You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/cache/lru_cache.cc

800 lines
25 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "cache/lru_cache.h"
#include <cassert>
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
#include <cstdint>
#include <cstdio>
#include "monitoring/perf_context_imp.h"
#include "monitoring/statistics.h"
#include "util/mutexlock.h"
namespace ROCKSDB_NAMESPACE {
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
LRUHandleTable::LRUHandleTable(int max_upper_hash_bits)
: length_bits_(/* historical starting size*/ 4),
list_(new LRUHandle* [size_t{1} << length_bits_] {}),
elems_(0),
max_length_bits_(max_upper_hash_bits) {}
LRUHandleTable::~LRUHandleTable() {
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
ApplyToEntriesRange(
[](LRUHandle* h) {
if (!h->HasRefs()) {
h->Free();
}
},
0, uint32_t{1} << length_bits_);
}
LRUHandle* LRUHandleTable::Lookup(const Slice& key, uint32_t hash) {
return *FindPointer(key, hash);
}
LRUHandle* LRUHandleTable::Insert(LRUHandle* h) {
LRUHandle** ptr = FindPointer(h->key(), h->hash);
LRUHandle* old = *ptr;
h->next_hash = (old == nullptr ? nullptr : old->next_hash);
*ptr = h;
if (old == nullptr) {
++elems_;
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
if ((elems_ >> length_bits_) > 0) { // elems_ >= length
// Since each cache entry is fairly large, we aim for a small
// average linked list length (<= 1).
Resize();
}
}
return old;
}
LRUHandle* LRUHandleTable::Remove(const Slice& key, uint32_t hash) {
LRUHandle** ptr = FindPointer(key, hash);
LRUHandle* result = *ptr;
if (result != nullptr) {
*ptr = result->next_hash;
--elems_;
}
return result;
}
LRUHandle** LRUHandleTable::FindPointer(const Slice& key, uint32_t hash) {
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
LRUHandle** ptr = &list_[hash >> (32 - length_bits_)];
while (*ptr != nullptr && ((*ptr)->hash != hash || key != (*ptr)->key())) {
ptr = &(*ptr)->next_hash;
}
return ptr;
}
void LRUHandleTable::Resize() {
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
if (length_bits_ >= max_length_bits_) {
// Due to reaching limit of hash information, if we made the table
// bigger, we would allocate more addresses but only the same
// number would be used.
return;
}
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
if (length_bits_ >= 31) {
// Avoid undefined behavior shifting uint32_t by 32
return;
}
uint32_t old_length = uint32_t{1} << length_bits_;
int new_length_bits = length_bits_ + 1;
std::unique_ptr<LRUHandle* []> new_list {
new LRUHandle* [size_t{1} << new_length_bits] {}
};
uint32_t count = 0;
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
for (uint32_t i = 0; i < old_length; i++) {
LRUHandle* h = list_[i];
while (h != nullptr) {
LRUHandle* next = h->next_hash;
uint32_t hash = h->hash;
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
LRUHandle** ptr = &new_list[hash >> (32 - new_length_bits)];
h->next_hash = *ptr;
*ptr = h;
h = next;
count++;
}
}
assert(elems_ == count);
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
list_ = std::move(new_list);
length_bits_ = new_length_bits;
}
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
4 years ago
LRUCacheShard::LRUCacheShard(
size_t capacity, bool strict_capacity_limit, double high_pri_pool_ratio,
bool use_adaptive_mutex, CacheMetadataChargePolicy metadata_charge_policy,
int max_upper_hash_bits,
const std::shared_ptr<SecondaryCache>& secondary_cache)
: capacity_(0),
high_pri_pool_usage_(0),
strict_capacity_limit_(strict_capacity_limit),
high_pri_pool_ratio_(high_pri_pool_ratio),
high_pri_pool_capacity_(0),
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
table_(max_upper_hash_bits),
usage_(0),
lru_usage_(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
4 years ago
mutex_(use_adaptive_mutex),
secondary_cache_(secondary_cache) {
set_metadata_charge_policy(metadata_charge_policy);
// Make empty circular linked list
lru_.next = &lru_;
lru_.prev = &lru_;
lru_low_pri_ = &lru_;
SetCapacity(capacity);
}
void LRUCacheShard::EraseUnRefEntries() {
autovector<LRUHandle*> last_reference_list;
{
MutexLock l(&mutex_);
while (lru_.next != &lru_) {
LRUHandle* old = lru_.next;
// LRU list contains only elements which can be evicted
assert(old->InCache() && !old->HasRefs());
LRU_Remove(old);
table_.Remove(old->key(), old->hash);
old->SetInCache(false);
size_t total_charge = old->CalcTotalCharge(metadata_charge_policy_);
assert(usage_ >= total_charge);
usage_ -= total_charge;
last_reference_list.push_back(old);
}
}
for (auto entry : last_reference_list) {
entry->Free();
}
}
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
void LRUCacheShard::ApplyToSomeEntries(
const std::function<void(const Slice& key, void* value, size_t charge,
DeleterFn deleter)>& callback,
uint32_t average_entries_per_lock, uint32_t* state) {
// The state is essentially going to be the starting hash, which works
// nicely even if we resize between calls because we use upper-most
// hash bits for table indexes.
