|
|
|
// 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) 2012 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 "util/hash.h"
|
|
|
|
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
#include <cstring>
|
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 <type_traits>
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
#include "test_util/testharness.h"
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
#include "util/coding.h"
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
#include "util/coding_lean.h"
|
|
|
|
#include "util/hash128.h"
|
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 "util/math.h"
|
|
|
|
#include "util/math128.h"
|
|
|
|
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
using ROCKSDB_NAMESPACE::BijectiveHash2x64;
|
|
|
|
using ROCKSDB_NAMESPACE::BijectiveUnhash2x64;
|
|
|
|
using ROCKSDB_NAMESPACE::DecodeFixed64;
|
|
|
|
using ROCKSDB_NAMESPACE::EncodeFixed32;
|
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
|
|
|
using ROCKSDB_NAMESPACE::EndianSwapValue;
|
|
|
|
using ROCKSDB_NAMESPACE::GetSliceHash64;
|
|
|
|
using ROCKSDB_NAMESPACE::Hash;
|
|
|
|
using ROCKSDB_NAMESPACE::Hash128;
|
Built-in support for generating unique IDs, bug fix (#8708)
Summary:
Env::GenerateUniqueId() works fine on Windows and on POSIX
where /proc/sys/kernel/random/uuid exists. Our other implementation is
flawed and easily produces collision in a new multi-threaded test.
As we rely more heavily on DB session ID uniqueness, this becomes a
serious issue.
This change combines several individually suitable entropy sources
for reliable generation of random unique IDs, with goal of uniqueness
and portability, not cryptographic strength nor maximum speed.
Specifically:
* Moves code for getting UUIDs from the OS to port::GenerateRfcUuid
rather than in Env implementation details. Callers are now told whether
the operation fails or succeeds.
* Adds an internal API GenerateRawUniqueId for generating high-quality
128-bit unique identifiers, by combining entropy from three "tracks":
* Lots of info from default Env like time, process id, and hostname.
* std::random_device
* port::GenerateRfcUuid (when working)
* Built-in implementations of Env::GenerateUniqueId() will now always
produce an RFC 4122 UUID string, either from platform-specific API or
by converting the output of GenerateRawUniqueId.
DB session IDs now use GenerateRawUniqueId while DB IDs (not as
critical) try to use port::GenerateRfcUuid but fall back on
GenerateRawUniqueId with conversion to an RFC 4122 UUID.
GenerateRawUniqueId is declared and defined under env/ rather than util/
or even port/ because of the Env dependency.
Likely follow-up: enhance GenerateRawUniqueId to be faster after the
first call and to guarantee uniqueness within the lifetime of a single
process (imparting the same property onto DB session IDs).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8708
Test Plan:
A new mini-stress test in env_test checks the various public
and internal APIs for uniqueness, including each track of
GenerateRawUniqueId individually. We can't hope to verify anywhere close
to 128 bits of entropy, but it can at least detect flaws as bad as the
old code. Serial execution of the new tests takes about 350 ms on
my machine.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D30563780
Pulled By: pdillinger
fbshipit-source-id: de4c9ff4b2f581cf784fcedb5f39f16e5185c364
3 years ago
|
|
|
using ROCKSDB_NAMESPACE::Hash2x64;
|
|
|
|
using ROCKSDB_NAMESPACE::Hash64;
|
|
|
|
using ROCKSDB_NAMESPACE::Lower32of64;
|
|
|
|
using ROCKSDB_NAMESPACE::Lower64of128;
|
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
|
|
|
using ROCKSDB_NAMESPACE::ReverseBits;
|
|
|
|
using ROCKSDB_NAMESPACE::Slice;
|
|
|
|
using ROCKSDB_NAMESPACE::Unsigned128;
|
|
|
|
using ROCKSDB_NAMESPACE::Upper32of64;
|
|
|
|
using ROCKSDB_NAMESPACE::Upper64of128;
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
|
|
|
|
// The hash algorithm is part of the file format, for example for the Bloom
|
|
|
|
// filters. Test that the hash values are stable for a set of random strings of
|
|
|
|
// varying lengths.
|
|
|
|
TEST(HashTest, Values) {
|
|
|
|
constexpr uint32_t kSeed = 0xbc9f1d34; // Same as BloomHash.
|
|
|
|
|
|
|
|
EXPECT_EQ(Hash("", 0, kSeed), 3164544308u);
|
|
|
|
EXPECT_EQ(Hash("\x08", 1, kSeed), 422599524u);
|
|
|
|
EXPECT_EQ(Hash("\x17", 1, kSeed), 3168152998u);
|
|
|
|
EXPECT_EQ(Hash("\x9a", 1, kSeed), 3195034349u);
|
|
|
|
EXPECT_EQ(Hash("\x1c", 1, kSeed), 2651681383u);
|
|
|
|
EXPECT_EQ(Hash("\x4d\x76", 2, kSeed), 2447836956u);
|
|
|
|
EXPECT_EQ(Hash("\x52\xd5", 2, kSeed), 3854228105u);
|
|
|
|
EXPECT_EQ(Hash("\x91\xf7", 2, kSeed), 31066776u);
|
|
|
|
EXPECT_EQ(Hash("\xd6\x27", 2, kSeed), 1806091603u);
|
|
|
|
EXPECT_EQ(Hash("\x30\x46\x0b", 3, kSeed), 3808221797u);
|
|
|
|
EXPECT_EQ(Hash("\x56\xdc\xd6", 3, kSeed), 2157698265u);
|
|
|
|
EXPECT_EQ(Hash("\xd4\x52\x33", 3, kSeed), 1721992661u);
|
|
|
|
EXPECT_EQ(Hash("\x6a\xb5\xf4", 3, kSeed), 2469105222u);
|
|
|
|
EXPECT_EQ(Hash("\x67\x53\x81\x1c", 4, kSeed), 118283265u);
