSDB_SUPPORT_THREAD_LOCAL and use thread_local (from C++11) * TODO in later PR: clean up that obsolete flag * Fix a cosmetic typo in c.h (https://github.com/facebook/rocksdb/issues/9488) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9481 Test Plan: CircleCI config substantially updated. * Upgrade to latest Ubuntu images for each release * Generally prefer Ubuntu 20, but keep a couple Ubuntu 16 builds with oldest supported compilers, to ensure compatibility * Remove .circleci/cat_ignore_eagain except for Ubuntu 16 builds, because this is to work around a kernel bug that should not affect anything but Ubuntu 16. * Remove designated gcc-9 build, because the default linux build now uses GCC 9 from Ubuntu 20. * Add some `apt-key add` to fix some apt "couldn't be verified" errors * Generally drop SKIP_LINK=1; work-around no longer needed * Generally `add-apt-repository` before `apt-get update` as manual testing indicated the reverse might not work. Travis: * Use gcc-7 by default (remove specific gcc-7 and gcc-4.8 builds) * TODO in later PR: fix s390x "Assembler messages: Error: invalid switch -march=z14" failure AppVeyor: * Completely dropped because we are dropping VS2015 support and CircleCI covers VS >= 2017 Also local testing with old gflags (out of necessity when using ROCKSDB_NO_FBCODE=1). Reviewed By: mrambacher Differential Revision: D33946377 Pulled By: pdillinger fbshipit-source-id: ae077c823905b45370a26c0103ada119459da6c1
2022-02-05 01:12:03 +00:00
auto item1 = std::make_unique<TestItem>(str1.data(), str1.length());
ASSERT_TRUE(cache->Insert("k1", item1.get(), nullptr, str1.length())
.IsInvalidArgument());
ASSERT_OK(cache->Insert("k1", item1.get(), &LRUSecondaryCacheTest::helper_,
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 05:57:51 +00:00
str1.length()));
Require C++17 (#9481) Summary: Drop support for some old compilers by requiring C++17 standard (or higher). See https://github.com/facebook/rocksdb/issues/9388 First modification based on this is to remove some conditional compilation in slice.h (also better for ODR) Also in this PR: * Fix some Makefile formatting that seems to affect ASSERT_STATUS_CHECKED config in some cases * Add c_test to NON_PARALLEL_TEST in Makefile * Fix a clang-analyze reported "potential leak" in lru_cache_test * Better "compatibility" definition of DEFINE_uint32 for old versions of gflags * Fix a linking problem with shared libraries in Makefile (`./random_test: error while loading shared libraries: librocksdb.so.6.29: cannot open shared object file: No such file or directory`) * Always set ROCKSDB_SUPPORT_THREAD_LOCAL and use thread_local (from C++11) * TODO in later PR: clean up that obsolete flag * Fix a cosmetic typo in c.h (https://github.com/facebook/rocksdb/issues/9488) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9481 Test Plan: CircleCI config substantially updated. * Upgrade to latest Ubuntu images for each release * Generally prefer Ubuntu 20, but keep a couple Ubuntu 16 builds with oldest supported compilers, to ensure compatibility * Remove .circleci/cat_ignore_eagain except for Ubuntu 16 builds, because this is to work around a kernel bug that should not affect anything but Ubuntu 16. * Remove designated gcc-9 build, because the default linux build now uses GCC 9 from Ubuntu 20. * Add some `apt-key add` to fix some apt "couldn't be verified" errors * Generally drop SKIP_LINK=1; work-around no longer needed * Generally `add-apt-repository` before `apt-get update` as manual testing indicated the reverse might not work. Travis: * Use gcc-7 by default (remove specific gcc-7 and gcc-4.8 builds) * TODO in later PR: fix s390x "Assembler messages: Error: invalid switch -march=z14" failure AppVeyor: * Completely dropped because we are dropping VS2015 support and CircleCI covers VS >= 2017 Also local testing with old gflags (out of necessity when using ROCKSDB_NO_FBCODE=1). Reviewed By: mrambacher Differential Revision: D33946377 Pulled By: pdillinger fbshipit-source-id: ae077c823905b45370a26c0103ada119459da6c1
2022-02-05 01:12:03 +00:00
item1.release(); // Appease clang-analyze "potential memory leak"
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 05:57:51 +00:00
Cache::Handle* handle;
handle = cache->Lookup("k2", nullptr, test_item_creator, Cache::Priority::LOW,
true);
ASSERT_EQ(handle, nullptr);
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, false);
ASSERT_EQ(handle, nullptr);
cache.reset();
secondary_cache.reset();
}
TEST_F(LRUSecondaryCacheTest, SaveFailTest) {
LRUCacheOptions opts(1024, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
Random rnd(301);
std::string str1 = rnd.RandomString(1020);
TestItem* item1 = new TestItem(str1.data(), str1.length());
ASSERT_OK(cache->Insert("k1", item1, &LRUSecondaryCacheTest::helper_fail_,
str1.length()));
std::string str2 = rnd.RandomString(1020);
TestItem* item2 = new TestItem(str2.data(), str2.length());
// k1 should be demoted to NVM
ASSERT_OK(cache->Insert("k2", item2, &LRUSecondaryCacheTest::helper_fail_,
str2.length()));
Cache::Handle* handle;
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_fail_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
cache->Release(handle);
// This lookup should fail, since k1 demotion would have failed
handle = cache->Lookup("k1", &LRUSecondaryCacheTest::helper_fail_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_EQ(handle, nullptr);
// Since k1 didn't get promoted, k2 should still be in cache
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_fail_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
cache->Release(handle);
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 1u);
cache.reset();
secondary_cache.reset();
}
TEST_F(LRUSecondaryCacheTest, CreateFailTest) {
LRUCacheOptions opts(1024, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
Random rnd(301);
std::string str1 = rnd.RandomString(1020);
TestItem* item1 = new TestItem(str1.data(), str1.length());
ASSERT_OK(cache->Insert("k1", item1, &LRUSecondaryCacheTest::helper_,
str1.length()));
std::string str2 = rnd.RandomString(1020);
TestItem* item2 = new TestItem(str2.data(), str2.length());
// k1 should be demoted to NVM
ASSERT_OK(cache->Insert("k2", item2, &LRUSecondaryCacheTest::helper_,
str2.length()));
Cache::Handle* handle;
SetFailCreate(true);
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
cache->Release(handle);
// This lookup should fail, since k1 creation would have failed
handle = cache->Lookup("k1", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_EQ(handle, nullptr);