MutexLock l(&mutex_);
uint32_t length_bits = table_.GetLengthBits();
uint32_t length = uint32_t{1} << length_bits;
assert(average_entries_per_lock > 0);
// Assuming we are called with same average_entries_per_lock repeatedly,
// this simplifies some logic (index_end will not overflow)
assert(average_entries_per_lock < length || *state == 0);
uint32_t index_begin = *state >> (32 - length_bits);
uint32_t index_end = index_begin + average_entries_per_lock;
if (index_end >= length) {
// Going to end
index_end = length;
*state = UINT32_MAX;
} else {
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
*state = index_end << (32 - length_bits);
}
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
table_.ApplyToEntriesRange(
[callback](LRUHandle* 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
4 years ago
DeleterFn deleter = h->IsSecondaryCacheCompatible()
? h->info_.helper->del_cb
: h->info_.deleter;
callback(h->key(), h->value, h->charge, deleter);
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
},
index_begin, index_end);
}
void LRUCacheShard::TEST_GetLRUList(LRUHandle** lru, LRUHandle** lru_low_pri) {
MutexLock l(&mutex_);
*lru = &lru_;
*lru_low_pri = lru_low_pri_;
}
size_t LRUCacheShard::TEST_GetLRUSize() {
MutexLock l(&mutex_);
LRUHandle* lru_handle = lru_.next;
size_t lru_size = 0;
while (lru_handle != &lru_) {
lru_size++;
lru_handle = lru_handle->next;
}
return lru_size;
}
double LRUCacheShard::GetHighPriPoolRatio() {
MutexLock l(&mutex_);
return high_pri_pool_ratio_;
}
void LRUCacheShard::LRU_Remove(LRUHandle* e) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
assert(e->next != nullptr);
assert(e->prev != nullptr);
if (lru_low_pri_ == e) {
lru_low_pri_ = e->prev;
}
e->next->prev = e->prev;
e->prev->next = e->next;
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
e->prev = e->next = nullptr;
size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
assert(lru_usage_ >= total_charge);
lru_usage_ -= total_charge;
if (e->InHighPriPool()) {
assert(high_pri_pool_usage_ >= total_charge);
high_pri_pool_usage_ -= total_charge;
}
}
void LRUCacheShard::LRU_Insert(LRUHandle* e) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
assert(e->next == nullptr);
assert(e->prev == nullptr);
size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
if (high_pri_pool_ratio_ > 0 && (e->IsHighPri() || e->HasHit())) {
// Inset "e" to head of LRU list.
e->next = &lru_;
e->prev = lru_.prev;
e->prev->next = e;
e->next->prev = e;
e->SetInHighPriPool(true);
high_pri_pool_usage_ += total_charge;
MaintainPoolSize();
} else {
// Insert "e" to the head of low-pri pool. Note that when
// high_pri_pool_ratio is 0, head of low-pri pool is also head of LRU list.
e->next = lru_low_pri_->next;
e->prev = lru_low_pri_;
e->prev->next = e;
e->next->prev = e;
e->SetInHighPriPool(false);
lru_low_pri_ = e;
}
lru_usage_ += total_charge;
}
void LRUCacheShard::MaintainPoolSize() {
while (high_pri_pool_usage_ > high_pri_pool_capacity_) {
// Overflow last entry in high-pri pool to low-pri pool.
lru_low_pri_ = lru_low_pri_->next;
assert(lru_low_pri_ != &lru_);
lru_low_pri_->SetInHighPriPool(false);
size_t total_charge =
lru_low_pri_->CalcTotalCharge(metadata_charge_policy_);
assert(high_pri_pool_usage_ >= total_charge);
high_pri_pool_usage_ -= total_charge;
}
}
void LRUCacheShard::EvictFromLRU(size_t charge,
autovector<LRUHandle*>* deleted) {
while ((usage_ + charge) > capacity_ && lru_.next != &lru_) {
LRUHandle* old = lru_.next;
// LRU list contains only elements which can be evicted
assert(old->InCache() && !old->HasRefs());
LRU_Remove(old);
table_.Remove(old->key(), old->hash);
old->SetInCache(false);
size_t old_total_charge = old->CalcTotalCharge(metadata_charge_policy_);
assert(usage_ >= old_total_charge);
usage_ -= old_total_charge;
deleted->push_back(old);
}
}
void LRUCacheShard::SetCapacity(size_t capacity) {
autovector<LRUHandle*> last_reference_list;
{
MutexLock l(&mutex_);
capacity_ = capacity;
high_pri_pool_capacity_ = capacity_ * high_pri_pool_ratio_;
EvictFromLRU(0, &last_reference_list);
}
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
4 years ago
// Try to insert the evicted entries into tiered cache
// Free the entries outside of mutex for performance reasons
for (auto entry : last_reference_list) {
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
4 years ago
if (secondary_cache_ && entry->IsSecondaryCacheCompatible() &&
!entry->IsPromoted()) {
secondary_cache_->Insert(entry->key(), entry->value, entry->info_.helper)
.PermitUncheckedError();
}
entry->Free();
}
}
void LRUCacheShard::SetStrictCapacityLimit(bool strict_capacity_limit) {
MutexLock l(&mutex_);
strict_capacity_limit_ = strict_capacity_limit;
}
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
4 years ago
Status LRUCacheShard::InsertItem(LRUHandle* e, Cache::Handle** handle,
bool free_handle_on_fail) {
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
4 years ago
Status s = Status::OK();
autovector<LRUHandle*> last_reference_list;
size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
{
MutexLock l(&mutex_);
// Free the space following strict LRU policy until enough space
// is freed or the lru list is empty
EvictFromLRU(total_charge, &last_reference_list);
if ((usage_ + total_charge) > capacity_ &&
(strict_capacity_limit_ || 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
4 years ago
e->SetInCache(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
4 years ago
if (handle == nullptr) {
// Don't insert the entry but still return ok, as if the entry inserted
// into cache and get evicted immediately.
last_reference_list.push_back(e);
} else {
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
4 years ago
if (free_handle_on_fail) {
delete[] reinterpret_cast<char*>(e);
*handle = nullptr;
}
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
4 years ago
s = Status::Incomplete("Insert failed due to LRU cache being full.");
}
} else {
// Insert into the cache. Note that the cache might get larger than its
// capacity if not enough space was freed up.