|
|
|
|
EXPECT_EQ(Hash("\x69\xb8\xc0\x88", 4, kSeed), 3416318611u);
|
|
|
|
EXPECT_EQ(Hash("\x1e\x84\xaf\x2d", 4, kSeed), 3315003572u);
|
|
|
|
EXPECT_EQ(Hash("\x46\xdc\x54\xbe", 4, kSeed), 447346355u);
|
|
|
|
EXPECT_EQ(Hash("\xd0\x7a\x6e\xea\x56", 5, kSeed), 4255445370u);
|
|
|
|
EXPECT_EQ(Hash("\x86\x83\xd5\xa4\xd8", 5, kSeed), 2390603402u);
|
|
|
|
EXPECT_EQ(Hash("\xb7\x46\xbb\x77\xce", 5, kSeed), 2048907743u);
|
|
|
|
EXPECT_EQ(Hash("\x6c\xa8\xbc\xe5\x99", 5, kSeed), 2177978500u);
|
|
|
|
EXPECT_EQ(Hash("\x5c\x5e\xe1\xa0\x73\x81", 6, kSeed), 1036846008u);
|
|
|
|
EXPECT_EQ(Hash("\x08\x5d\x73\x1c\xe5\x2e", 6, kSeed), 229980482u);
|
|
|
|
EXPECT_EQ(Hash("\x42\xfb\xf2\x52\xb4\x10", 6, kSeed), 3655585422u);
|
|
|
|
EXPECT_EQ(Hash("\x73\xe1\xff\x56\x9c\xce", 6, kSeed), 3502708029u);
|
|
|
|
EXPECT_EQ(Hash("\x5c\xbe\x97\x75\x54\x9a\x52", 7, kSeed), 815120748u);
|
|
|
|
EXPECT_EQ(Hash("\x16\x82\x39\x49\x88\x2b\x36", 7, kSeed), 3056033698u);
|
|
|
|
EXPECT_EQ(Hash("\x59\x77\xf0\xa7\x24\xf4\x78", 7, kSeed), 587205227u);
|
|
|
|
EXPECT_EQ(Hash("\xd3\xa5\x7c\x0e\xc0\x02\x07", 7, kSeed), 2030937252u);
|
|
|
|
EXPECT_EQ(Hash("\x31\x1b\x98\x75\x96\x22\xd3\x9a", 8, kSeed), 469635402u);
|
|
|
|
EXPECT_EQ(Hash("\x38\xd6\xf7\x28\x20\xb4\x8a\xe9", 8, kSeed), 3530274698u);
|
|
|
|
EXPECT_EQ(Hash("\xbb\x18\x5d\xf4\x12\x03\xf7\x99", 8, kSeed), 1974545809u);
|
|
|
|
EXPECT_EQ(Hash("\x80\xd4\x3b\x3b\xae\x22\xa2\x78", 8, kSeed), 3563570120u);
|
|
|
|
EXPECT_EQ(Hash("\x1a\xb5\xd0\xfe\xab\xc3\x61\xb2\x99", 9, kSeed),
|
|
|
|
2706087434u);
|
|
|
|
EXPECT_EQ(Hash("\x8e\x4a\xc3\x18\x20\x2f\x06\xe6\x3c", 9, kSeed),
|
|
|
|
1534654151u);
|
|
|
|
EXPECT_EQ(Hash("\xb6\xc0\xdd\x05\x3f\xc4\x86\x4c\xef", 9, kSeed),
|
|
|
|
2355554696u);
|
|
|
|
EXPECT_EQ(Hash("\x9a\x5f\x78\x0d\xaf\x50\xe1\x1f\x55", 9, kSeed),
|
|
|
|
1400800912u);
|
|
|
|
EXPECT_EQ(Hash("\x22\x6f\x39\x1f\xf8\xdd\x4f\x52\x17\x94", 10, kSeed),
|
|
|
|
3420325137u);
|
|
|
|
EXPECT_EQ(Hash("\x32\x89\x2a\x75\x48\x3a\x4a\x02\x69\xdd", 10, kSeed),
|
|
|
|
3427803584u);
|
|
|
|
EXPECT_EQ(Hash("\x06\x92\x5c\xf4\x88\x0e\x7e\x68\x38\x3e", 10, kSeed),
|
|
|
|
1152407945u);
|
|
|
|
EXPECT_EQ(Hash("\xbd\x2c\x63\x38\xbf\xe9\x78\xb7\xbf\x15", 10, kSeed),
|
|
|
|
3382479516u);
|
|
|
|
}
|
|
|
|
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
// The hash algorithm is part of the file format, for example for the Bloom
|
|
|
|
// filters.
|
|
|
|
TEST(HashTest, Hash64Misc) {
|
|
|
|
constexpr uint32_t kSeed = 0; // Same as GetSliceHash64
|
|
|
|
|
|
|
|
for (char fill : {'\0', 'a', '1', '\xff'}) {
|
|
|
|
const size_t max_size = 1000;
|
|
|
|
const std::string str(max_size, fill);
|
|
|
|
|
|
|
|
for (size_t size = 0; size <= max_size; ++size) {
|
|
|
|
uint64_t here = Hash64(str.data(), size, kSeed);
|
|
|
|
|
|
|
|
// Must be same as unseeded Hash64 and GetSliceHash64
|
|
|
|
EXPECT_EQ(here, Hash64(str.data(), size));
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
EXPECT_EQ(here, GetSliceHash64(Slice(str.data(), size)));
|
|
|
|
|
|
|
|
// Upper and Lower must reconstruct hash
|
|
|
|
EXPECT_EQ(here, (uint64_t{Upper32of64(here)} << 32) | Lower32of64(here));
|
|
|
|
EXPECT_EQ(here, (uint64_t{Upper32of64(here)} << 32) + Lower32of64(here));
|
|
|
|
EXPECT_EQ(here, (uint64_t{Upper32of64(here)} << 32) ^ Lower32of64(here));
|
|
|
|
|
|
|
|
// Seed changes hash value (with high probability)
|
|
|
|
for (uint64_t var_seed = 1; var_seed != 0; var_seed <<= 1) {
|
|
|
|
EXPECT_NE(here, Hash64(str.data(), size, var_seed));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Size changes hash value (with high probability)
|
|
|
|
size_t max_smaller_by = std::min(size_t{30}, size);
|
|
|
|
for (size_t smaller_by = 1; smaller_by <= max_smaller_by; ++smaller_by) {
|
|
|
|
EXPECT_NE(here, Hash64(str.data(), size - smaller_by, kSeed));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that hash values are "non-trivial" for "trivial" inputs
|
|
|
|
TEST(HashTest, Hash64Trivial) {
|
|
|
|
// Thorough test too slow for regression testing
|
|
|
|
constexpr bool thorough = false;
|
|
|
|
|
|
|
|
// For various seeds, make sure hash of empty string is not zero.
|
|
|
|
constexpr uint64_t max_seed = thorough ? 0x1000000 : 0x10000;
|
|
|
|
for (uint64_t seed = 0; seed < max_seed; ++seed) {
|
|
|
|
uint64_t here = Hash64("", 0, seed);
|
|
|
|
EXPECT_NE(Lower32of64(here), 0u);
|
|
|
|
EXPECT_NE(Upper32of64(here), 0u);
|
|
|
|
}
|
|
|
|
|
|
|
|
// For standard seed, make sure hash of small strings are not zero
|
|
|
|
constexpr uint32_t kSeed = 0; // Same as GetSliceHash64
|
|
|
|
char input[4];
|
|
|
|
constexpr int max_len = thorough ? 3 : 2;
|
|
|
|
for (int len = 1; len <= max_len; ++len) {
|
|
|
|
for (uint32_t i = 0; (i >> (len * 8)) == 0; ++i) {
|
|
|
|
EncodeFixed32(input, i);
|
|
|
|
uint64_t here = Hash64(input, len, kSeed);
|
|
|
|
EXPECT_NE(Lower32of64(here), 0u);
|
|
|
|
EXPECT_NE(Upper32of64(here), 0u);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that the hash values are stable for a set of random strings of
|
|
|
|
// varying small lengths.