// Since k1 didn't get promoted, k2 should still be in cache
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
cache->Release(handle);
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 1u);
cache.reset();
secondary_cache.reset();
}
TEST_F(LRUSecondaryCacheTest, FullCapacityTest) {
LRUCacheOptions opts(1024, 0, /*_strict_capacity_limit=*/true, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
Random rnd(301);
std::string str1 = rnd.RandomString(1020);
TestItem* item1 = new TestItem(str1.data(), str1.length());
ASSERT_OK(cache->Insert("k1", item1, &LRUSecondaryCacheTest::helper_,
str1.length()));
std::string str2 = rnd.RandomString(1020);
TestItem* item2 = new TestItem(str2.data(), str2.length());
// k1 should be demoted to NVM
ASSERT_OK(cache->Insert("k2", item2, &LRUSecondaryCacheTest::helper_,
str2.length()));
Cache::Handle* handle;
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
// k1 promotion should fail due to the block cache being at capacity,
// but the lookup should still succeed
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 05:57:51 +00:00
Cache::Handle* handle2;
handle2 = cache->Lookup("k1", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
ASSERT_NE(handle2, nullptr);
// Since k1 didn't get inserted, k2 should still be in cache
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 05:57:51 +00:00
cache->Release(handle);
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
cache->Release(handle2);
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 05:57:51 +00:00
handle = cache->Lookup("k2", &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, true);
ASSERT_NE(handle, nullptr);
cache->Release(handle);
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 1u);
cache.reset();
secondary_cache.reset();
}
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 01:28:28 +00:00
// In this test, the block cache size is set to 4096, after insert 6 KV-pairs
// and flush, there are 5 blocks in this SST file, 2 data blocks and 3 meta
// blocks. block_1 size is 4096 and block_2 size is 2056. The total size
// of the meta blocks are about 900 to 1000. Therefore, in any situation,
// if we try to insert block_1 to the block cache, it will always fails. Only
// block_2 will be successfully inserted into the block cache.
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheCorrectness1) {
LRUCacheOptions opts(4 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
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 01:28:28 +00:00
// Set the file paranoid check, so after flush, the file will be read
// all the blocks will be accessed.
options.paranoid_file_checks = true;
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
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 01:28:28 +00:00
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
// After Flush is successful, RocksDB will do the paranoid check for the new
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 01:28:28 +00:00
// SST file. Meta blocks are always cached in the block cache and they
// will not be evicted. When block_2 is cache miss and read out, it is
// inserted to the block cache. Note that, block_1 is never successfully
// inserted to the block cache. Here are 2 lookups in the secondary cache
// for block_1 and block_2
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Compact("a", "z");
// Compaction will create the iterator to scan the whole file. So all the
// blocks are needed. Meta blocks are always cached. When block_1 is read
// out, block_2 is evicted from block cache and inserted to secondary
// cache.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 3u);
std::string v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// The first data block is not in the cache, similarly, trigger the block
// cache Lookup and secondary cache lookup for block_1. But block_1 will not
// be inserted successfully due to the size. Currently, cache only has
// the meta blocks.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// The second data block is not in the cache, similarly, trigger the block
// cache Lookup and secondary cache lookup for block_2 and block_2 is found
// in the secondary cache. Now block cache has block_2
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// block_2 is in the block cache. There is a block cache hit. No need to
// lookup or insert the secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Lookup the first data block, not in the block cache, so lookup the
// secondary cache. Also not in the secondary cache. After Get, still
// block_1 is will not be cached.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 6u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Lookup the first data block, not in the block cache, so lookup the
// secondary cache. Also not in the secondary cache. After Get, still
// block_1 is will not be cached.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 7u);
Destroy(options);
}
// In this test, the block cache size is set to 6100, after insert 6 KV-pairs
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 01:28:28 +00:00
// and flush, there are 5 blocks in this SST file, 2 data blocks and 3 meta
// blocks. block_1 size is 4096 and block_2 size is 2056. The total size
// of the meta blocks are about 900 to 1000. Therefore, we can successfully
// insert and cache block_1 in the block cache (this is the different place
// from TestSecondaryCacheCorrectness1)
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheCorrectness2) {
LRUCacheOptions opts(6100, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
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 01:28:28 +00:00
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.paranoid_file_checks = true;
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
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 01:28:28 +00:00
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
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 01:28:28 +00:00
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
// After Flush is successful, RocksDB will do the paranoid check for the new
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 01:28:28 +00:00
// SST file. Meta blocks are always cached in the block cache and they
// will not be evicted. When block_2 is cache miss and read out, it is
// inserted to the block cache. Thefore, block_1 is evicted from block
// cache and successfully inserted to the secondary cache. Here are 2
// lookups in the secondary cache for block_1 and block_2.