LRUHandle* old = table_.Insert(e);
usage_ += total_charge;
if (old != nullptr) {
s = Status::OkOverwritten();
assert(old->InCache());
old->SetInCache(false);
if (!old->HasRefs()) {
// old is on LRU because it's in cache and its reference count is 0
LRU_Remove(old);
size_t old_total_charge =
old->CalcTotalCharge(metadata_charge_policy_);
assert(usage_ >= old_total_charge);
usage_ -= old_total_charge;
last_reference_list.push_back(old);
}
}
if (handle == nullptr) {
LRU_Insert(e);
} else {
// If caller already holds a ref, no need to take one here
if (!e->HasRefs()) {
e->Ref();
}
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
4 years ago
*handle = reinterpret_cast<Cache::Handle*>(e);
}
}
}
// Try to insert the evicted entries into the secondary cache
// Free the entries here outside of mutex for performance reasons
for (auto entry : last_reference_list) {
if (secondary_cache_ && entry->IsSecondaryCacheCompatible() &&
!entry->IsPromoted()) {
secondary_cache_->Insert(entry->key(), entry->value, entry->info_.helper)
.PermitUncheckedError();
}
entry->Free();
}
return s;
}
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
4 years ago
void LRUCacheShard::Promote(LRUHandle* e) {
SecondaryCacheResultHandle* secondary_handle = e->sec_handle;
assert(secondary_handle->IsReady());
e->SetIncomplete(false);
e->SetInCache(true);
e->SetPromoted(true);
e->value = secondary_handle->Value();
e->charge = secondary_handle->Size();
delete secondary_handle;
// This call could fail if the cache is over capacity and
// strict_capacity_limit_ is true. In such a case, we don't want
// InsertItem() to free the handle, since the item is already in memory
// and the caller will most likely just read from disk if we erase it here.
if (e->value) {
Cache::Handle* handle = reinterpret_cast<Cache::Handle*>(e);
Status s = InsertItem(e, &handle, /*free_handle_on_fail=*/false);
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
4 years ago
// Item is in memory, but not accounted against the cache capacity.
// When the handle is released, the item should get deleted
assert(!e->InCache());
}
} else {
// Since the secondary cache lookup failed, mark the item as not in cache
// Don't charge the cache as its only metadata that'll shortly be released
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
4 years ago
MutexLock l(&mutex_);
e->charge = 0;
e->SetInCache(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
4 years ago
Cache::Handle* LRUCacheShard::Lookup(
const Slice& key, uint32_t hash,
const ShardedCache::CacheItemHelper* helper,
const ShardedCache::CreateCallback& create_cb, Cache::Priority priority,
bool wait, Statistics* stats) {
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
4 years ago
LRUHandle* e = nullptr;
{
MutexLock l(&mutex_);
e = table_.Lookup(key, hash);
if (e != nullptr) {
assert(e->InCache());
if (!e->HasRefs()) {
// The entry is in LRU since it's in hash and has no external references
LRU_Remove(e);
}
e->Ref();
e->SetHit();
}
}
// If handle table lookup failed, then allocate a handle outside the
// mutex if we're going to lookup in the secondary cache
// Only support synchronous for now
// TODO: Support asynchronous lookup in secondary cache
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
4 years ago
if (!e && secondary_cache_ && helper && helper->saveto_cb) {
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
4 years ago
// For objects from the secondary cache, we expect the caller to provide
// a way to create/delete the primary cache object. The only case where
// a deleter would not be required is for dummy entries inserted for
// accounting purposes, which we won't demote to the secondary cache
// anyway.
assert(create_cb && helper->del_cb);
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
4 years ago
std::unique_ptr<SecondaryCacheResultHandle> 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
4 years ago
secondary_cache_->Lookup(key, create_cb, wait);
if (secondary_handle != nullptr) {
e = reinterpret_cast<LRUHandle*>(
new char[sizeof(LRUHandle) - 1 + key.size()]);
e->flags = 0;
e->SetSecondaryCacheCompatible(true);
e->info_.helper = helper;
e->key_length = key.size();
e->hash = hash;
e->refs = 0;
e->next = e->prev = nullptr;
e->SetPriority(priority);
memcpy(e->key_data, key.data(), key.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
4 years ago
e->value = nullptr;
e->sec_handle = secondary_handle.release();
e->Ref();
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
4 years ago
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
4 years ago
if (wait) {
Promote(e);
if (!e->value) {
// The secondary cache returned a handle, but the lookup failed
e->Unref();
e->Free();
e = nullptr;
} else {
PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
RecordTick(stats, SECONDARY_CACHE_HITS);
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
4 years ago
}
} else {
// If wait is false, we always return a handle and let the caller
// release the handle after checking for success or failure
e->SetIncomplete(true);
// This may be slightly inaccurate, if the lookup eventually fails.
// But the probability is very low.
PERF_COUNTER_ADD(secondary_cache_hit_count, 1);
RecordTick(stats, SECONDARY_CACHE_HITS);
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
4 years ago
}
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
}
}
return reinterpret_cast<Cache::Handle*>(e);
}
bool LRUCacheShard::Ref(Cache::Handle* h) {
LRUHandle* e = reinterpret_cast<LRUHandle*>(h);
MutexLock l(&mutex_);
// To create another reference - entry must be already externally referenced
assert(e->HasRefs());
e->Ref();
return true;
}
void LRUCacheShard::SetHighPriorityPoolRatio(double high_pri_pool_ratio) {
MutexLock l(&mutex_);
high_pri_pool_ratio_ = high_pri_pool_ratio;
high_pri_pool_capacity_ = capacity_ * high_pri_pool_ratio_;
MaintainPoolSize();
}
bool LRUCacheShard::Release(Cache::Handle* handle, bool force_erase) {
if (handle == nullptr) {
return false;
}
LRUHandle* e = reinterpret_cast<LRUHandle*>(handle);
bool last_reference = false;
{
MutexLock l(&mutex_);
last_reference = e->Unref();
if (last_reference && e->InCache()) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
// The item is still in cache, and nobody else holds a reference to it
if (usage_ > capacity_ || force_erase) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
// The LRU list must be empty since the cache is full
assert(lru_.next == &lru_ || force_erase);
// Take this opportunity and remove the item
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
table_.Remove(e->key(), e->hash);
e->SetInCache(false);
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
} else {
// Put the item back on the LRU list, and don't free it
LRU_Insert(e);
last_reference = false;
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
}
}
// If it was the last reference, and the entry is either not secondary
// cache compatible (i.e a dummy entry for accounting), or is secondary
// cache compatible and has a non-null value, then decrement the cache
// usage. If value is null in the latter case, taht means the lookup
// failed and we didn't charge the cache.