|
|
|
|
TEST(HashTest, Hash64SmallValueSchema) {
|
|
|
|
constexpr uint32_t kSeed = 0; // Same as GetSliceHash64
|
|
|
|
|
|
|
|
EXPECT_EQ(Hash64("", 0, kSeed), uint64_t{5999572062939766020u});
|
|
|
|
EXPECT_EQ(Hash64("\x08", 1, kSeed), uint64_t{583283813901344696u});
|
|
|
|
EXPECT_EQ(Hash64("\x17", 1, kSeed), uint64_t{16175549975585474943u});
|
|
|
|
EXPECT_EQ(Hash64("\x9a", 1, kSeed), uint64_t{16322991629225003903u});
|
|
|
|
EXPECT_EQ(Hash64("\x1c", 1, kSeed), uint64_t{13269285487706833447u});
|
|
|
|
EXPECT_EQ(Hash64("\x4d\x76", 2, kSeed), uint64_t{6859542833406258115u});
|
|
|
|
EXPECT_EQ(Hash64("\x52\xd5", 2, kSeed), uint64_t{4919611532550636959u});
|
|
|
|
EXPECT_EQ(Hash64("\x91\xf7", 2, kSeed), uint64_t{14199427467559720719u});
|
|
|
|
EXPECT_EQ(Hash64("\xd6\x27", 2, kSeed), uint64_t{12292689282614532691u});
|
|
|
|
EXPECT_EQ(Hash64("\x30\x46\x0b", 3, kSeed), uint64_t{11404699285340020889u});
|
|
|
|
EXPECT_EQ(Hash64("\x56\xdc\xd6", 3, kSeed), uint64_t{12404347133785524237u});
|
|
|
|
EXPECT_EQ(Hash64("\xd4\x52\x33", 3, kSeed), uint64_t{15853805298481534034u});
|
|
|
|
EXPECT_EQ(Hash64("\x6a\xb5\xf4", 3, kSeed), uint64_t{16863488758399383382u});
|
|
|
|
EXPECT_EQ(Hash64("\x67\x53\x81\x1c", 4, kSeed),
|
|
|
|
uint64_t{9010661983527562386u});
|
|
|
|
EXPECT_EQ(Hash64("\x69\xb8\xc0\x88", 4, kSeed),
|
|
|
|
uint64_t{6611781377647041447u});
|
|
|
|
EXPECT_EQ(Hash64("\x1e\x84\xaf\x2d", 4, kSeed),
|
|
|
|
uint64_t{15290969111616346501u});
|
|
|
|
EXPECT_EQ(Hash64("\x46\xdc\x54\xbe", 4, kSeed),
|
|
|
|
uint64_t{7063754590279313623u});
|
|
|
|
EXPECT_EQ(Hash64("\xd0\x7a\x6e\xea\x56", 5, kSeed),
|
|
|
|
uint64_t{6384167718754869899u});
|
|
|
|
EXPECT_EQ(Hash64("\x86\x83\xd5\xa4\xd8", 5, kSeed),
|
|
|
|
uint64_t{16874407254108011067u});
|
|
|
|
EXPECT_EQ(Hash64("\xb7\x46\xbb\x77\xce", 5, kSeed),
|
|
|
|
uint64_t{16809880630149135206u});
|
|
|
|
EXPECT_EQ(Hash64("\x6c\xa8\xbc\xe5\x99", 5, kSeed),
|
|
|
|
uint64_t{1249038833153141148u});
|
|
|
|
EXPECT_EQ(Hash64("\x5c\x5e\xe1\xa0\x73\x81", 6, kSeed),
|
|
|
|
uint64_t{17358142495308219330u});
|
|
|
|
EXPECT_EQ(Hash64("\x08\x5d\x73\x1c\xe5\x2e", 6, kSeed),
|
|
|
|
uint64_t{4237646583134806322u});
|
|
|
|
EXPECT_EQ(Hash64("\x42\xfb\xf2\x52\xb4\x10", 6, kSeed),
|
|
|
|
uint64_t{4373664924115234051u});
|
|
|
|
EXPECT_EQ(Hash64("\x73\xe1\xff\x56\x9c\xce", 6, kSeed),
|
|
|
|
uint64_t{12012981210634596029u});
|
|
|
|
EXPECT_EQ(Hash64("\x5c\xbe\x97\x75\x54\x9a\x52", 7, kSeed),
|
|
|
|
uint64_t{5716522398211028826u});
|
|
|
|
EXPECT_EQ(Hash64("\x16\x82\x39\x49\x88\x2b\x36", 7, kSeed),
|
|
|
|
uint64_t{15604531309862565013u});
|
|
|
|
EXPECT_EQ(Hash64("\x59\x77\xf0\xa7\x24\xf4\x78", 7, kSeed),
|
|
|
|
uint64_t{8601330687345614172u});
|
|
|
|
EXPECT_EQ(Hash64("\xd3\xa5\x7c\x0e\xc0\x02\x07", 7, kSeed),
|
|
|
|
uint64_t{8088079329364056942u});
|
|
|
|
EXPECT_EQ(Hash64("\x31\x1b\x98\x75\x96\x22\xd3\x9a", 8, kSeed),
|
|
|
|
uint64_t{9844314944338447628u});
|
|
|
|
EXPECT_EQ(Hash64("\x38\xd6\xf7\x28\x20\xb4\x8a\xe9", 8, kSeed),
|
|
|
|
uint64_t{10973293517982163143u});
|
|
|
|
EXPECT_EQ(Hash64("\xbb\x18\x5d\xf4\x12\x03\xf7\x99", 8, kSeed),
|
|
|
|
uint64_t{9986007080564743219u});
|
|
|
|
EXPECT_EQ(Hash64("\x80\xd4\x3b\x3b\xae\x22\xa2\x78", 8, kSeed),
|
|
|
|
uint64_t{1729303145008254458u});
|
|
|
|
EXPECT_EQ(Hash64("\x1a\xb5\xd0\xfe\xab\xc3\x61\xb2\x99", 9, kSeed),
|
|
|
|
uint64_t{13253403748084181481u});
|
|
|
|
EXPECT_EQ(Hash64("\x8e\x4a\xc3\x18\x20\x2f\x06\xe6\x3c", 9, kSeed),
|
|
|
|
uint64_t{7768754303876232188u});
|
|
|
|
EXPECT_EQ(Hash64("\xb6\xc0\xdd\x05\x3f\xc4\x86\x4c\xef", 9, kSeed),
|
|
|
|
uint64_t{12439346786701492u});
|
|
|
|
EXPECT_EQ(Hash64("\x9a\x5f\x78\x0d\xaf\x50\xe1\x1f\x55", 9, kSeed),
|
|
|
|
uint64_t{10841838338450144690u});
|
|
|
|
EXPECT_EQ(Hash64("\x22\x6f\x39\x1f\xf8\xdd\x4f\x52\x17\x94", 10, kSeed),
|
|
|
|
uint64_t{12883919702069153152u});
|
|
|
|
EXPECT_EQ(Hash64("\x32\x89\x2a\x75\x48\x3a\x4a\x02\x69\xdd", 10, kSeed),
|
|
|
|
uint64_t{12692903507676842188u});
|
|
|
|
EXPECT_EQ(Hash64("\x06\x92\x5c\xf4\x88\x0e\x7e\x68\x38\x3e", 10, kSeed),
|
|
|
|
uint64_t{6540985900674032620u});
|
|
|
|
EXPECT_EQ(Hash64("\xbd\x2c\x63\x38\xbf\xe9\x78\xb7\xbf\x15", 10, kSeed),
|
|
|
|
uint64_t{10551812464348219044u});
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string Hash64TestDescriptor(const char *repeat, size_t limit) {
|
|
|
|
const char *mod61_encode =
|
|
|
|
"abcdefghijklmnopqrstuvwxyz123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
|
|
|
|
|
|
|
|
std::string input;
|
|
|
|
while (input.size() < limit) {
|
|
|
|
input.append(repeat);
|
|
|
|
}
|
|
|
|
std::string rv;
|
|
|
|
for (size_t i = 0; i < limit; ++i) {
|
|
|
|
uint64_t h = GetSliceHash64(Slice(input.data(), i));
|
|
|
|
rv.append(1, mod61_encode[static_cast<size_t>(h % 61)]);
|
|
|
|
}
|
|
|
|
return rv;
|
|
|
|
}
|
|
|
|
|
|
|
|
// XXPH3 changes its algorithm for various sizes up through 250 bytes, so
|
Add new persistent 64-bit hash (#5984)
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
5 years ago
|
|
|
// we need to check the stability of larger sizes also.
