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/utilities/cache_dump_load_impl.cc

474 lines
18 KiB

// Copyright (c) Facebook, Inc. and its affiliates. 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).
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
3 years ago
#include "cache/cache_key.h"
#include "table/block_based/block_based_table_reader.h"
#ifndef ROCKSDB_LITE
#include "utilities/cache_dump_load_impl.h"
#include "cache/cache_entry_roles.h"
#include "file/writable_file_writer.h"
#include "port/lang.h"
#include "rocksdb/env.h"
#include "rocksdb/file_system.h"
#include "rocksdb/utilities/ldb_cmd.h"
#include "table/format.h"
#include "util/crc32c.h"
namespace ROCKSDB_NAMESPACE {
// Set the dump filter with a list of DBs. Block cache may be shared by multipe
// DBs and we may only want to dump out the blocks belonging to certain DB(s).
// Therefore, a filter is need to decide if the key of the block satisfy the
// requirement.
Status CacheDumperImpl::SetDumpFilter(std::vector<DB*> db_list) {
Status s = Status::OK();
for (size_t i = 0; i < db_list.size(); i++) {
assert(i < db_list.size());
TablePropertiesCollection ptc;
assert(db_list[i] != nullptr);
s = db_list[i]->GetPropertiesOfAllTables(&ptc);
if (!s.ok()) {
return s;
}
for (auto id = ptc.begin(); id != ptc.end(); id++) {
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
3 years ago
OffsetableCacheKey base;
// We only want to save cache entries that are portable to another
// DB::Open, so only save entries with stable keys.
bool is_stable;
Derive cache keys from SST unique IDs (#10394) Summary: ... so that cache keys can be derived from DB manifest data before reading the file from storage--so that every part of the file can potentially go in a persistent cache. See updated comments in cache_key.cc for technical details. Importantly, the new cache key encoding uses some fancy but efficient math to pack data into the cache key without depending on the sizes of the various pieces. This simplifies some existing code creating cache keys, like cache warming before the file size is known. This should provide us an essentially permanent mapping between SST unique IDs and base cache keys, with the ability to "upgrade" SST unique IDs (and thus cache keys) with new SST format_versions. These cache keys are of similar, perhaps indistinguishable quality to the previous generation. Before this change (see "corrected" days between collision): ``` ./cache_bench -stress_cache_key -sck_keep_bits=43 18 collisions after 2 x 90 days, est 10 days between (1.15292e+19 corrected) ``` After this change (keep 43 bits, up through 50, to validate "trajectory" is ok on "corrected" days between collision): ``` 19 collisions after 3 x 90 days, est 14.2105 days between (1.63836e+19 corrected) 16 collisions after 5 x 90 days, est 28.125 days between (1.6213e+19 corrected) 15 collisions after 7 x 90 days, est 42 days between (1.21057e+19 corrected) 15 collisions after 17 x 90 days, est 102 days between (1.46997e+19 corrected) 15 collisions after 49 x 90 days, est 294 days between (2.11849e+19 corrected) 15 collisions after 62 x 90 days, est 372 days between (1.34027e+19 corrected) 15 collisions after 53 x 90 days, est 318 days between (5.72858e+18 corrected) 15 collisions after 309 x 90 days, est 1854 days between (1.66994e+19 corrected) ``` However, the change does modify (probably weaken) the "guaranteed unique" promise from this > SST files generated in a single process are guaranteed to have unique cache keys, unless/until number session ids * max file number = 2**86 to this (see https://github.com/facebook/rocksdb/issues/10388) > With the DB id limitation, we only have nice guaranteed unique cache keys for files generated in a single process until biggest session_id_counter and offset_in_file reach combined 64 bits I don't think this is a practical concern, though. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10394 Test Plan: unit tests updated, see simulation results above Reviewed By: jay-zhuang Differential Revision: D38667529 Pulled By: pdillinger fbshipit-source-id: 49af3fe7f47e5b61162809a78b76c769fd519fba
2 years ago
BlockBasedTable::SetupBaseCacheKey(id->second.get(),
/*cur_db_session_id*/ "",
/*cur_file_num*/ 0, &base, &is_stable);
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
3 years ago
if (is_stable) {
Slice prefix_slice = base.CommonPrefixSlice();
assert(prefix_slice.size() == OffsetableCacheKey::kCommonPrefixSize);
prefix_filter_.insert(prefix_slice.ToString());
}
}
}
return s;
}
// This is the main function to dump out the cache block entries to the writer.