if (last_reference && (!e->IsSecondaryCacheCompatible() || e->value)) {
size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
assert(usage_ >= total_charge);
usage_ -= total_charge;
}
}
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
// Free the entry here outside of mutex for performance reasons
if (last_reference) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
e->Free();
}
return last_reference;
}
Status LRUCacheShard::Insert(const Slice& key, uint32_t hash, void* value,
size_t charge,
void (*deleter)(const Slice& key, void* value),
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
4 years ago
const Cache::CacheItemHelper* helper,
Cache::Handle** handle, Cache::Priority priority) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
// Allocate the memory here outside of the mutex
// If the cache is full, we'll have to release it
// It shouldn't happen very often though.
LRUHandle* e = reinterpret_cast<LRUHandle*>(
add simulator Cache as class SimCache/SimLRUCache(with test) Summary: add class SimCache(base class with instrumentation api) and SimLRUCache(derived class with detailed implementation) which is used as an instrumented block cache that can predict hit rate for different cache size Test Plan: Add a test case in `db_block_cache_test.cc` called `SimCacheTest` to test basic logic of SimCache. Also add option `-simcache_size` in db_bench. if set with a value other than -1, then the benchmark will use this value as the size of the simulator cache and finally output the simulation result. ``` [gzh@dev9927.prn1 ~/local/rocksdb] ./db_bench -benchmarks "fillseq,readrandom" -cache_size 1000000 -simcache_size 1000000 RocksDB: version 4.8 Date: Tue May 17 16:56:16 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 0 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 6.809 micros/op 146874 ops/sec; 16.2 MB/s DB path: [/tmp/rocksdbtest-112628/dbbench] readrandom : 6.343 micros/op 157665 ops/sec; 17.4 MB/s (1000000 of 1000000 found) SIMULATOR CACHE STATISTICS: SimCache LOOKUPs: 986559 SimCache HITs: 264760 SimCache HITRATE: 26.84% [gzh@dev9927.prn1 ~/local/rocksdb] ./db_bench -benchmarks "fillseq,readrandom" -cache_size 1000000 -simcache_size 10000000 RocksDB: version 4.8 Date: Tue May 17 16:57:10 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 0 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 5.066 micros/op 197394 ops/sec; 21.8 MB/s DB path: [/tmp/rocksdbtest-112628/dbbench] readrandom : 6.457 micros/op 154870 ops/sec; 17.1 MB/s (1000000 of 1000000 found) SIMULATOR CACHE STATISTICS: SimCache LOOKUPs: 1059764 SimCache HITs: 374501 SimCache HITRATE: 35.34% [gzh@dev9927.prn1 ~/local/rocksdb] ./db_bench -benchmarks "fillseq,readrandom" -cache_size 1000000 -simcache_size 100000000 RocksDB: version 4.8 Date: Tue May 17 16:57:32 2016 CPU: 32 * Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz CPUCache: 20480 KB Keys: 16 bytes each Values: 100 bytes each (50 bytes after compression) Entries: 1000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 110.6 MB (estimated) FileSize: 62.9 MB (estimated) Write rate: 0 bytes/second Compression: Snappy Memtablerep: skip_list Perf Level: 0 WARNING: Assertions are enabled; benchmarks unnecessarily slow ------------------------------------------------ DB path: [/tmp/rocksdbtest-112628/dbbench] fillseq : 5.632 micros/op 177572 ops/sec; 19.6 MB/s DB path: [/tmp/rocksdbtest-112628/dbbench] readrandom : 6.892 micros/op 145094 ops/sec; 16.1 MB/s (1000000 of 1000000 found) SIMULATOR CACHE STATISTICS: SimCache LOOKUPs: 1150767 SimCache HITs: 1034535 SimCache HITRATE: 89.90% ``` Reviewers: IslamAbdelRahman, andrewkr, sdong Reviewed By: sdong Subscribers: MarkCallaghan, andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D57999
9 years ago
new char[sizeof(LRUHandle) - 1 + key.size()]);
e->value = value;
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
4 years ago
e->flags = 0;
if (helper) {
e->SetSecondaryCacheCompatible(true);
e->info_.helper = helper;
} else {
#ifdef __SANITIZE_THREAD__
e->is_secondary_cache_compatible_for_tsan = false;
#endif // __SANITIZE_THREAD__
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
4 years ago
e->info_.deleter = deleter;
}
e->charge = charge;
e->key_length = key.size();
e->hash = hash;
e->refs = 0;
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
e->next = e->prev = nullptr;
e->SetInCache(true);
e->SetPriority(priority);
memcpy(e->key_data, key.data(), key.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
4 years ago
return InsertItem(e, handle, /* free_handle_on_fail */ true);
}
void LRUCacheShard::Erase(const Slice& key, uint32_t hash) {
LRUHandle* e;
bool last_reference = false;
{
MutexLock l(&mutex_);
e = table_.Remove(key, hash);
if (e != nullptr) {
assert(e->InCache());
e->SetInCache(false);
if (!e->HasRefs()) {
// The entry is in LRU since it's in hash and has no external references
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
LRU_Remove(e);
size_t total_charge = e->CalcTotalCharge(metadata_charge_policy_);
assert(usage_ >= total_charge);
usage_ -= total_charge;
last_reference = true;
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
}
}
}
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
// Free the entry here outside of mutex for performance reasons
// last_reference will only be true if e != nullptr
if (last_reference) {