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Compact("a", "z");
// Compaction will create the iterator to scan the whole file. So all the
// blocks are needed. After Flush, only block_2 is cached in block cache
// and block_1 is in the secondary cache. So when read block_1, it is
// read out from secondary cache and inserted to block cache. At the same
// time, block_2 is inserted to secondary cache. Now, secondary cache has
// both block_1 and block_2. After compaction, block_1 is in the cache.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 3u);
std::string v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// This Get needs to access block_1, since block_1 is cached in block cache
// there is no secondary cache lookup.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 3u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// This Get needs to access block_2 which is not in the block cache. So
// it will lookup the secondary cache for block_2 and cache it in the
// block_cache.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// This Get needs to access block_2 which is already in the block cache.
// No need to lookup secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// This Get needs to access block_1, since block_1 is not in block cache
// there is one econdary cache lookup. Then, block_1 is cached in the
// block cache.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// This Get needs to access block_1, since block_1 is cached in block cache
// there is no secondary cache lookup.
ASSERT_EQ(secondary_cache->num_inserts(), 2u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
Destroy(options);
}
// The block cache size is set to 1024*1024, after insert 6 KV-pairs
// and flush, there are 5 blocks in this SST file, 2 data blocks and 3 meta
// blocks. block_1 size is 4096 and block_2 size is 2056. The total size
// of the meta blocks are about 900 to 1000. Therefore, we can successfully
// cache all the blocks in the block cache and there is not secondary cache
// insertion. 2 lookup is needed for the blocks.
TEST_F(DBSecondaryCacheTest, NoSecondaryCacheInsertion) {
LRUCacheOptions opts(1024 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.paranoid_file_checks = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
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 01:28:28 +00:00
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
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 01:28:28 +00:00
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1000);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
// After Flush is successful, RocksDB will do the paranoid check for the new
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 01:28:28 +00:00
// SST file. Meta blocks are always cached in the block cache and they
// will not be evicted. Now, block cache is large enough, it cache
// both block_1 and block_2. When first time read block_1 and block_2
// there are cache misses. So 2 secondary cache lookups are needed for
// the 2 blocks
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Compact("a", "z");
// Compaction will iterate the whole SST file. Since all the data blocks
// are in the block cache. No need to lookup the secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
std::string v = Get(Key(0));
ASSERT_EQ(1000, v.size());
// Since the block cache is large enough, all the blocks are cached. we
// do not need to lookup the seondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Destroy(options);
}
TEST_F(DBSecondaryCacheTest, SecondaryCacheIntensiveTesting) {
LRUCacheOptions opts(8 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
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 01:28:28 +00:00
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
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 01:28:28 +00:00
Random rnd(301);
const int N = 256;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1000);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
Compact("a", "z");
Random r_index(47);
std::string v;
for (int i = 0; i < 1000; i++) {
uint32_t key_i = r_index.Next() % N;
v = Get(Key(key_i));
}
// We have over 200 data blocks there will be multiple insertion
// and lookups.
ASSERT_GE(secondary_cache->num_inserts(), 1u);
ASSERT_GE(secondary_cache->num_lookups(), 1u);
Destroy(options);
}
// In this test, the block cache size is set to 4096, after insert 6 KV-pairs
// and flush, there are 5 blocks in this SST file, 2 data blocks and 3 meta
// blocks. block_1 size is 4096 and block_2 size is 2056. The total size
// of the meta blocks are about 900 to 1000. Therefore, in any situation,
// if we try to insert block_1 to the block cache, it will always fails. Only
// block_2 will be successfully inserted into the block cache.
TEST_F(DBSecondaryCacheTest, SecondaryCacheFailureTest) {
LRUCacheOptions opts(4 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.paranoid_file_checks = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
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 01:28:28 +00:00
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
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 01:28:28 +00:00
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
// After Flush is successful, RocksDB will do the paranoid check for the new
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 01:28:28 +00:00
// SST file. Meta blocks are always cached in the block cache and they
// will not be evicted. When block_2 is cache miss and read out, it is
// inserted to the block cache. Note that, block_1 is never successfully
// inserted to the block cache. Here are 2 lookups in the secondary cache
// for block_1 and block_2
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
// Fail the insertion, in LRU cache, the secondary insertion returned status
// is not checked, therefore, the DB will not be influenced.