// The writer may create a file or write to other systems. Currently, we will
// iterate the whole block cache, get the blocks, and write them to the writer
IOStatus CacheDumperImpl::DumpCacheEntriesToWriter() {
// Prepare stage, check the parameters.
if (cache_ == nullptr) {
return IOStatus::InvalidArgument("Cache is null");
}
if (writer_ == nullptr) {
return IOStatus::InvalidArgument("CacheDumpWriter is null");
}
// Set the system clock
if (options_.clock == nullptr) {
return IOStatus::InvalidArgument("System clock is null");
}
clock_ = options_.clock;
// We copy the Cache Deleter Role Map as its member.
role_map_ = CopyCacheDeleterRoleMap();
// Set the sequence number
sequence_num_ = 0;
// Dump stage, first, we write the hader
IOStatus io_s = WriteHeader();
if (!io_s.ok()) {
return io_s;
}
// Then, we iterate the block cache and dump out the blocks that are not
// filtered out.
cache_->ApplyToAllEntries(DumpOneBlockCallBack(), {});
// Finally, write the footer
io_s = WriteFooter();
if (!io_s.ok()) {
return io_s;
}
io_s = writer_->Close();
return io_s;
}
// Check if we need to filter out the block based on its key
bool CacheDumperImpl::ShouldFilterOut(const Slice& key) {
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
3 years ago
if (key.size() < OffsetableCacheKey::kCommonPrefixSize) {
return /*filter out*/ true;
}
New stable, fixed-length cache keys (#9126) Summary: This change standardizes on a new 16-byte cache key format for block cache (incl compressed and secondary) and persistent cache (but not table cache and row cache). The goal is a really fast cache key with practically ideal stability and uniqueness properties without external dependencies (e.g. from FileSystem). A fixed key size of 16 bytes should enable future optimizations to the concurrent hash table for block cache, which is a heavy CPU user / bottleneck, but there appears to be measurable performance improvement even with no changes to LRUCache. This change replaces a lot of disjointed and ugly code handling cache keys with calls to a simple, clean new internal API (cache_key.h). (Preserving the old cache key logic under an option would be very ugly and likely negate the performance gain of the new approach. Complete replacement carries some inherent risk, but I think that's acceptable with sufficient analysis and testing.) The scheme for encoding new cache keys is complicated but explained in cache_key.cc. Also: EndianSwapValue is moved to math.h to be next to other bit operations. (Explains some new include "math.h".) ReverseBits operation added and unit tests added to hash_test for both. Fixes https://github.com/facebook/rocksdb/issues/7405 (presuming a root cause) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9126 Test Plan: ### Basic correctness Several tests needed updates to work with the new functionality, mostly because we are no longer relying on filesystem for stable cache keys so table builders & readers need more context info to agree on cache keys. This functionality is so core, a huge number of existing tests exercise the cache key functionality. ### Performance Create db with `TEST_TMPDIR=/dev/shm ./db_bench -bloom_bits=10 -benchmarks=fillrandom -num=3000000 -partition_index_and_filters` And test performance with `TEST_TMPDIR=/dev/shm ./db_bench -readonly -use_existing_db -bloom_bits=10 -benchmarks=readrandom -num=3000000 -duration=30 -cache_index_and_filter_blocks -cache_size=250000 -threads=4` using DEBUG_LEVEL=0 and simultaneous before & after runs. Before ops/sec, avg over 100 runs: 121924 After ops/sec, avg over 100 runs: 125385 (+2.8%) ### Collision probability I have built a tool, ./cache_bench -stress_cache_key to broadly simulate host-wide cache activity over many months, by making some pessimistic simplifying