Modifed the LRU cache eviction code so that it doesn't evict blocks which have exteranl references Summary: Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense: - blocks are still in RAM, so the RAM usage reported by the cache is incorrect - if the same block is needed by another iterator, it will be loaded and decompressed again This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache. This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache. Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this. table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved Test Plan: Ran tests, made sure they pass Reviewers: sdong, ljin Differential Revision: https://reviews.facebook.net/D25503
10 years ago
e->Free();
}
}
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
4 years ago
bool LRUCacheShard::IsReady(Cache::Handle* handle) {
LRUHandle* e = reinterpret_cast<LRUHandle*>(handle);
MutexLock l(&mutex_);
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
4 years ago
bool ready = true;
if (e->IsPending()) {
assert(secondary_cache_);
assert(e->sec_handle);
ready = e->sec_handle->IsReady();
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
4 years ago
}
return ready;
}
size_t LRUCacheShard::GetUsage() const {
MutexLock l(&mutex_);
return usage_;
}
size_t LRUCacheShard::GetPinnedUsage() const {
MutexLock l(&mutex_);
assert(usage_ >= lru_usage_);
return usage_ - lru_usage_;
}
std::string LRUCacheShard::GetPrintableOptions() const {
const int kBufferSize = 200;
char buffer[kBufferSize];
{
MutexLock l(&mutex_);
snprintf(buffer, kBufferSize, " high_pri_pool_ratio: %.3lf\n",
high_pri_pool_ratio_);
}
return std::string(buffer);
}
LRUCache::LRUCache(size_t capacity, int num_shard_bits,
bool strict_capacity_limit, double high_pri_pool_ratio,
std::shared_ptr<MemoryAllocator> allocator,
bool use_adaptive_mutex,
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
4 years ago
CacheMetadataChargePolicy metadata_charge_policy,
const std::shared_ptr<SecondaryCache>& secondary_cache)
: ShardedCache(capacity, num_shard_bits, strict_capacity_limit,
std::move(allocator)) {
num_shards_ = 1 << num_shard_bits;
shards_ = reinterpret_cast<LRUCacheShard*>(
port::cacheline_aligned_alloc(sizeof(LRUCacheShard) * num_shards_));
size_t per_shard = (capacity + (num_shards_ - 1)) / num_shards_;
for (int i = 0; i < num_shards_; i++) {
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
4 years ago
new (&shards_[i]) LRUCacheShard(
per_shard, strict_capacity_limit, high_pri_pool_ratio,
use_adaptive_mutex, metadata_charge_policy,
/* max_upper_hash_bits */ 32 - num_shard_bits, secondary_cache);
}
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
4 years ago
secondary_cache_ = secondary_cache;
}
LRUCache::~LRUCache() {
if (shards_ != nullptr) {
assert(num_shards_ > 0);
for (int i = 0; i < num_shards_; i++) {
shards_[i].~LRUCacheShard();
}
port::cacheline_aligned_free(shards_);
}
}
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
CacheShard* LRUCache::GetShard(uint32_t shard) {
return reinterpret_cast<CacheShard*>(&shards_[shard]);
}
New Cache API for gathering statistics (#8225) Summary: Adds a new Cache::ApplyToAllEntries API that we expect to use (in follow-up PRs) for efficiently gathering block cache statistics. Notable features vs. old ApplyToAllCacheEntries: * Includes key and deleter (in addition to value and charge). We could have passed in a Handle but then more virtual function calls would be needed to get the "fields" of each entry. We expect to use the 'deleter' to identify the origin of entries, perhaps even more. * Heavily tuned to minimize latency impact on operating cache. It does this by iterating over small sections of each cache shard while cycling through the shards. * Supports tuning roughly how many entries to operate on for each lock acquire and release, to control the impact on the latency of other operations without excessive lock acquire & release. The right balance can depend on the cost of the callback. Good default seems to be around 256. * There should be no need to disable thread safety. (I would expect uncontended locks to be sufficiently fast.) I have enhanced cache_bench to validate this approach: * Reports a histogram of ns per operation, so we can look at the ditribution of times, not just throughput (average). * Can add a thread for simulated "gather stats" which calls ApplyToAllEntries at a specified interval. We also generate a histogram of time to run ApplyToAllEntries. To make the iteration over some entries of each shard work as cleanly as possible, even with resize between next set of entries, I have re-arranged which hash bits are used for sharding and which for indexing within a shard. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8225 Test Plan: A couple of unit tests are added, but primary validation is manual, as the primary risk is to performance. The primary validation is using cache_bench to ensure that neither the minor hashing changes nor the simulated stats gathering significantly impact QPS or latency distribution. Note that adding op latency histogram seriously impacts the benchmark QPS, so for a fair baseline, we need the cache_bench changes (except remove simulated stat gathering to make it compile). In short, we don't see any reproducible difference in ops/sec or op latency unless we are gathering stats nearly continuously. Test uses 10GB block cache with 8KB values to be somewhat realistic in the number of items to iterate over. Baseline typical output: ``` Complete in 92.017 s; Rough parallel ops/sec = 869401 Thread ops/sec = 54662 Operation latency (ns): Count: 80000000 Average: 11223.9494 StdDev: 29.61 Min: 0 Median: 7759.3973 Max: 9620500 Percentiles: P50: 7759.40 P75: 14190.73 P99: 46922.75 P99.9: 77509.84 P99.99: 217030.58 ------------------------------------------------------ [ 0, 1 ] 68 0.000% 0.000% ( 2900, 4400 ] 89 0.000% 0.000% ( 4400, 6600 ] 33630240 42.038% 42.038% ######## ( 6600, 9900 ] 18129842 22.662% 64.700% ##### ( 9900, 14000 ] 7877533 9.847% 74.547% ## ( 14000, 22000 ] 15193238 18.992% 93.539% #### ( 22000, 33000 ] 3037061 3.796% 97.335% # ( 33000, 50000 ] 1626316 2.033% 99.368% ( 50000, 75000 ] 421532 0.527% 99.895% ( 75000, 110000 ] 56910 0.071% 99.966% ( 110000, 170000 ] 16134 0.020% 99.986% ( 170000, 250000 ] 5166 0.006% 99.993% ( 250000, 380000 ] 3017 0.004% 99.996% ( 380000, 570000 ] 1337 0.002% 99.998% ( 570000, 860000 ] 805 0.001% 99.999% ( 860000, 1200000 ] 319 0.000% 100.000% ( 1200000, 1900000 ] 231 0.000% 100.000% ( 1900000, 2900000 ] 100 0.000% 100.000% ( 2900000, 4300000 ] 39 0.000% 100.000% ( 4300000, 6500000 ] 16 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ``` New, gather_stats=false. Median thread ops/sec of 5 runs: ``` Complete in 92.030 s; Rough parallel ops/sec = 869285 Thread ops/sec = 54458 Operation latency (ns): Count: 80000000 Average: 11298.1027 StdDev: 42.18 Min: 0 Median: 7722.0822 Max: 6398720 Percentiles: P50: 7722.08 P75: 14294.68 P99: 47522.95 P99.9: 85292.16 P99.99: 228077.78 ------------------------------------------------------ [ 0, 1 ] 109 0.000% 0.000% ( 2900, 4400 ] 793 0.001% 0.001% ( 4400, 6600 ] 34054563 42.568% 42.569% ######### ( 6600, 9900 ] 17482646 21.853% 64.423% #### ( 9900, 14000 ] 7908180 9.885% 74.308% ## ( 14000, 22000 ] 15032072 18.790% 93.098% #### ( 22000, 33000 ] 3237834 4.047% 97.145% # ( 33000, 50000 ] 1736882 2.171% 99.316% ( 50000, 75000 ] 446851 0.559% 99.875% ( 75000, 110000 ] 68251 0.085% 99.960% ( 110000, 170000 ] 18592 0.023% 99.983% ( 170000, 250000 ] 7200 0.009% 99.992% ( 250000, 380000 ] 3334 0.004% 99.997% ( 380000, 570000 ] 1393 0.002% 99.998% ( 570000, 860000 ] 700 0.001% 99.999% ( 860000, 1200000 ] 293 0.000% 100.000% ( 1200000, 1900000 ] 196 0.000% 100.000% ( 1900000, 2900000 ] 69 0.000% 100.000% ( 2900000, 4300000 ] 32 0.000% 100.000% ( 4300000, 6500000 ] 10 0.000% 100.000% ``` New, gather_stats=true, 1 second delay between scans. Scans take about 1 second here so it's spending about 50% time scanning. Still the effect on ops/sec and latency seems to be in the noise. Median thread ops/sec of 5 runs: ``` Complete in 91.890 s; Rough parallel ops/sec = 870608 Thread ops/sec = 54551 Operation latency (ns): Count: 80000000 Average: 11311.2629 StdDev: 45.28 Min: 0 Median: 7686.5458 Max: 10018340 Percentiles: P50: 7686.55 P75: 14481.95 P99: 47232.60 P99.9: 79230.18 P99.99: 232998.86 ------------------------------------------------------ [ 0, 1 ] 71 0.000% 0.000% ( 2900, 4400 ] 291 0.000% 0.000% ( 4400, 6600 ] 34492060 43.115% 43.116% ######### ( 6600, 9900 ] 16727328 20.909% 64.025% #### ( 9900, 14000 ] 7845828 9.807% 73.832% ## ( 14000, 22000 ] 15510654 19.388% 93.220% #### ( 22000, 33000 ] 3216533 4.021% 97.241% # ( 33000, 50000 ] 1680859 2.101% 99.342% ( 50000, 75000 ] 439059 0.549% 99.891% ( 75000, 110000 ] 60540 0.076% 99.967% ( 110000, 170000 ] 14649 0.018% 99.985% ( 170000, 250000 ] 5242 0.007% 99.991% ( 250000, 380000 ] 3260 0.004% 99.995% ( 380000, 570000 ] 1599 0.002% 99.997% ( 570000, 860000 ] 1043 0.001% 99.999% ( 860000, 1200000 ] 471 0.001% 99.999% ( 1200000, 1900000 ] 275 0.000% 100.000% ( 1900000, 2900000 ] 143 0.000% 100.000% ( 2900000, 4300000 ] 60 0.000% 100.000% ( 4300000, 6500000 ] 27 0.000% 100.000% ( 6500000, 9800000 ] 7 0.000% 100.000% ( 9800000, 14000000 ] 1 0.000% 100.000% Gather stats latency (us): Count: 46 Average: 980387.5870 StdDev: 60911.18 Min: 879155 Median: 1033777.7778 Max: 1261431 Percentiles: P50: 1033777.78 P75: 1120666.67 P99: 1261431.00 P99.9: 1261431.00 P99.99: 1261431.00 ------------------------------------------------------ ( 860000, 1200000 ] 45 97.826% 97.826% #################### ( 1200000, 1900000 ] 1 2.174% 100.000% Most recent cache entry stats: Number of entries: 1295133 Total charge: 9.88 GB Average key size: 23.4982 Average charge: 8.00 KB Unique deleters: 3 ``` Reviewed By: mrambacher Differential Revision: D28295742 Pulled By: pdillinger fbshipit-source-id: bbc4a552f91ba0fe10e5cc025c42cef5a81f2b95
4 years ago
const CacheShard* LRUCache::GetShard(uint32_t shard) const {
return reinterpret_cast<CacheShard*>(&shards_[shard]);
}
void* LRUCache::Value(Handle* handle) {
return reinterpret_cast<const LRUHandle*>(handle)->value;
}
size_t LRUCache::GetCharge(Handle* handle) const {
return reinterpret_cast<const LRUHandle*>(handle)->charge;
}
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
4 years ago
Cache::DeleterFn LRUCache::GetDeleter(Handle* handle) const {
auto h = reinterpret_cast<const LRUHandle*>(handle);
if (h->IsSecondaryCacheCompatible()) {
return h->info_.helper->del_cb;
} else {
return h->info_.deleter;
}
}
uint32_t LRUCache::GetHash(Handle* handle) const {
return reinterpret_cast<const LRUHandle*>(handle)->hash;
}
void LRUCache::DisownData() {
// Do not drop data if compile with ASAN to suppress leak warning.