|
|
|
|
TEST(HashTest, Hash64LargeValueSchema) {
|
|
|
|
// Each of these derives a "descriptor" from the hash values for all
|
|
|
|
// lengths up to 430.
|
|
|
|
// Note that "c" is common for the zero-length string.
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash64TestDescriptor("foo", 430),
|
|
|
|
"cRhyWsY67B6klRA1udmOuiYuX7IthyGBKqbeosz2hzVglWCmQx8nEdnpkvPfYX56Up2OWOTV"
|
|
|
|
"lTzfAoYwvtqKzjD8E9xttR2unelbXbIV67NUe6bOO23BxaSFRcA3njGu5cUWfgwOqNoTsszp"
|
|
|
|
"uPvKRP6qaUR5VdoBkJUCFIefd7edlNK5mv6JYWaGdwxehg65hTkTmjZoPKxTZo4PLyzbL9U4"
|
|
|
|
"xt12ITSfeP2MfBHuLI2z2pDlBb44UQKVMx27LEoAHsdLp3WfWfgH3sdRBRCHm33UxCM4QmE2"
|
|
|
|
"xJ7gqSvNwTeH7v9GlC8zWbGroyD3UVNeShMLx29O7tH1biemLULwAHyIw8zdtLMDpEJ8m2ic"
|
|
|
|
"l6Lb4fDuuFNAs1GCVUthjK8CV8SWI8Rsz5THSwn5CGhpqUwSZcFknjwWIl5rNCvDxXJqYr");
|
|
|
|
// Note that "1EeRk" is common for "Rocks"
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash64TestDescriptor("Rocks", 430),
|
|
|
|
"c1EeRkrzgOYWLA8PuhJrwTePJewoB44WdXYDfhbk3ZxTqqg25WlPExDl7IKIQLJvnA6gJxxn"
|
|
|
|
"9TCSLkFGfJeXehaSS1GBqWSzfhEH4VXiXIUCuxJXxtKXcSC6FrNIQGTZbYDiUOLD6Y5inzrF"
|
|
|
|
"9etwQhXUBanw55xAUdNMFQAm2GjJ6UDWp2mISLiMMkLjANWMKLaZMqaFLX37qB4MRO1ooVRv"
|
|
|
|
"zSvaNRSCLxlggQCasQq8icWjzf3HjBlZtU6pd4rkaUxSzHqmo9oM5MghbU5Rtxg8wEfO7lVN"
|
|
|
|
"5wdMONYecslQTwjZUpO1K3LDf3K3XK6sUXM6ShQQ3RHmMn2acB4YtTZ3QQcHYJSOHn2DuWpa"
|
|
|
|
"Q8RqzX5lab92YmOLaCdOHq1BPsM7SIBzMdLgePNsJ1vvMALxAaoDUHPxoFLO2wx18IXnyX");
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash64TestDescriptor("RocksDB", 430),
|
|
|
|
"c1EeRkukbkb28wLTahwD2sfUhZzaBEnF8SVrxnPVB6A7b8CaAl3UKsDZISF92GSq2wDCukOq"
|
|
|
|
"Jgrsp7A3KZhDiLW8dFXp8UPqPxMCRlMdZeVeJ2dJxrmA6cyt99zkQFj7ELbut6jAeVqARFnw"
|
|
|
|
"fnWVXOsaLrq7bDCbMcns2DKvTaaqTCLMYxI7nhtLpFN1jR755FRQFcOzrrDbh7QhypjdvlYw"
|
|
|
|
"cdAMSZgp9JMHxbM23wPSuH6BOFgxejz35PScZfhDPvTOxIy1jc3MZsWrMC3P324zNolO7JdW"
|
|
|
|
"CX2I5UDKjjaEJfxbgVgJIXxtQGlmj2xkO5sPpjULQV4X2HlY7FQleJ4QRaJIB4buhCA4vUTF"
|
|
|
|
"eMFlxCIYUpTCsal2qsmnGOWa8WCcefrohMjDj1fjzSvSaQwlpyR1GZHF2uPOoQagiCpHpm");
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(HashTest, Hash128Misc) {
|
|
|
|
constexpr uint32_t kSeed = 0; // Same as GetSliceHash128
|
|
|
|
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
for (char fill : {'\0', 'a', '1', '\xff', 'e'}) {
|
|
|
|
const size_t max_size = 1000;
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
std::string str(max_size, fill);
|
|
|
|
|
|
|
|
if (fill == 'e') {
|
|
|
|
// Use different characters to check endianness handling
|
|
|
|
for (size_t i = 0; i < str.size(); ++i) {
|
|
|
|
str[i] += static_cast<char>(i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (size_t size = 0; size <= max_size; ++size) {
|
|
|
|
Unsigned128 here = Hash128(str.data(), size, kSeed);
|
|
|
|
|
|
|
|
// Must be same as unseeded Hash128 and GetSliceHash128
|
|
|
|
EXPECT_EQ(here, Hash128(str.data(), size));
|
|
|
|
EXPECT_EQ(here, GetSliceHash128(Slice(str.data(), size)));
|
Built-in support for generating unique IDs, bug fix (#8708)
Summary:
Env::GenerateUniqueId() works fine on Windows and on POSIX
where /proc/sys/kernel/random/uuid exists. Our other implementation is
flawed and easily produces collision in a new multi-threaded test.
As we rely more heavily on DB session ID uniqueness, this becomes a
serious issue.
This change combines several individually suitable entropy sources
for reliable generation of random unique IDs, with goal of uniqueness
and portability, not cryptographic strength nor maximum speed.
Specifically:
* Moves code for getting UUIDs from the OS to port::GenerateRfcUuid
rather than in Env implementation details. Callers are now told whether
the operation fails or succeeds.