assumptions: * Every generated file has a cache entry for every byte offset in the file (contiguous range of cache keys) * All of every file is cached for its entire lifetime We use a simple table with skewed address assignment and replacement on address collision to simulate files coming & going, with quite a variance (super-Poisson) in ages. Some output with `./cache_bench -stress_cache_key -sck_keep_bits=40`: ``` Total cache or DBs size: 32TiB Writing 925.926 MiB/s or 76.2939TiB/day Multiply by 9.22337e+18 to correct for simulation losses (but still assume whole file cached) ``` These come from default settings of 2.5M files per day of 32 MB each, and `-sck_keep_bits=40` means that to represent a single file, we are only keeping 40 bits of the 128-bit cache key. With file size of 2\*\*25 contiguous keys (pessimistic), our simulation is about 2\*\*(128-40-25) or about 9 billion billion times more prone to collision than reality. More default assumptions, relatively pessimistic: * 100 DBs in same process (doesn't matter much) * Re-open DB in same process (new session ID related to old session ID) on average every 100 files generated * Restart process (all new session IDs unrelated to old) 24 times per day After enough data, we get a result at the end: ``` (keep 40 bits) 17 collisions after 2 x 90 days, est 10.5882 days between (9.76592e+19 corrected) ``` If we believe the (pessimistic) simulation and the mathematical generalization, we would need to run a billion machines all for 97 billion days to expect a cache key collision. To help verify that our generalization ("corrected") is robust, we can make our simulation more precise with `-sck_keep_bits=41` and `42`, which takes more running time to get enough data: ``` (keep 41 bits) 16 collisions after 4 x 90 days, est 22.5 days between (1.03763e+20 corrected) (keep 42 bits) 19 collisions after 10 x 90 days, est 47.3684 days between (1.09224e+20 corrected) ``` The generalized prediction still holds. With the `-sck_randomize` option, we can see that we are beating "random" cache keys (except offsets still non-randomized) by a modest amount (roughly 20x less collision prone than random), which should make us reasonably comfortable even in "degenerate" cases: ``` 197 collisions after 1 x 90 days, est 0.456853 days between (4.21372e+18 corrected) ``` I've run other tests to validate other conditions behave as expected, never behaving "worse than random" unless we start chopping off structured data. Reviewed By: zhichao-cao Differential Revision: D33171746 Pulled By: pdillinger fbshipit-source-id: f16a57e369ed37be5e7e33525ace848d0537c88f
3 years ago
Slice key_prefix(key.data(), OffsetableCacheKey::kCommonPrefixSize);
std::string prefix = key_prefix.ToString();
// Filter out if not found
return prefix_filter_.find(prefix) == prefix_filter_.end();
}
// This is the callback function which will be applied to
// Cache::ApplyToAllEntries. In this callback function, we will get the block
// type, decide if the block needs to be dumped based on the filter, and write
// the block through the provided writer.
std::function<void(const Slice&, void*, size_t, Cache::DeleterFn)>
CacheDumperImpl::DumpOneBlockCallBack() {
return [&](const Slice& key, void* value, size_t /*charge*/,
Cache::DeleterFn deleter) {
// Step 1: get the type of the block from role_map_
auto e = role_map_.find(deleter);
CacheEntryRole role;
CacheDumpUnitType type = CacheDumpUnitType::kBlockTypeMax;
if (e == role_map_.end()) {
role = CacheEntryRole::kMisc;
} else {
role = e->second;
}
bool filter_out = false;
// Step 2: based on the key prefix, check if the block should be filter out.
if (ShouldFilterOut(key)) {
filter_out = true;
}
// Step 3: based on the block type, get the block raw pointer and length.