secondary_cache->InjectFailure();
Compact("a", "z");
// Compaction will create the iterator to scan the whole file. So all the
// blocks are needed. Meta blocks are always cached. When block_1 is read
// out, block_2 is evicted from block cache and inserted to secondary
// cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 3u);
std::string v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// The first data block is not in the cache, similarly, trigger the block
// cache Lookup and secondary cache lookup for block_1. But block_1 will not
// be inserted successfully due to the size. Currently, cache only has
// the meta blocks.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// The second data block is not in the cache, similarly, trigger the block
// cache Lookup and secondary cache lookup for block_2 and block_2 is found
// in the secondary cache. Now block cache has block_2
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// block_2 is in the block cache. There is a block cache hit. No need to
// lookup or insert the secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 5u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Lookup the first data block, not in the block cache, so lookup the
// secondary cache. Also not in the secondary cache. After Get, still
// block_1 is will not be cached.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 6u);
v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Lookup the first data block, not in the block cache, so lookup the
// secondary cache. Also not in the secondary cache. After Get, still
// block_1 is will not be cached.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 7u);
secondary_cache->ResetInjectFailure();
Destroy(options);
}
Parallelize secondary cache lookup in MultiGet (#8405) Summary: Implement the ```WaitAll()``` interface in ```LRUCache``` to allow callers to issue multiple lookups in parallel and wait for all of them to complete. Modify ```MultiGet``` to use this to parallelize the secondary cache lookups in order to reduce the overall latency. A call to ```cache->Lookup()``` returns a handle that has an incomplete value (nullptr), and the caller can call ```cache->IsReady()``` to check whether the lookup is complete, and pass a vector of handles to ```WaitAll``` to wait for completion. If any of the lookups fail, ```MultiGet``` will read the block from the SST file. Another change in this PR is to rename ```SecondaryCacheHandle``` to ```SecondaryCacheResultHandle``` as it more accurately describes the return result of the secondary cache lookup, which is more like a future. Tests: 1. Add unit tests in lru_cache_test 2. Benchmark results with no secondary cache configured Master - ``` readrandom : 41.175 micros/op 388562 ops/sec; 106.7 MB/s (7277999 of 7277999 found) readrandom : 41.217 micros/op 388160 ops/sec; 106.6 MB/s (7274999 of 7274999 found) multireadrandom : 10.309 micros/op 1552082 ops/sec; (28908992 of 28908992 found) multireadrandom : 10.321 micros/op 1550218 ops/sec; (29081984 of 29081984 found) ``` This PR - ``` readrandom : 41.158 micros/op 388723 ops/sec; 106.8 MB/s (7290999 of 7290999 found) readrandom : 41.185 micros/op 388463 ops/sec; 106.7 MB/s (7287999 of 7287999 found) multireadrandom : 10.277 micros/op 1556801 ops/sec; (29346944 of 29346944 found) multireadrandom : 10.253 micros/op 1560539 ops/sec; (29274944 of 29274944 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8405 Reviewed By: zhichao-cao Differential Revision: D29190509 Pulled By: anand1976 fbshipit-source-id: 6f8eff6246712af8a297cfe22ea0d1c3b2a01bb0
2021-06-18 16:35:03 +00:00
TEST_F(LRUSecondaryCacheTest, BasicWaitAllTest) {
LRUCacheOptions opts(1024, 2, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(32 * 1024);
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
const int num_keys = 32;
Random rnd(301);
std::vector<std::string> values;
for (int i = 0; i < num_keys; ++i) {
std::string str = rnd.RandomString(1020);
values.emplace_back(str);
TestItem* item = new TestItem(str.data(), str.length());
ASSERT_OK(cache->Insert("k" + std::to_string(i), item,
&LRUSecondaryCacheTest::helper_, str.length()));
}
// Force all entries to be evicted to the secondary cache
cache->SetCapacity(0);
ASSERT_EQ(secondary_cache->num_inserts(), 32u);
cache->SetCapacity(32 * 1024);
secondary_cache->SetResultMap(
{{"k3", TestSecondaryCache::ResultType::DEFER},
{"k4", TestSecondaryCache::ResultType::DEFER_AND_FAIL},
{"k5", TestSecondaryCache::ResultType::FAIL}});
std::vector<Cache::Handle*> results;
for (int i = 0; i < 6; ++i) {
results.emplace_back(
cache->Lookup("k" + std::to_string(i), &LRUSecondaryCacheTest::helper_,
test_item_creator, Cache::Priority::LOW, false));
}
cache->WaitAll(results);
for (int i = 0; i < 6; ++i) {
if (i == 4) {
ASSERT_EQ(cache->Value(results[i]), nullptr);
} else if (i == 5) {
ASSERT_EQ(results[i], nullptr);
continue;
} else {
TestItem* item = static_cast<TestItem*>(cache->Value(results[i]));
ASSERT_EQ(item->ToString(), values[i]);
}
cache->Release(results[i]);
}
cache.reset();
secondary_cache.reset();
}
// In this test, we have one KV pair per data block. We indirectly determine
// the cache key associated with each data block (and thus each KV) by using
// a sync point callback in TestSecondaryCache::Lookup. We then control the
// lookup result by setting the ResultMap.
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheMultiGet) {