* Adds an internal API GenerateRawUniqueId for generating high-quality
128-bit unique identifiers, by combining entropy from three "tracks":
* Lots of info from default Env like time, process id, and hostname.
* std::random_device
* port::GenerateRfcUuid (when working)
* Built-in implementations of Env::GenerateUniqueId() will now always
produce an RFC 4122 UUID string, either from platform-specific API or
by converting the output of GenerateRawUniqueId.
DB session IDs now use GenerateRawUniqueId while DB IDs (not as
critical) try to use port::GenerateRfcUuid but fall back on
GenerateRawUniqueId with conversion to an RFC 4122 UUID.
GenerateRawUniqueId is declared and defined under env/ rather than util/
or even port/ because of the Env dependency.
Likely follow-up: enhance GenerateRawUniqueId to be faster after the
first call and to guarantee uniqueness within the lifetime of a single
process (imparting the same property onto DB session IDs).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8708
Test Plan:
A new mini-stress test in env_test checks the various public
and internal APIs for uniqueness, including each track of
GenerateRawUniqueId individually. We can't hope to verify anywhere close
to 128 bits of entropy, but it can at least detect flaws as bad as the
old code. Serial execution of the new tests takes about 350 ms on
my machine.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D30563780
Pulled By: pdillinger
fbshipit-source-id: de4c9ff4b2f581cf784fcedb5f39f16e5185c364
3 years ago
|
|
|
{
|
|
|
|
uint64_t hi, lo;
|
|
|
|
Hash2x64(str.data(), size, &hi, &lo);
|
|
|
|
EXPECT_EQ(Lower64of128(here), lo);
|
|
|
|
EXPECT_EQ(Upper64of128(here), hi);
|
|
|
|
}
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
if (size == 16) {
|
|
|
|
const uint64_t in_hi = DecodeFixed64(str.data() + 8);
|
|
|
|
const uint64_t in_lo = DecodeFixed64(str.data());
|
|
|
|
uint64_t hi, lo;
|
|
|
|
BijectiveHash2x64(in_hi, in_lo, &hi, &lo);
|
|
|
|
EXPECT_EQ(Lower64of128(here), lo);
|
|
|
|
EXPECT_EQ(Upper64of128(here), hi);
|
|
|
|
uint64_t un_hi, un_lo;
|
|
|
|
BijectiveUnhash2x64(hi, lo, &un_hi, &un_lo);
|
|
|
|
EXPECT_EQ(in_lo, un_lo);
|
|
|
|
EXPECT_EQ(in_hi, un_hi);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Upper and Lower must reconstruct hash
|
|
|
|
EXPECT_EQ(here,
|
|
|
|
(Unsigned128{Upper64of128(here)} << 64) | Lower64of128(here));
|
|
|
|
EXPECT_EQ(here,
|
|
|
|
(Unsigned128{Upper64of128(here)} << 64) ^ Lower64of128(here));
|
|
|
|
|
|
|
|
// Seed changes hash value (with high probability)
|
|
|
|
for (uint64_t var_seed = 1; var_seed != 0; var_seed <<= 1) {
|
Experimental support for SST unique IDs (#8990)
Summary:
* New public header unique_id.h and function GetUniqueIdFromTableProperties
which computes a universally unique identifier based on table properties
of table files from recent RocksDB versions.
* Generation of DB session IDs is refactored so that they are
guaranteed unique in the lifetime of a process running RocksDB.
(SemiStructuredUniqueIdGen, new test included.) Along with file numbers,
this enables SST unique IDs to be guaranteed unique among SSTs generated
in a single process, and "better than random" between processes.
See https://github.com/pdillinger/unique_id
* In addition to public API producing 'external' unique IDs, there is a function
for producing 'internal' unique IDs, with functions for converting between the
two. In short, the external ID is "safe" for things people might do with it, and
the internal ID enables more "power user" features for the future. Specifically,
the external ID goes through a hashing layer so that any subset of bits in the
external ID can be used as a hash of the full ID, while also preserving
uniqueness guarantees in the first 128 bits (bijective both on first 128 bits
and on full 192 bits).
Intended follow-up:
* Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into
the third 64-bit value of the unique ID.)
* Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990
Test Plan:
Unit tests added, and checking of unique ids in stress test.
NOTE in stress test we do not generate nearly enough files to thoroughly
stress uniqueness, but the test trims off pieces of the ID to check for
uniqueness so that we can infer (with some assumptions) stronger
properties in the aggregate.
Reviewed By: zhichao-cao, mrambacher
Differential Revision: D31582865
Pulled By: pdillinger
fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
|
|
|
Unsigned128 seeded = Hash128(str.data(), size, var_seed);
|
|
|
|
EXPECT_NE(here, seeded);
|
|
|
|
// Must match seeded Hash2x64
|
|
|
|
{
|
|
|
|
uint64_t hi, lo;
|
|
|
|
Hash2x64(str.data(), size, var_seed, &hi, &lo);
|
|
|
|
EXPECT_EQ(Lower64of128(seeded), lo);
|
|
|
|
EXPECT_EQ(Upper64of128(seeded), hi);
|
|
|
|
}
|
|
|
|
if (size == 16) {
|
|
|
|
const uint64_t in_hi = DecodeFixed64(str.data() + 8);
|
|
|
|
const uint64_t in_lo = DecodeFixed64(str.data());
|
|
|
|
uint64_t hi, lo;
|
|
|
|
BijectiveHash2x64(in_hi, in_lo, var_seed, &hi, &lo);
|
|
|
|
EXPECT_EQ(Lower64of128(seeded), lo);
|
|
|
|
EXPECT_EQ(Upper64of128(seeded), hi);
|
|
|
|
uint64_t un_hi, un_lo;
|
|
|
|
BijectiveUnhash2x64(hi, lo, var_seed, &un_hi, &un_lo);
|
|
|
|
EXPECT_EQ(in_lo, un_lo);
|
|
|
|
EXPECT_EQ(in_hi, un_hi);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Size changes hash value (with high probability)
|
|
|
|
size_t max_smaller_by = std::min(size_t{30}, size);
|
|
|
|
for (size_t smaller_by = 1; smaller_by <= max_smaller_by; ++smaller_by) {
|
|
|
|
EXPECT_NE(here, Hash128(str.data(), size - smaller_by, kSeed));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that hash values are "non-trivial" for "trivial" inputs
|
|
|
|
TEST(HashTest, Hash128Trivial) {
|
|
|
|
// Thorough test too slow for regression testing
|
|
|
|
constexpr bool thorough = false;
|
|
|
|
|
|
|
|
// For various seeds, make sure hash of empty string is not zero.