const char* block_start = nullptr;
size_t block_len = 0;
switch (role) {
case CacheEntryRole::kDataBlock:
type = CacheDumpUnitType::kData;
block_start = (static_cast<Block*>(value))->data();
block_len = (static_cast<Block*>(value))->size();
break;
case CacheEntryRole::kFilterBlock:
type = CacheDumpUnitType::kFilter;
block_start = (static_cast<ParsedFullFilterBlock*>(value))
->GetBlockContentsData()
.data();
block_len = (static_cast<ParsedFullFilterBlock*>(value))
->GetBlockContentsData()
.size();
break;
case CacheEntryRole::kFilterMetaBlock:
type = CacheDumpUnitType::kFilterMetaBlock;
block_start = (static_cast<Block*>(value))->data();
block_len = (static_cast<Block*>(value))->size();
break;
case CacheEntryRole::kIndexBlock:
type = CacheDumpUnitType::kIndex;
block_start = (static_cast<Block*>(value))->data();
block_len = (static_cast<Block*>(value))->size();
break;
Remove deprecated block-based filter (#10184) Summary: In https://github.com/facebook/rocksdb/issues/9535, release 7.0, we hid the old block-based filter from being created using the public API, because of its inefficiency. Although we normally maintain read compatibility on old DBs forever, filters are not required for reading a DB, only for optimizing read performance. Thus, it should be acceptable to remove this code and the substantial maintenance burden it carries as useful features are developed and validated (such as user timestamp). This change completely removes the code for reading and writing the old block-based filters, net removing about 1370 lines of code no longer needed. Options removed from testing / benchmarking tools. The prior existence is only evident in a couple of places: * `CacheEntryRole::kDeprecatedFilterBlock` - We can update this public API enum in a major release to minimize source code incompatibilities. * A warning is logged when an old table file is opened that used the old block-based filter. This is provided as a courtesy, and would be a pain to unit test, so manual testing should suffice. Unfortunately, sst_dump does not tell you whether a file uses block-based filter, and the structure of the code makes it very difficult to fix. * To detect that case, `kObsoleteFilterBlockPrefix` (renamed from `kFilterBlockPrefix`) for metaindex is maintained (for now). Other notes: * In some cases where numbers are associated with filter configurations, we have had to update the assigned numbers so that they all correspond to something that exists. * Fixed potential stat counting bug by assuming `filter_checked = false` for cases like `filter == nullptr` rather than assuming `filter_checked = true` * Removed obsolete `block_offset` and `prefix_extractor` parameters from several functions. * Removed some unnecessary checks `if (!table_prefix_extractor() && !prefix_extractor)` because the caller guarantees the prefix extractor exists and is compatible Pull Request resolved: https://github.com/facebook/rocksdb/pull/10184 Test Plan: tests updated, manually test new warning in LOG using base version to generate a DB Reviewed By: riversand963 Differential Revision: D37212647 Pulled By: pdillinger fbshipit-source-id: 06ee020d8de3b81260ffc36ad0c1202cbf463a80
2 years ago
case CacheEntryRole::kDeprecatedFilterBlock:
// Obsolete
filter_out = true;
break;
case CacheEntryRole::kMisc:
filter_out = true;
break;
case CacheEntryRole::kOtherBlock:
filter_out = true;
break;
case CacheEntryRole::kWriteBuffer:
filter_out = true;
break;
default:
filter_out = true;
}
// Step 4: if the block should not be filter out, write the block to the
// CacheDumpWriter
if (!filter_out && block_start != nullptr) {
char* buffer = new char[block_len];
memcpy(buffer, block_start, block_len);
WriteCacheBlock(type, key, (void*)buffer, block_len)
.PermitUncheckedError();
delete[] buffer;
}
};
}
// Write the raw block to the writer. It takes the timestamp of the block being
// copied from block cache, block type, key, block pointer, raw block size and
// the block checksum as the input. When writing the raw block, we first create
// the dump unit and encoude it to a string. Then, we calculate the checksum of
// the how dump unit string and store it in the dump unit metadata.
// First, we write the metadata first, which is a fixed size string. Then, we
// Append the dump unit string to the writer.
IOStatus CacheDumperImpl::WriteRawBlock(uint64_t timestamp,
CacheDumpUnitType type,
const Slice& key, void* value,
size_t len, uint32_t checksum) {
// First, serilize the block information in a string
DumpUnit dump_unit;
dump_unit.timestamp = timestamp;
dump_unit.key = key;
dump_unit.type = type;
dump_unit.value_len = len;
dump_unit.value = value;
dump_unit.value_checksum = checksum;
std::string encoded_data;
CacheDumperHelper::EncodeDumpUnit(dump_unit, &encoded_data);
// Second, create the metadata, which contains a sequence number, the dump
// unit string checksum and the string size. The sequence number monotonically
// increases from 0.