Use deleters to label cache entries and collect stats (#8297) Summary: This change gathers and publishes statistics about the kinds of items in block cache. This is especially important for profiling relative usage of cache by index vs. filter vs. data blocks. It works by iterating over the cache during periodic stats dump (InternalStats, stats_dump_period_sec) or on demand when DB::Get(Map)Property(kBlockCacheEntryStats), except that for efficiency and sharing among column families, saved data from the last scan is used when the data is not considered too old. The new information can be seen in info LOG, for example: Block cache LRUCache@0x7fca62229330 capacity: 95.37 MB collections: 8 last_copies: 0 last_secs: 0.00178 secs_since: 0 Block cache entry stats(count,size,portion): DataBlock(7092,28.24 MB,29.6136%) FilterBlock(215,867.90 KB,0.888728%) FilterMetaBlock(2,5.31 KB,0.00544%) IndexBlock(217,180.11 KB,0.184432%) WriteBuffer(1,256.00 KB,0.262144%) Misc(1,0.00 KB,0%) And also through DB::GetProperty and GetMapProperty (here using ldb just for demonstration): $ ./ldb --db=/dev/shm/dbbench/ get_property rocksdb.block-cache-entry-stats rocksdb.block-cache-entry-stats.bytes.data-block: 0 rocksdb.block-cache-entry-stats.bytes.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-block: 0 rocksdb.block-cache-entry-stats.bytes.filter-meta-block: 0 rocksdb.block-cache-entry-stats.bytes.index-block: 178992 rocksdb.block-cache-entry-stats.bytes.misc: 0 rocksdb.block-cache-entry-stats.bytes.other-block: 0 rocksdb.block-cache-entry-stats.bytes.write-buffer: 0 rocksdb.block-cache-entry-stats.capacity: 8388608 rocksdb.block-cache-entry-stats.count.data-block: 0 rocksdb.block-cache-entry-stats.count.deprecated-filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-block: 0 rocksdb.block-cache-entry-stats.count.filter-meta-block: 0 rocksdb.block-cache-entry-stats.count.index-block: 215 rocksdb.block-cache-entry-stats.count.misc: 1 rocksdb.block-cache-entry-stats.count.other-block: 0 rocksdb.block-cache-entry-stats.count.write-buffer: 0 rocksdb.block-cache-entry-stats.id: LRUCache@0x7f3636661290 rocksdb.block-cache-entry-stats.percent.data-block: 0.000000 rocksdb.block-cache-entry-stats.percent.deprecated-filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-block: 0.000000 rocksdb.block-cache-entry-stats.percent.filter-meta-block: 0.000000 rocksdb.block-cache-entry-stats.percent.index-block: 2.133751 rocksdb.block-cache-entry-stats.percent.misc: 0.000000 rocksdb.block-cache-entry-stats.percent.other-block: 0.000000 rocksdb.block-cache-entry-stats.percent.write-buffer: 0.000000 rocksdb.block-cache-entry-stats.secs_for_last_collection: 0.000052 rocksdb.block-cache-entry-stats.secs_since_last_collection: 0 Solution detail - We need some way to flag what kind of blocks each entry belongs to, preferably without changing the Cache API. One of the complications is that Cache is a general interface that could have other users that don't adhere to whichever convention we decide on for keys and values. Or we would pay for an extra field in the Handle that would only be used for this purpose. This change uses a back-door approach, the deleter, to indicate the "role" of a Cache entry (in addition to the value type, implicitly). This has the added benefit of ensuring proper code origin whenever we recognize a particular role for a cache entry; if the entry came from some other part of the code, it will use an unrecognized deleter, which we simply attribute to the "Misc" role. An internal API makes for simple instantiation and automatic registration of Cache deleters for a given value type and "role". Another internal API, CacheEntryStatsCollector, solves the problem of caching the results of a scan and sharing them, to ensure scans are neither excessive nor redundant so as not to harm Cache performance. Because code is added to BlocklikeTraits, it is pulled out of block_based_table_reader.cc into its own file. This is a reformulation of https://github.com/facebook/rocksdb/issues/8276, without the type checking option (could still be added), and with actual stat gathering. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8297 Test Plan: manual testing with db_bench, and a couple of basic unit tests Reviewed By: ltamasi Differential Revision: D28488721 Pulled By: pdillinger fbshipit-source-id: 472f524a9691b5afb107934be2d41d84f2b129fb
4 years ago
#ifndef MUST_FREE_HEAP_ALLOCATIONS
shards_ = nullptr;
num_shards_ = 0;
#endif
}
size_t LRUCache::TEST_GetLRUSize() {
size_t lru_size_of_all_shards = 0;
for (int i = 0; i < num_shards_; i++) {
lru_size_of_all_shards += shards_[i].TEST_GetLRUSize();
}
return lru_size_of_all_shards;
}
double LRUCache::GetHighPriPoolRatio() {
double result = 0.0;
if (num_shards_ > 0) {
result = shards_[0].GetHighPriPoolRatio();
}