|
|
|
|
constexpr uint64_t max_seed = thorough ? 0x1000000 : 0x10000;
|
|
|
|
for (uint64_t seed = 0; seed < max_seed; ++seed) {
|
|
|
|
Unsigned128 here = Hash128("", 0, seed);
|
|
|
|
EXPECT_NE(Lower64of128(here), 0u);
|
|
|
|
EXPECT_NE(Upper64of128(here), 0u);
|
|
|
|
}
|
|
|
|
|
|
|
|
// For standard seed, make sure hash of small strings are not zero
|
|
|
|
constexpr uint32_t kSeed = 0; // Same as GetSliceHash128
|
|
|
|
char input[4];
|
|
|
|
constexpr int max_len = thorough ? 3 : 2;
|
|
|
|
for (int len = 1; len <= max_len; ++len) {
|
|
|
|
for (uint32_t i = 0; (i >> (len * 8)) == 0; ++i) {
|
|
|
|
EncodeFixed32(input, i);
|
|
|
|
Unsigned128 here = Hash128(input, len, kSeed);
|
|
|
|
EXPECT_NE(Lower64of128(here), 0u);
|
|
|
|
EXPECT_NE(Upper64of128(here), 0u);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string Hash128TestDescriptor(const char *repeat, size_t limit) {
|
|
|
|
const char *mod61_encode =
|
|
|
|
"abcdefghijklmnopqrstuvwxyz123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
|
|
|
|
|
|
|
|
std::string input;
|
|
|
|
while (input.size() < limit) {
|
|
|
|
input.append(repeat);
|
|
|
|
}
|
|
|
|
std::string rv;
|
|
|
|
for (size_t i = 0; i < limit; ++i) {
|
|
|
|
auto h = GetSliceHash128(Slice(input.data(), i));
|
|
|
|
uint64_t h2 = Upper64of128(h) + Lower64of128(h);
|
|
|
|
rv.append(1, mod61_encode[static_cast<size_t>(h2 % 61)]);
|
|
|
|
}
|
|
|
|
return rv;
|
|
|
|
}
|
|
|
|
|
|
|
|
// XXH3 changes its algorithm for various sizes up through 250 bytes, so
|
|
|
|
// we need to check the stability of larger sizes also.
|
|
|
|
TEST(HashTest, Hash128ValueSchema) {
|
|
|
|
// Each of these derives a "descriptor" from the hash values for all
|
|
|
|
// lengths up to 430.
|
|
|
|
// Note that "b" is common for the zero-length string.
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash128TestDescriptor("foo", 430),
|
|
|
|
"bUMA3As8n9I4vNGhThXlEevxZlyMcbb6TYAlIKJ2f5ponsv99q962rYclQ7u3gfnRdCDQ5JI"
|
|
|
|
"2LrGUaCycbXrvLFe4SjgRb9RQwCfrnmNQ7VSEwSKMnkGCK3bDbXSrnIh5qLXdtvIZklbJpGH"
|
|
|
|
"Dqr93BlqF9ubTnOSYkSdx89XvQqflMIW8bjfQp9BPjQejWOeEQspnN1D3sfgVdFhpaQdHYA5"
|
|
|
|
"pI2XcPlCMFPxvrFuRr7joaDvjNe9IUZaunLPMewuXmC3EL95h52Ju3D7y9RNKhgYxMTrA84B"
|
|
|
|
"yJrMvyjdm3vlBxet4EN7v2GEyjbGuaZW9UL6lrX6PghJDg7ACfLGdxNbH3qXM4zaiG2RKnL5"
|
|
|
|
"S3WXKR78RBB5fRFQ8KDIEQjHFvSNsc3GrAEi6W8P2lv8JMTzjBODO2uN4wadVQFT9wpGfV");
|
|
|
|
// Note that "35D2v" is common for "Rocks"
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash128TestDescriptor("Rocks", 430),
|
|
|
|
"b35D2vzvklFVDqJmyLRXyApwGGO3EAT3swhe8XJAN3mY2UVPglzdmydxcba6JI2tSvwO6zSu"
|
|
|
|
"ANpjSM7tc9G5iMhsa7R8GfyCXRO1TnLg7HvdWNdgGGBirxZR68BgT7TQsYJt6zyEyISeXI1n"
|
|
|
|
"MXA48Xo7dWfJeYN6Z4KWlqZY7TgFXGbks9AX4ehZNSGtIhdO5i58qlgVX1bEejeOVaCcjC79"
|
|
|
|
"67DrMfOKds7rUQzjBa77sMPcoPW1vu6ljGJPZH3XkRyDMZ1twxXKkNxN3tE8nR7JHwyqBAxE"
|
|
|
|
"fTcjbOWrLZ1irWxRSombD8sGDEmclgF11IxqEhe3Rt7gyofO3nExGckKkS9KfRqsCHbiUyva"
|
|
|
|
"JGkJwUHRXaZnh58b4i1Ei9aQKZjXlvIVDixoZrjcNaH5XJIJlRZce9Z9t82wYapTpckYSg");
|
|
|
|
EXPECT_EQ(
|
|
|
|
Hash128TestDescriptor("RocksDB", 430),
|
|
|
|
"b35D2vFUst3XDZCRlSrhmYYakmqImV97LbBsV6EZlOEQpUPH1d1sD3xMKAPlA5UErHehg5O7"
|
|
|
|
"n966fZqhAf3hRc24kGCLfNAWjyUa7vSNOx3IcPoTyVRFZeFlcCtfl7t1QJumHOCpS33EBmBF"
|
|
|
|
"hvK13QjBbDWYWeHQhJhgV9Mqbx17TIcvUkEnYZxb8IzWNmjVsJG44Z7v52DjGj1ZzS62S2Vv"
|
|
|
|
"qWcDO7apvH5VHg68E9Wl6nXP21vlmUqEH9GeWRehfWVvY7mUpsAg5drHHQyDSdiMceiUuUxJ"
|
|
|
|
"XJqHFcDdzbbPk7xDvbLgWCKvH8k3MpQNWOmbSSRDdAP6nGlDjoTToYkcqVREHJzztSWAAq5h"
|
|
|
|
"GHSUNJ6OxsMHhf8EhXfHtKyUzRmPtjYyeckQcGmrQfFFLidc6cjMDKCdBG6c6HVBrS7H2R");
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(FastRange32Test, Values) {
|
|
|
|
using ROCKSDB_NAMESPACE::FastRange32;
|
|
|
|
// Zero range
|
|
|
|
EXPECT_EQ(FastRange32(0, 0), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(123, 0), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 0), 0U);
|
|
|
|
|
|
|
|
// One range
|
|
|
|
EXPECT_EQ(FastRange32(0, 1), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(123, 1), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 1), 0U);
|
|
|
|
|
|
|
|
// Two range
|
|
|
|
EXPECT_EQ(FastRange32(0, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(123, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(0x7fffffff, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(0x80000000, 2), 1U);
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 2), 1U);
|
|
|
|
|
|
|
|
// Seven range
|
|
|
|
EXPECT_EQ(FastRange32(0, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(123, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(613566756, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(613566757, 7), 1U);
|
|
|
|
EXPECT_EQ(FastRange32(1227133513, 7), 1U);
|
|
|
|
EXPECT_EQ(FastRange32(1227133514, 7), 2U);
|
|
|
|
// etc.