LRUCacheOptions opts(1 << 20, 0, false, 0.5, nullptr, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
table_options.cache_index_and_filter_blocks = false;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.paranoid_file_checks = true;
DestroyAndReopen(options);
Random rnd(301);
const int N = 8;
std::vector<std::string> keys;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(4000);
keys.emplace_back(p_v);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
// After Flush is successful, RocksDB does the paranoid check for the new
// SST file. This will try to lookup all data blocks in the secondary
// cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 8u);
cache->SetCapacity(0);
ASSERT_EQ(secondary_cache->num_inserts(), 8u);
cache->SetCapacity(1 << 20);
std::vector<std::string> cache_keys;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"TestSecondaryCache::Lookup", [&cache_keys](void* key) -> void {
cache_keys.emplace_back(*(static_cast<std::string*>(key)));
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
for (int i = 0; i < N; ++i) {
std::string v = Get(Key(i));
ASSERT_EQ(4000, v.size());
ASSERT_EQ(v, keys[i]);
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ASSERT_EQ(secondary_cache->num_lookups(), 16u);
cache->SetCapacity(0);
cache->SetCapacity(1 << 20);
ASSERT_EQ(Get(Key(2)), keys[2]);
ASSERT_EQ(Get(Key(7)), keys[7]);
secondary_cache->SetResultMap(
{{cache_keys[3], TestSecondaryCache::ResultType::DEFER},
{cache_keys[4], TestSecondaryCache::ResultType::DEFER_AND_FAIL},
{cache_keys[5], TestSecondaryCache::ResultType::FAIL}});
std::vector<std::string> mget_keys(
{Key(0), Key(1), Key(2), Key(3), Key(4), Key(5), Key(6), Key(7)});
std::vector<PinnableSlice> values(mget_keys.size());
std::vector<Status> s(keys.size());
std::vector<Slice> key_slices;
for (const std::string& key : mget_keys) {
key_slices.emplace_back(key);
}
uint32_t num_lookups = secondary_cache->num_lookups();
dbfull()->MultiGet(ReadOptions(), dbfull()->DefaultColumnFamily(),
key_slices.size(), key_slices.data(), values.data(),
s.data(), false);
ASSERT_EQ(secondary_cache->num_lookups(), num_lookups + 5);
for (int i = 0; i < N; ++i) {
ASSERT_OK(s[i]);
ASSERT_EQ(values[i].ToString(), keys[i]);
values[i].Reset();
}
Destroy(options);
}
class LRUCacheWithStat : public LRUCache {
public:
LRUCacheWithStat(
size_t _capacity, int _num_shard_bits, bool _strict_capacity_limit,
double _high_pri_pool_ratio,
std::shared_ptr<MemoryAllocator> _memory_allocator = nullptr,
bool _use_adaptive_mutex = kDefaultToAdaptiveMutex,
CacheMetadataChargePolicy _metadata_charge_policy =
kDontChargeCacheMetadata,
const std::shared_ptr<SecondaryCache>& _secondary_cache = nullptr)
: LRUCache(_capacity, _num_shard_bits, _strict_capacity_limit,
_high_pri_pool_ratio, _memory_allocator, _use_adaptive_mutex,
_metadata_charge_policy, _secondary_cache) {
insert_count_ = 0;
lookup_count_ = 0;
}
~LRUCacheWithStat() {}
Status Insert(const Slice& key, void* value, size_t charge, DeleterFn deleter,
Handle** handle, Priority priority) override {
insert_count_++;
return LRUCache::Insert(key, value, charge, deleter, handle, priority);
}
Status Insert(const Slice& key, void* value, const CacheItemHelper* helper,
size_t chargge, Handle** handle = nullptr,
Priority priority = Priority::LOW) override {
insert_count_++;
return LRUCache::Insert(key, value, helper, chargge, handle, priority);
}
Handle* Lookup(const Slice& key, Statistics* stats) override {
lookup_count_++;
return LRUCache::Lookup(key, stats);
}
Handle* Lookup(const Slice& key, const CacheItemHelper* helper,
const CreateCallback& create_cb, Priority priority, bool wait,
Statistics* stats = nullptr) override {
lookup_count_++;
return LRUCache::Lookup(key, helper, create_cb, priority, wait, stats);
}
uint32_t GetInsertCount() { return insert_count_; }
uint32_t GetLookupcount() { return lookup_count_; }
void ResetCount() {
insert_count_ = 0;
lookup_count_ = 0;
}
private:
uint32_t insert_count_;
uint32_t lookup_count_;
};
#ifndef ROCKSDB_LITE
TEST_F(DBSecondaryCacheTest, LRUCacheDumpLoadBasic) {
LRUCacheOptions cache_opts(1024 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
LRUCacheWithStat* tmp_cache = new LRUCacheWithStat(
cache_opts.capacity, cache_opts.num_shard_bits,
cache_opts.strict_capacity_limit, cache_opts.high_pri_pool_ratio,
cache_opts.memory_allocator, cache_opts.use_adaptive_mutex,
cache_opts.metadata_charge_policy, cache_opts.secondary_cache);
std::shared_ptr<Cache> cache(tmp_cache);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
DestroyAndReopen(options);
fault_fs_->SetFailGetUniqueId(true);
Random rnd(301);
const int N = 256;
std::vector<std::string> value;
char buf[1000];
memset(buf, 'a', 1000);
value.resize(N);
for (int i = 0; i < N; i++) {
// std::string p_v = rnd.RandomString(1000);
std::string p_v(buf, 1000);
value[i] = p_v;
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
Compact("a", "z");
// do th eread for all the key value pairs, so all the blocks should be in
// cache
uint32_t start_insert = tmp_cache->GetInsertCount();
uint32_t start_lookup = tmp_cache->GetLookupcount();
std::string v;
for (int i = 0; i < N; i++) {
v = Get(Key(i));
ASSERT_EQ(v, value[i]);
}
uint32_t dump_insert = tmp_cache->GetInsertCount() - start_insert;
uint32_t dump_lookup = tmp_cache->GetLookupcount() - start_lookup;
ASSERT_EQ(63,
static_cast<int>(dump_insert)); // the insert in the block cache
ASSERT_EQ(256,
static_cast<int>(dump_lookup)); // the lookup in the block cache