DumpUnitMeta unit_meta;
unit_meta.sequence_num = sequence_num_;
sequence_num_++;
unit_meta.dump_unit_checksum =
crc32c::Value(encoded_data.c_str(), encoded_data.size());
unit_meta.dump_unit_size = static_cast<uint64_t>(encoded_data.size());
std::string encoded_meta;
CacheDumperHelper::EncodeDumpUnitMeta(unit_meta, &encoded_meta);
// We write the metadata first.
assert(writer_ != nullptr);
IOStatus io_s = writer_->WriteMetadata(Slice(encoded_meta));
if (!io_s.ok()) {
return io_s;
}
// followed by the dump unit.
return writer_->WritePacket(Slice(encoded_data));
}
// Before we write any block, we write the header first to store the cache dump
// format version, rocksdb version, and brief intro.
IOStatus CacheDumperImpl::WriteHeader() {
std::string header_key = "header";
std::ostringstream s;
s << kTraceMagic << "\t"
<< "Cache dump format version: " << kCacheDumpMajorVersion << "."
<< kCacheDumpMinorVersion << "\t"
<< "RocksDB Version: " << kMajorVersion << "." << kMinorVersion << "\t"
<< "Format: dump_unit_metadata <sequence_number, dump_unit_checksum, "
"dump_unit_size>, dump_unit <timestamp, key, block_type, "
"block_size, raw_block, raw_block_checksum> cache_value\n";
std::string header_value(s.str());
CacheDumpUnitType type = CacheDumpUnitType::kHeader;
uint64_t timestamp = clock_->NowMicros();
uint32_t header_checksum =
crc32c::Value(header_value.c_str(), header_value.size());
return WriteRawBlock(timestamp, type, Slice(header_key),
(void*)header_value.c_str(), header_value.size(),
header_checksum);
}
// Write the block dumped from cache
IOStatus CacheDumperImpl::WriteCacheBlock(const CacheDumpUnitType type,
const Slice& key, void* value,
size_t len) {
uint64_t timestamp = clock_->NowMicros();
uint32_t value_checksum = crc32c::Value((char*)value, len);
return WriteRawBlock(timestamp, type, key, value, len, value_checksum);
}
// Write the footer after all the blocks are stored to indicate the ending.
IOStatus CacheDumperImpl::WriteFooter() {
std::string footer_key = "footer";
std::ostringstream s;
std::string footer_value("cache dump completed");
CacheDumpUnitType type = CacheDumpUnitType::kFooter;
uint64_t timestamp = clock_->NowMicros();
uint32_t footer_checksum =
crc32c::Value(footer_value.c_str(), footer_value.size());
return WriteRawBlock(timestamp, type, Slice(footer_key),
(void*)footer_value.c_str(), footer_value.size(),
footer_checksum);
}
// This is the main function to restore the cache entries to secondary cache.
// First, we check if all the arguments are valid. Then, we read the block
// sequentially from the reader and insert them to the secondary cache.
IOStatus CacheDumpedLoaderImpl::RestoreCacheEntriesToSecondaryCache() {
// TODO: remove this line when options are used in the loader
(void)options_;
// Step 1: we check if all the arguments are valid
if (secondary_cache_ == nullptr) {
return IOStatus::InvalidArgument("Secondary Cache is null");
}
if (reader_ == nullptr) {
return IOStatus::InvalidArgument("CacheDumpReader is null");
}
// we copy the Cache Deleter Role Map as its member.
role_map_ = CopyCacheDeleterRoleMap();
// Step 2: read the header
// TODO: we need to check the cache dump format version and RocksDB version
// after the header is read out.