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 7), 6U);
|
|
|
|
|
|
|
|
// Big
|
|
|
|
EXPECT_EQ(FastRange32(1, 0x80000000), 0U);
|
|
|
|
EXPECT_EQ(FastRange32(2, 0x80000000), 1U);
|
|
|
|
EXPECT_EQ(FastRange32(4, 0x7fffffff), 1U);
|
|
|
|
EXPECT_EQ(FastRange32(4, 0x80000000), 2U);
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 0x7fffffff), 0x7ffffffeU);
|
|
|
|
EXPECT_EQ(FastRange32(0xffffffff, 0x80000000), 0x7fffffffU);
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(FastRange64Test, Values) {
|
|
|
|
using ROCKSDB_NAMESPACE::FastRange64;
|
|
|
|
// Zero range
|
|
|
|
EXPECT_EQ(FastRange64(0, 0), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(123, 0), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFF, 0), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 0), 0U);
|
|
|
|
|
|
|
|
// One range
|
|
|
|
EXPECT_EQ(FastRange64(0, 1), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(123, 1), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFF, 1), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 1), 0U);
|
|
|
|
|
|
|
|
// Two range
|
|
|
|
EXPECT_EQ(FastRange64(0, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(123, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFF, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0x7fffFFFFffffFFFF, 2), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0x8000000000000000, 2), 1U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 2), 1U);
|
|
|
|
|
|
|
|
// Seven range
|
|
|
|
EXPECT_EQ(FastRange64(0, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(123, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFF, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(2635249153387078802, 7), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(2635249153387078803, 7), 1U);
|
|
|
|
EXPECT_EQ(FastRange64(5270498306774157604, 7), 1U);
|
|
|
|
EXPECT_EQ(FastRange64(5270498306774157605, 7), 2U);
|
|
|
|
EXPECT_EQ(FastRange64(0x7fffFFFFffffFFFF, 7), 3U);
|
|
|
|
EXPECT_EQ(FastRange64(0x8000000000000000, 7), 3U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 7), 6U);
|
|
|
|
|
|
|
|
// Big but 32-bit range
|
|
|
|
EXPECT_EQ(FastRange64(0x100000000, 0x80000000), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0x200000000, 0x80000000), 1U);
|
|
|
|
EXPECT_EQ(FastRange64(0x400000000, 0x7fffFFFF), 1U);
|
|
|
|
EXPECT_EQ(FastRange64(0x400000000, 0x80000000), 2U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 0x7fffFFFF), 0x7fffFFFEU);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 0x80000000), 0x7fffFFFFU);
|
|
|
|
|
|
|
|
// Big, > 32-bit range
|
|
|
|
#if SIZE_MAX == UINT64_MAX
|
|
|
|
EXPECT_EQ(FastRange64(0x7fffFFFFffffFFFF, 0x4200000002), 0x2100000000U);
|
|
|
|
EXPECT_EQ(FastRange64(0x8000000000000000, 0x4200000002), 0x2100000001U);
|
|
|
|
|
|
|
|
EXPECT_EQ(FastRange64(0x0000000000000000, 420000000002), 0U);
|
|
|
|
EXPECT_EQ(FastRange64(0x7fffFFFFffffFFFF, 420000000002), 210000000000U);
|
|
|
|
EXPECT_EQ(FastRange64(0x8000000000000000, 420000000002), 210000000001U);
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 420000000002), 420000000001U);
|
|
|
|
|
|
|
|
EXPECT_EQ(FastRange64(0xffffFFFFffffFFFF, 0xffffFFFFffffFFFF),
|
|
|
|
0xffffFFFFffffFFFEU);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(FastRangeGenericTest, Values) {
|
|
|
|
using ROCKSDB_NAMESPACE::FastRangeGeneric;
|
|
|
|
// Generic (including big and small)
|
|
|
|
// Note that FastRangeGeneric is also tested indirectly above via
|
|
|
|
// FastRange32 and FastRange64.
|
|
|
|
EXPECT_EQ(
|
|
|
|
FastRangeGeneric(uint64_t{0x8000000000000000}, uint64_t{420000000002}),
|
|
|
|
uint64_t{210000000001});
|
|
|
|
EXPECT_EQ(FastRangeGeneric(uint64_t{0x8000000000000000}, uint16_t{12468}),
|
|
|
|
uint16_t{6234});
|
|
|
|
EXPECT_EQ(FastRangeGeneric(uint32_t{0x80000000}, uint16_t{12468}),
|
|
|
|
uint16_t{6234});
|
|
|
|
// Not recommended for typical use because for example this could fail on
|
|
|
|
// some platforms and pass on others:
|
|
|
|
//EXPECT_EQ(FastRangeGeneric(static_cast<unsigned long>(0x80000000),
|
|
|
|
// uint16_t{12468}),
|
|
|
|
// uint16_t{6234});
|
|
|
|
}
|
|
|
|
|
|
|
|
// for inspection of disassembly
|
|
|
|
uint32_t FastRange32(uint32_t hash, uint32_t range) {
|
|
|
|
return ROCKSDB_NAMESPACE::FastRange32(hash, range);
|
|
|
|
}
|
|
|
|
|
|
|
|
// for inspection of disassembly
|
|
|
|
size_t FastRange64(uint64_t hash, size_t range) {
|
|
|
|
return ROCKSDB_NAMESPACE::FastRange64(hash, range);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Tests for math.h / math128.h (not worth a separate test binary)
|
|
|
|
using ROCKSDB_NAMESPACE::BitParity;
|
|
|
|
using ROCKSDB_NAMESPACE::BitsSetToOne;
|
Meta-internal folly integration with F14FastMap (#9546)
Summary:
Especially after updating to C++17, I don't see a compelling case for
*requiring* any folly components in RocksDB. I was able to purge the existing
hard dependencies, and it can be quite difficult to strip out non-trivial components
from folly for use in RocksDB. (The prospect of doing that on F14 has changed
my mind on the best approach here.)
But this change creates an optional integration where we can plug in
components from folly at compile time, starting here with F14FastMap to replace
std::unordered_map when possible (probably no public APIs for example). I have
replaced the biggest CPU users of std::unordered_map with compile-time
pluggable UnorderedMap which will use F14FastMap when USE_FOLLY is set.
USE_FOLLY is always set in the Meta-internal buck build, and a simulation of
that is in the Makefile for public CI testing. A full folly build is not needed, but
checking out the full folly repo is much simpler for getting the dependency,
and anything else we might want to optionally integrate in the future.
Some picky details:
* I don't think the distributed mutex stuff is actually used, so it was easy to remove.
* I implemented an alternative to `folly::constexpr_log2` (which is much easier
in C++17 than C++11) so that I could pull out the hard dependencies on
`ConstexprMath.h`
* I had to add noexcept move constructors/operators to some types to make
F14's complainUnlessNothrowMoveAndDestroy check happy, and I added a
macro to make that easier in some common cases.
* Updated Meta-internal buck build to use folly F14Map (always)
No updates to HISTORY.md nor INSTALL.md as this is not (yet?) considered a
production integration for open source users.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9546
Test Plan:
CircleCI tests updated so that a couple of them use folly.
Most internal unit & stress/crash tests updated to use Meta-internal latest folly.
(Note: they should probably use buck but they currently use Makefile.)