// We have enough blocks in the block cache
CacheDumpOptions cd_options;
cd_options.clock = fault_env_->GetSystemClock().get();
std::string dump_path = db_->GetName() + "/cache_dump";
std::unique_ptr<CacheDumpWriter> dump_writer;
Status s = NewToFileCacheDumpWriter(fault_fs_, FileOptions(), dump_path,
&dump_writer);
ASSERT_OK(s);
std::unique_ptr<CacheDumper> cache_dumper;
s = NewDefaultCacheDumper(cd_options, cache, std::move(dump_writer),
&cache_dumper);
ASSERT_OK(s);
std::vector<DB*> db_list;
db_list.push_back(db_);
s = cache_dumper->SetDumpFilter(db_list);
ASSERT_OK(s);
s = cache_dumper->DumpCacheEntriesToWriter();
ASSERT_OK(s);
cache_dumper.reset();
// we have a new cache it is empty, then, before we do the Get, we do the
// dumpload
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048 * 1024);
cache_opts.secondary_cache = secondary_cache;
tmp_cache = new LRUCacheWithStat(
cache_opts.capacity, cache_opts.num_shard_bits,
cache_opts.strict_capacity_limit, cache_opts.high_pri_pool_ratio,
cache_opts.memory_allocator, cache_opts.use_adaptive_mutex,
cache_opts.metadata_charge_policy, cache_opts.secondary_cache);
std::shared_ptr<Cache> cache_new(tmp_cache);
table_options.block_cache = cache_new;
table_options.block_size = 4 * 1024;
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
// start to load the data to new block cache
start_insert = secondary_cache->num_inserts();
start_lookup = secondary_cache->num_lookups();
std::unique_ptr<CacheDumpReader> dump_reader;
s = NewFromFileCacheDumpReader(fault_fs_, FileOptions(), dump_path,
&dump_reader);
ASSERT_OK(s);
std::unique_ptr<CacheDumpedLoader> cache_loader;
s = NewDefaultCacheDumpedLoader(cd_options, table_options, secondary_cache,
std::move(dump_reader), &cache_loader);
ASSERT_OK(s);
s = cache_loader->RestoreCacheEntriesToSecondaryCache();
ASSERT_OK(s);
uint32_t load_insert = secondary_cache->num_inserts() - start_insert;
uint32_t load_lookup = secondary_cache->num_lookups() - start_lookup;
// check the number we inserted
ASSERT_EQ(64, static_cast<int>(load_insert));
ASSERT_EQ(0, static_cast<int>(load_lookup));
ASSERT_OK(s);
Reopen(options);
// After load, we do the Get again
start_insert = secondary_cache->num_inserts();
start_lookup = secondary_cache->num_lookups();
uint32_t cache_insert = tmp_cache->GetInsertCount();
uint32_t cache_lookup = tmp_cache->GetLookupcount();
for (int i = 0; i < N; i++) {
v = Get(Key(i));
ASSERT_EQ(v, value[i]);
}
uint32_t final_insert = secondary_cache->num_inserts() - start_insert;
uint32_t final_lookup = secondary_cache->num_lookups() - start_lookup;
// no insert to secondary cache
ASSERT_EQ(0, static_cast<int>(final_insert));
// lookup the secondary to get all blocks
ASSERT_EQ(64, static_cast<int>(final_lookup));
uint32_t block_insert = tmp_cache->GetInsertCount() - cache_insert;
uint32_t block_lookup = tmp_cache->GetLookupcount() - cache_lookup;
// Check the new block cache insert and lookup, should be no insert since all
// blocks are from the secondary cache.
ASSERT_EQ(0, static_cast<int>(block_insert));
ASSERT_EQ(256, static_cast<int>(block_lookup));
fault_fs_->SetFailGetUniqueId(false);
Destroy(options);
}
TEST_F(DBSecondaryCacheTest, LRUCacheDumpLoadWithFilter) {
LRUCacheOptions cache_opts(1024 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
LRUCacheWithStat* tmp_cache = new LRUCacheWithStat(
cache_opts.capacity, cache_opts.num_shard_bits,
cache_opts.strict_capacity_limit, cache_opts.high_pri_pool_ratio,
cache_opts.memory_allocator, cache_opts.use_adaptive_mutex,
cache_opts.metadata_charge_policy, cache_opts.secondary_cache);
std::shared_ptr<Cache> cache(tmp_cache);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
std::string dbname1 = test::PerThreadDBPath("db_1");
ASSERT_OK(DestroyDB(dbname1, options));
DB* db1 = nullptr;
ASSERT_OK(DB::Open(options, dbname1, &db1));
std::string dbname2 = test::PerThreadDBPath("db_2");
ASSERT_OK(DestroyDB(dbname2, options));
DB* db2 = nullptr;
ASSERT_OK(DB::Open(options, dbname2, &db2));
fault_fs_->SetFailGetUniqueId(true);
// write the KVs to db1
Random rnd(301);
const int N = 256;
std::vector<std::string> value1;
WriteOptions wo;
char buf[1000];
memset(buf, 'a', 1000);
value1.resize(N);
for (int i = 0; i < N; i++) {
std::string p_v(buf, 1000);
value1[i] = p_v;
ASSERT_OK(db1->Put(wo, Key(i), p_v));
}
ASSERT_OK(db1->Flush(FlushOptions()));
Slice bg("a");
Slice ed("b");
ASSERT_OK(db1->CompactRange(CompactRangeOptions(), &bg, &ed));
// Write the KVs to DB2
std::vector<std::string> value2;
memset(buf, 'b', 1000);
value2.resize(N);
for (int i = 0; i < N; i++) {
std::string p_v(buf, 1000);
value2[i] = p_v;
ASSERT_OK(db2->Put(wo, Key(i), p_v));
}
ASSERT_OK(db2->Flush(FlushOptions()));
ASSERT_OK(db2->CompactRange(CompactRangeOptions(), &bg, &ed));
// do th eread for all the key value pairs, so all the blocks should be in
// cache
uint32_t start_insert = tmp_cache->GetInsertCount();
uint32_t start_lookup = tmp_cache->GetLookupcount();
ReadOptions ro;
std::string v;
for (int i = 0; i < N; i++) {
ASSERT_OK(db1->Get(ro, Key(i), &v));
ASSERT_EQ(v, value1[i]);
}
for (int i = 0; i < N; i++) {
ASSERT_OK(db2->Get(ro, Key(i), &v));
ASSERT_EQ(v, value2[i]);
}
uint32_t dump_insert = tmp_cache->GetInsertCount() - start_insert;
uint32_t dump_lookup = tmp_cache->GetLookupcount() - start_lookup;
ASSERT_EQ(128,
static_cast<int>(dump_insert)); // the insert in the block cache
ASSERT_EQ(512,
static_cast<int>(dump_lookup)); // the lookup in the block cache