return result;
}
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
4 years ago
void LRUCache::WaitAll(std::vector<Handle*>& handles) {
if (secondary_cache_) {
std::vector<SecondaryCacheResultHandle*> sec_handles;
sec_handles.reserve(handles.size());
for (Handle* handle : handles) {
if (!handle) {
continue;
}
LRUHandle* lru_handle = reinterpret_cast<LRUHandle*>(handle);
if (!lru_handle->IsPending()) {
continue;
}
sec_handles.emplace_back(lru_handle->sec_handle);
}
secondary_cache_->WaitAll(sec_handles);
for (Handle* handle : handles) {
if (!handle) {
continue;
}
LRUHandle* lru_handle = reinterpret_cast<LRUHandle*>(handle);
if (!lru_handle->IsPending()) {
continue;
}
uint32_t hash = GetHash(handle);
LRUCacheShard* shard = static_cast<LRUCacheShard*>(GetShard(Shard(hash)));
shard->Promote(lru_handle);
}
}
}
std::shared_ptr<Cache> NewLRUCache(
size_t capacity, int num_shard_bits, bool strict_capacity_limit,
double high_pri_pool_ratio,
std::shared_ptr<MemoryAllocator> memory_allocator, bool use_adaptive_mutex,
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
4 years ago
CacheMetadataChargePolicy metadata_charge_policy,
const std::shared_ptr<SecondaryCache>& secondary_cache) {
if (num_shard_bits >= 20) {
return nullptr; // the cache cannot be sharded into too many fine pieces
}
if (high_pri_pool_ratio < 0.0 || high_pri_pool_ratio > 1.0) {
// invalid high_pri_pool_ratio
return nullptr;
}
if (num_shard_bits < 0) {
num_shard_bits = GetDefaultCacheShardBits(capacity);
}
return std::make_shared<LRUCache>(
capacity, num_shard_bits, strict_capacity_limit, high_pri_pool_ratio,
Initial support for secondary cache in LRUCache (#8271) Summary: Defined the abstract interface for a secondary cache in include/rocksdb/secondary_cache.h, and updated LRUCacheOptions to take a std::shared_ptr<SecondaryCache>. An item is initially inserted into the LRU (primary) cache. When it ages out and evicted from memory, its inserted into the secondary cache. On a LRU cache miss and successful lookup in the secondary cache, the item is promoted to the LRU cache. Only support synchronous lookup currently. The secondary cache would be used to implement a persistent (flash cache) or compressed cache. Tests: Results from cache_bench and db_bench don't show any regression due to these changes. cache_bench results before and after this change - Command ```./cache_bench -ops_per_thread=10000000 -threads=1``` Before ```Complete in 40.688 s; QPS = 245774``` ```Complete in 40.486 s; QPS = 246996``` ```Complete in 42.019 s; QPS = 237989``` After ```Complete in 40.672 s; QPS = 245869``` ```Complete in 44.622 s; QPS = 224107``` ```Complete in 42.445 s; QPS = 235599``` db_bench results before this change, and with this change + https://github.com/facebook/rocksdb/issues/8213 and https://github.com/facebook/rocksdb/issues/8191 - Commands ```./db_bench --benchmarks="fillseq,compact" -num=30000000 -key_size=32 -value_size=256 -use_direct_io_for_flush_and_compaction=true -db=/home/anand76/nvm_cache/db -partition_index_and_filters=true``` ```./db_bench -db=/home/anand76/nvm_cache/db -use_existing_db=true -benchmarks=readrandom -num=30000000 -key_size=32 -value_size=256 -use_direct_reads=true -cache_size=1073741824 -cache_numshardbits=6 -cache_index_and_filter_blocks=true -read_random_exp_range=17 -statistics -partition_index_and_filters=true -threads=16 -duration=300``` Before ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 80.702 micros/op 198104 ops/sec; 54.4 MB/s (3708999 of 3708999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 87.124 micros/op 183625 ops/sec; 50.4 MB/s (3439999 of 3439999 found) ``` After ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 77.653 micros/op 206025 ops/sec; 56.6 MB/s (3866999 of 3866999 found) ``` ``` DB path: [/home/anand76/nvm_cache/db] readrandom : 84.962 micros/op 188299 ops/sec; 51.7 MB/s (3535999 of 3535999 found) ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8271 Reviewed By: zhichao-cao Differential Revision: D28357511 Pulled By: anand1976 fbshipit-source-id: d1cfa236f00e649a18c53328be10a8062a4b6da2
4 years ago
std::move(memory_allocator), use_adaptive_mutex, metadata_charge_policy,
secondary_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
4 years ago
std::shared_ptr<Cache> NewLRUCache(const LRUCacheOptions& cache_opts) {
return NewLRUCache(
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> NewLRUCache(
size_t capacity, int num_shard_bits, bool strict_capacity_limit,
double high_pri_pool_ratio,
std::shared_ptr<MemoryAllocator> memory_allocator, bool use_adaptive_mutex,
CacheMetadataChargePolicy metadata_charge_policy) {
return NewLRUCache(capacity, num_shard_bits, strict_capacity_limit,
high_pri_pool_ratio, memory_allocator, use_adaptive_mutex,
metadata_charge_policy, nullptr);
}
} // namespace ROCKSDB_NAMESPACE