IOStatus io_s;
DumpUnit dump_unit;
std::string data;
io_s = ReadHeader(&data, &dump_unit);
if (!io_s.ok()) {
return io_s;
}
// Step 3: read out the rest of the blocks from the reader. The loop will stop
// either I/O status is not ok or we reach to the the end.
while (io_s.ok() && dump_unit.type != CacheDumpUnitType::kFooter) {
dump_unit.reset();
data.clear();
// read the content and store in the dump_unit
io_s = ReadCacheBlock(&data, &dump_unit);
if (!io_s.ok()) {
break;
}
// create the raw_block_content based on the information in the dump_unit
BlockContents raw_block_contents(
Slice((char*)dump_unit.value, dump_unit.value_len));
Cache::CacheItemHelper* helper = nullptr;
Statistics* statistics = nullptr;
Status s = Status::OK();
// according to the block type, get the helper callback function and create
// the corresponding block
switch (dump_unit.type) {
case CacheDumpUnitType::kFilter: {
helper = BlocklikeTraits<ParsedFullFilterBlock>::GetCacheItemHelper(
BlockType::kFilter);
std::unique_ptr<ParsedFullFilterBlock> block_holder;
block_holder.reset(BlocklikeTraits<ParsedFullFilterBlock>::Create(
std::move(raw_block_contents), toptions_.read_amp_bytes_per_bit,
statistics, false, toptions_.filter_policy.get()));
if (helper != nullptr) {
s = secondary_cache_->Insert(dump_unit.key,
(void*)(block_holder.get()), helper);
}
break;
}
case CacheDumpUnitType::kData: {
helper = BlocklikeTraits<Block>::GetCacheItemHelper(BlockType::kData);
std::unique_ptr<Block> block_holder;
block_holder.reset(BlocklikeTraits<Block>::Create(
std::move(raw_block_contents), toptions_.read_amp_bytes_per_bit,
statistics, false, toptions_.filter_policy.get()));
if (helper != nullptr) {
s = secondary_cache_->Insert(dump_unit.key,
(void*)(block_holder.get()), helper);
}
break;
}
case CacheDumpUnitType::kIndex: {
helper = BlocklikeTraits<Block>::GetCacheItemHelper(BlockType::kIndex);
std::unique_ptr<Block> block_holder;
block_holder.reset(BlocklikeTraits<Block>::Create(
std::move(raw_block_contents), 0, statistics, false,
toptions_.filter_policy.get()));
if (helper != nullptr) {
s = secondary_cache_->Insert(dump_unit.key,
(void*)(block_holder.get()), helper);
}
break;
}
case CacheDumpUnitType::kFilterMetaBlock: {
helper = BlocklikeTraits<Block>::GetCacheItemHelper(
BlockType::kFilterPartitionIndex);
std::unique_ptr<Block> block_holder;
block_holder.reset(BlocklikeTraits<Block>::Create(
std::move(raw_block_contents), toptions_.read_amp_bytes_per_bit,
statistics, false, toptions_.filter_policy.get()));
if (helper != nullptr) {
s = secondary_cache_->Insert(dump_unit.key,
(void*)(block_holder.get()), helper);
}
break;
}
case CacheDumpUnitType::kFooter:
break;
Remove deprecated block-based filter (#10184) Summary: In https://github.com/facebook/rocksdb/issues/9535, release 7.0, we hid the old block-based filter from being created using the public API, because of its inefficiency. Although we normally maintain read compatibility on old DBs forever, filters are not required for reading a DB, only for optimizing read performance. Thus, it should be acceptable to remove this code and the substantial maintenance burden it carries as useful features are developed and validated (such as user timestamp). This change completely removes the code for reading and writing the old block-based filters, net removing about 1370 lines of code no longer needed. Options removed from testing / benchmarking tools. The prior existence is only evident in a couple of places: * `CacheEntryRole::kDeprecatedFilterBlock` - We can update this public API enum in a major release to minimize source code incompatibilities. * A warning is logged when an old table file is opened that used the old block-based filter. This is provided as a courtesy, and would be a pain to unit test, so manual testing should suffice. Unfortunately, sst_dump does not tell you