// We have enough blocks in the block cache
CacheDumpOptions cd_options;
cd_options.clock = fault_env_->GetSystemClock().get();
std::string dump_path = db1->GetName() + "/cache_dump";
std::unique_ptr<CacheDumpWriter> dump_writer;
Status s = NewToFileCacheDumpWriter(fault_fs_, FileOptions(), dump_path,
&dump_writer);
ASSERT_OK(s);
std::unique_ptr<CacheDumper> cache_dumper;
s = NewDefaultCacheDumper(cd_options, cache, std::move(dump_writer),
&cache_dumper);
ASSERT_OK(s);
std::vector<DB*> db_list;
db_list.push_back(db1);
s = cache_dumper->SetDumpFilter(db_list);
ASSERT_OK(s);
s = cache_dumper->DumpCacheEntriesToWriter();
ASSERT_OK(s);
cache_dumper.reset();
// we have a new cache it is empty, then, before we do the Get, we do the
// dumpload
std::shared_ptr<TestSecondaryCache> secondary_cache =
std::make_shared<TestSecondaryCache>(2048 * 1024);
cache_opts.secondary_cache = secondary_cache;
tmp_cache = new LRUCacheWithStat(
cache_opts.capacity, cache_opts.num_shard_bits,
cache_opts.strict_capacity_limit, cache_opts.high_pri_pool_ratio,
cache_opts.memory_allocator, cache_opts.use_adaptive_mutex,
cache_opts.metadata_charge_policy, cache_opts.secondary_cache);
std::shared_ptr<Cache> cache_new(tmp_cache);
table_options.block_cache = cache_new;
table_options.block_size = 4 * 1024;
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
// Start the cache loading process
start_insert = secondary_cache->num_inserts();
start_lookup = secondary_cache->num_lookups();
std::unique_ptr<CacheDumpReader> dump_reader;
s = NewFromFileCacheDumpReader(fault_fs_, FileOptions(), dump_path,
&dump_reader);
ASSERT_OK(s);
std::unique_ptr<CacheDumpedLoader> cache_loader;
s = NewDefaultCacheDumpedLoader(cd_options, table_options, secondary_cache,
std::move(dump_reader), &cache_loader);
ASSERT_OK(s);
s = cache_loader->RestoreCacheEntriesToSecondaryCache();
ASSERT_OK(s);
uint32_t load_insert = secondary_cache->num_inserts() - start_insert;
uint32_t load_lookup = secondary_cache->num_lookups() - start_lookup;
// check the number we inserted
ASSERT_EQ(64, static_cast<int>(load_insert));
ASSERT_EQ(0, static_cast<int>(load_lookup));
ASSERT_OK(s);
ASSERT_OK(db1->Close());
delete db1;
ASSERT_OK(DB::Open(options, dbname1, &db1));
// After load, we do the Get again. To validate the cache, we do not allow any
// I/O, so we set the file system to false.
IOStatus error_msg = IOStatus::IOError("Retryable IO Error");
fault_fs_->SetFilesystemActive(false, error_msg);
start_insert = secondary_cache->num_inserts();
start_lookup = secondary_cache->num_lookups();
uint32_t cache_insert = tmp_cache->GetInsertCount();
uint32_t cache_lookup = tmp_cache->GetLookupcount();
for (int i = 0; i < N; i++) {
ASSERT_OK(db1->Get(ro, Key(i), &v));
ASSERT_EQ(v, value1[i]);
}
uint32_t final_insert = secondary_cache->num_inserts() - start_insert;
uint32_t final_lookup = secondary_cache->num_lookups() - start_lookup;
// no insert to secondary cache
ASSERT_EQ(0, static_cast<int>(final_insert));
// lookup the secondary to get all blocks
ASSERT_EQ(64, static_cast<int>(final_lookup));
uint32_t block_insert = tmp_cache->GetInsertCount() - cache_insert;
uint32_t block_lookup = tmp_cache->GetLookupcount() - cache_lookup;
// Check the new block cache insert and lookup, should be no insert since all
// blocks are from the secondary cache.
ASSERT_EQ(0, static_cast<int>(block_insert));
ASSERT_EQ(256, static_cast<int>(block_lookup));
fault_fs_->SetFailGetUniqueId(false);
fault_fs_->SetFilesystemActive(true);
delete db1;
delete db2;
ASSERT_OK(DestroyDB(dbname1, options));
ASSERT_OK(DestroyDB(dbname2, options));
}
// Test the option not to use the secondary cache in a certain DB.
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheOptionBasic) {
LRUCacheOptions opts(4 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
options.lowest_used_cache_tier = CacheTier::kVolatileTier;
// Set the file paranoid check, so after flush, the file will be read
// all the blocks will be accessed.
options.paranoid_file_checks = true;
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i + 70), p_v));
}
ASSERT_OK(Flush());
// Flush will trigger the paranoid check and read blocks. But only block cache
// will be read. No operations for secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
Compact("a", "z");
// Compaction will also insert and evict blocks, no operations to the block
// cache. No operations for secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
std::string v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Check the data in first block. Cache miss, direclty read from SST file.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// Check the second block.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// block cache hit
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(70));
ASSERT_EQ(1007, v.size());
// Check the first block in the second SST file. Cache miss and trigger SST
// file read. No operations for secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(75));
ASSERT_EQ(1007, v.size());
// Check the second block in the second SST file. Cache miss and trigger SST
// file read. No operations for secondary cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
Destroy(options);
}
// We disable the secondary cache in DBOptions at first. Close and reopen the DB
// with new options, which set the lowest_used_cache_tier to
// kNonVolatileBlockTier. So secondary cache will be used.