whether a file uses block-based filter, and the structure of the code makes it very difficult to fix. * To detect that case, `kObsoleteFilterBlockPrefix` (renamed from `kFilterBlockPrefix`) for metaindex is maintained (for now). Other notes: * In some cases where numbers are associated with filter configurations, we have had to update the assigned numbers so that they all correspond to something that exists. * Fixed potential stat counting bug by assuming `filter_checked = false` for cases like `filter == nullptr` rather than assuming `filter_checked = true` * Removed obsolete `block_offset` and `prefix_extractor` parameters from several functions. * Removed some unnecessary checks `if (!table_prefix_extractor() && !prefix_extractor)` because the caller guarantees the prefix extractor exists and is compatible Pull Request resolved: https://github.com/facebook/rocksdb/pull/10184 Test Plan: tests updated, manually test new warning in LOG using base version to generate a DB Reviewed By: riversand963 Differential Revision: D37212647 Pulled By: pdillinger fbshipit-source-id: 06ee020d8de3b81260ffc36ad0c1202cbf463a80
2 years ago
case CacheDumpUnitType::kDeprecatedFilterBlock:
// Obsolete
break;
default:
continue;
}
if (!s.ok()) {
io_s = status_to_io_status(std::move(s));
}
}
if (dump_unit.type == CacheDumpUnitType::kFooter) {
return IOStatus::OK();
} else {
return io_s;
}
}
// Read and copy the dump unit metadata to std::string data, decode and create
// the unit metadata based on the string
IOStatus CacheDumpedLoaderImpl::ReadDumpUnitMeta(std::string* data,
DumpUnitMeta* unit_meta) {
assert(reader_ != nullptr);
assert(data != nullptr);
assert(unit_meta != nullptr);
IOStatus io_s = reader_->ReadMetadata(data);
if (!io_s.ok()) {
return io_s;
}
return status_to_io_status(
CacheDumperHelper::DecodeDumpUnitMeta(*data, unit_meta));
}
// Read and copy the dump unit to std::string data, decode and create the unit
// based on the string
IOStatus CacheDumpedLoaderImpl::ReadDumpUnit(size_t len, std::string* data,
DumpUnit* unit) {
assert(reader_ != nullptr);
assert(data != nullptr);
assert(unit != nullptr);
IOStatus io_s = reader_->ReadPacket(data);
if (!io_s.ok()) {
return io_s;
}
if (data->size() != len) {
return IOStatus::Corruption(
"The data being read out does not match the size stored in metadata!");
}
Slice block;
return status_to_io_status(CacheDumperHelper::DecodeDumpUnit(*data, unit));
}
// Read the header
IOStatus CacheDumpedLoaderImpl::ReadHeader(std::string* data,
DumpUnit* dump_unit) {
DumpUnitMeta header_meta;
header_meta.reset();
std::string meta_string;
IOStatus io_s = ReadDumpUnitMeta(&meta_string, &header_meta);
if (!io_s.ok()) {
return io_s;
}
io_s = ReadDumpUnit(header_meta.dump_unit_size, data, dump_unit);
if (!io_s.ok()) {
return io_s;
}
uint32_t unit_checksum = crc32c::Value(data->c_str(), data->size());
if (unit_checksum != header_meta.dump_unit_checksum) {
return IOStatus::Corruption("Read header unit corrupted!");
}
return io_s;
}
// Read the blocks after header is read out
IOStatus CacheDumpedLoaderImpl::ReadCacheBlock(std::string* data,
DumpUnit* dump_unit) {
// According to the write process, we read the dump_unit_metadata first
DumpUnitMeta unit_meta;
unit_meta.reset();
std::string unit_string;
IOStatus io_s = ReadDumpUnitMeta(&unit_string, &unit_meta);
if (!io_s.ok()) {
return io_s;
}
// Based on the information in the dump_unit_metadata, we read the dump_unit
// and verify if its content is correct.
io_s = ReadDumpUnit(unit_meta.dump_unit_size, data, dump_unit);
if (!io_s.ok()) {
return io_s;
}
uint32_t unit_checksum = crc32c::Value(data->c_str(), data->size());
if (unit_checksum != unit_meta.dump_unit_checksum) {
return IOStatus::Corruption(
"Checksum does not match! Read dumped unit corrupted!");
}
return io_s;
}
} // namespace ROCKSDB_NAMESPACE
#endif // ROCKSDB_LITE