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheOptionChange) {
LRUCacheOptions opts(4 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
fault_fs_->SetFailGetUniqueId(true);
options.lowest_used_cache_tier = CacheTier::kVolatileTier;
// Set the file paranoid check, so after flush, the file will be read
// all the blocks will be accessed.
options.paranoid_file_checks = true;
DestroyAndReopen(options);
std::string session_id;
ASSERT_OK(db_->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i), p_v));
}
ASSERT_OK(Flush());
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(Put(Key(i + 70), p_v));
}
ASSERT_OK(Flush());
// Flush will trigger the paranoid check and read blocks. But only block cache
// will be read.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
Compact("a", "z");
// Compaction will also insert and evict blocks, no operations to the block
// cache.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
std::string v = Get(Key(0));
ASSERT_EQ(1007, v.size());
// Check the data in first block. Cache miss, direclty read from SST file.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// Check the second block.
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
v = Get(Key(5));
ASSERT_EQ(1007, v.size());
// block cache hit
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
// Change the option to enable secondary cache after we Reopen the DB
options.lowest_used_cache_tier = CacheTier::kNonVolatileBlockTier;
Reopen(options);
v = Get(Key(70));
ASSERT_EQ(1007, v.size());
// Enable the secondary cache, trigger lookup of the first block in second SST
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 1u);
v = Get(Key(75));
ASSERT_EQ(1007, v.size());
// trigger lookup of the second block in second SST
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Destroy(options);
}
// Two DB test. We create 2 DBs sharing the same block cache and secondary
// cache. We diable the secondary cache option for DB2.
TEST_F(DBSecondaryCacheTest, TestSecondaryCacheOptionTwoDB) {
LRUCacheOptions opts(4 * 1024, 0, false, 0.5, nullptr,
kDefaultToAdaptiveMutex, kDontChargeCacheMetadata);
std::shared_ptr<TestSecondaryCache> secondary_cache(
new TestSecondaryCache(2048 * 1024));
opts.secondary_cache = secondary_cache;
std::shared_ptr<Cache> cache = NewLRUCache(opts);
BlockBasedTableOptions table_options;
table_options.block_cache = cache;
table_options.block_size = 4 * 1024;
Options options = GetDefaultOptions();
options.create_if_missing = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.env = fault_env_.get();
options.paranoid_file_checks = true;
std::string dbname1 = test::PerThreadDBPath("db_t_1");
ASSERT_OK(DestroyDB(dbname1, options));
DB* db1 = nullptr;
ASSERT_OK(DB::Open(options, dbname1, &db1));
std::string dbname2 = test::PerThreadDBPath("db_t_2");
ASSERT_OK(DestroyDB(dbname2, options));
DB* db2 = nullptr;
Options options2 = options;
options2.lowest_used_cache_tier = CacheTier::kVolatileTier;
ASSERT_OK(DB::Open(options2, dbname2, &db2));
fault_fs_->SetFailGetUniqueId(true);
// Set the file paranoid check, so after flush, the file will be read
// all the blocks will be accessed.
std::string session_id;
ASSERT_OK(db1->GetDbSessionId(session_id));
secondary_cache->SetDbSessionId(session_id);
WriteOptions wo;
Random rnd(301);
const int N = 6;
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(db1->Put(wo, Key(i), p_v));
}
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 0u);
ASSERT_OK(db1->Flush(FlushOptions()));
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
for (int i = 0; i < N; i++) {
std::string p_v = rnd.RandomString(1007);
ASSERT_OK(db2->Put(wo, Key(i), p_v));
}
// No change in the secondary cache, since it is disabled in DB2
ASSERT_EQ(secondary_cache->num_inserts(), 0u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
ASSERT_OK(db2->Flush(FlushOptions()));
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
Slice bg("a");
Slice ed("b");
ASSERT_OK(db1->CompactRange(CompactRangeOptions(), &bg, &ed));
ASSERT_OK(db2->CompactRange(CompactRangeOptions(), &bg, &ed));
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 2u);
ReadOptions ro;
std::string v;
ASSERT_OK(db1->Get(ro, Key(0), &v));
ASSERT_EQ(1007, v.size());
// DB 1 has lookup block 1 and it is miss in block cache, trigger secondary
// cache lookup
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 3u);
ASSERT_OK(db1->Get(ro, Key(5), &v));
ASSERT_EQ(1007, v.size());
// DB 1 lookup the second block and it is miss in block cache, trigger
// secondary cache lookup
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
ASSERT_OK(db2->Get(ro, Key(0), &v));
ASSERT_EQ(1007, v.size());
// For db2, it is not enabled with secondary cache, so no search in the
// secondary cache
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
ASSERT_OK(db2->Get(ro, Key(5), &v));
ASSERT_EQ(1007, v.size());
// For db2, it is not enabled with secondary cache, so no search in the
// secondary cache
ASSERT_EQ(secondary_cache->num_inserts(), 1u);
ASSERT_EQ(secondary_cache->num_lookups(), 4u);
fault_fs_->SetFailGetUniqueId(false);
fault_fs_->SetFilesystemActive(true);
delete db1;
delete db2;
ASSERT_OK(DestroyDB(dbname1, options));
ASSERT_OK(DestroyDB(dbname2, options));
}
#endif // ROCKSDB_LITE
} // namespace ROCKSDB_NAMESPACE
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}