Example performance improvement: when filter partitions are pinned in cache,
they are tracked by PartitionedFilterBlockReader::filter_map_ and we can build
a test that exercises that heavily. Build DB with
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters
```
and test with (simultaneous runs with & without folly, ~20 times each to see
convergence)
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench_folly -readonly -use_existing_db -benchmarks=readrandom -num=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters -duration=40 -pin_l0_filter_and_index_blocks_in_cache
```
Average ops/s no folly: 26229.2
Average ops/s with folly: 26853.3 (+2.4%)
Reviewed By: ajkr
Differential Revision: D34181736
Pulled By: pdillinger
fbshipit-source-id: ffa6ad5104c2880321d8a1aa7187e00ab0d02e94
3 years ago
|
|
|
using ROCKSDB_NAMESPACE::ConstexprFloorLog2;
|
|
|
|
using ROCKSDB_NAMESPACE::CountTrailingZeroBits;
|
|
|
|
using ROCKSDB_NAMESPACE::DecodeFixed128;
|
|
|
|
using ROCKSDB_NAMESPACE::DecodeFixedGeneric;
|
|
|
|
using ROCKSDB_NAMESPACE::EncodeFixed128;
|
|
|
|
using ROCKSDB_NAMESPACE::EncodeFixedGeneric;
|
|
|
|
using ROCKSDB_NAMESPACE::FloorLog2;
|
|
|
|
using ROCKSDB_NAMESPACE::Lower64of128;
|
|
|
|
using ROCKSDB_NAMESPACE::Multiply64to128;
|
|
|
|
using ROCKSDB_NAMESPACE::Unsigned128;
|
|
|
|
using ROCKSDB_NAMESPACE::Upper64of128;
|
|
|
|
|
|
|
|
template <typename T>
|
|
|
|
static void test_BitOps() {
|
|
|
|
// This complex code is to generalize to 128-bit values. Otherwise
|
|
|
|
// we could just use = static_cast<T>(0x5555555555555555ULL);
|
|
|
|
T everyOtherBit = 0;
|
|
|
|
for (unsigned i = 0; i < sizeof(T); ++i) {
|
|
|
|
everyOtherBit = (everyOtherBit << 8) | T{0x55};
|
|
|
|
}
|
|
|
|
|
|
|
|
// This one built using bit operations, as our 128-bit layer
|
|
|
|
// might not implement arithmetic such as subtraction.
|
|
|
|
T vm1 = 0; // "v minus one"
|
|
|
|
|
|
|
|
for (int i = 0; i < int{8 * sizeof(T)}; ++i) {
|
|
|
|
T v = T{1} << i;
|
|
|
|
// If we could directly use arithmetic:
|
|
|
|
// T vm1 = static_cast<T>(v - 1);
|
|
|
|
|
|
|
|
// FloorLog2
|
|
|
|
if (v > 0) {
|
|
|
|
EXPECT_EQ(FloorLog2(v), i);
|
Meta-internal folly integration with F14FastMap (#9546)
Summary:
Especially after updating to C++17, I don't see a compelling case for
*requiring* any folly components in RocksDB. I was able to purge the existing
hard dependencies, and it can be quite difficult to strip out non-trivial components
from folly for use in RocksDB. (The prospect of doing that on F14 has changed
my mind on the best approach here.)
But this change creates an optional integration where we can plug in
components from folly at compile time, starting here with F14FastMap to replace
std::unordered_map when possible (probably no public APIs for example). I have
replaced the biggest CPU users of std::unordered_map with compile-time
pluggable UnorderedMap which will use F14FastMap when USE_FOLLY is set.
USE_FOLLY is always set in the Meta-internal buck build, and a simulation of
that is in the Makefile for public CI testing. A full folly build is not needed, but
checking out the full folly repo is much simpler for getting the dependency,
and anything else we might want to optionally integrate in the future.
Some picky details:
* I don't think the distributed mutex stuff is actually used, so it was easy to remove.
* I implemented an alternative to `folly::constexpr_log2` (which is much easier
in C++17 than C++11) so that I could pull out the hard dependencies on
`ConstexprMath.h`
* I had to add noexcept move constructors/operators to some types to make
F14's complainUnlessNothrowMoveAndDestroy check happy, and I added a
macro to make that easier in some common cases.
* Updated Meta-internal buck build to use folly F14Map (always)
No updates to HISTORY.md nor INSTALL.md as this is not (yet?) considered a
production integration for open source users.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9546
Test Plan:
CircleCI tests updated so that a couple of them use folly.
Most internal unit & stress/crash tests updated to use Meta-internal latest folly.
(Note: they should probably use buck but they currently use Makefile.)
Example performance improvement: when filter partitions are pinned in cache,
they are tracked by PartitionedFilterBlockReader::filter_map_ and we can build
a test that exercises that heavily. Build DB with
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters
```
and test with (simultaneous runs with & without folly, ~20 times each to see
convergence)
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench_folly -readonly -use_existing_db -benchmarks=readrandom -num=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters -duration=40 -pin_l0_filter_and_index_blocks_in_cache
```
Average ops/s no folly: 26229.2
Average ops/s with folly: 26853.3 (+2.4%)
Reviewed By: ajkr
Differential Revision: D34181736
Pulled By: pdillinger
fbshipit-source-id: ffa6ad5104c2880321d8a1aa7187e00ab0d02e94
3 years ago
|
|
|
EXPECT_EQ(ConstexprFloorLog2(v), i);
|
|
|
|
}
|
|
|
|
if (vm1 > 0) {
|
|
|
|
EXPECT_EQ(FloorLog2(vm1), i - 1);
|
Meta-internal folly integration with F14FastMap (#9546)
Summary:
Especially after updating to C++17, I don't see a compelling case for
*requiring* any folly components in RocksDB. I was able to purge the existing
hard dependencies, and it can be quite difficult to strip out non-trivial components
from folly for use in RocksDB. (The prospect of doing that on F14 has changed
my mind on the best approach here.)
But this change creates an optional integration where we can plug in
components from folly at compile time, starting here with F14FastMap to replace
std::unordered_map when possible (probably no public APIs for example). I have
replaced the biggest CPU users of std::unordered_map with compile-time
pluggable UnorderedMap which will use F14FastMap when USE_FOLLY is set.
USE_FOLLY is always set in the Meta-internal buck build, and a simulation of
that is in the Makefile for public CI testing. A full folly build is not needed, but
checking out the full folly repo is much simpler for getting the dependency,
and anything else we might want to optionally integrate in the future.
Some picky details:
* I don't think the distributed mutex stuff is actually used, so it was easy to remove.
* I implemented an alternative to `folly::constexpr_log2` (which is much easier
in C++17 than C++11) so that I could pull out the hard dependencies on
`ConstexprMath.h`
* I had to add noexcept move constructors/operators to some types to make
F14's complainUnlessNothrowMoveAndDestroy check happy, and I added a
macro to make that easier in some common cases.
* Updated Meta-internal buck build to use folly F14Map (always)
No updates to HISTORY.md nor INSTALL.md as this is not (yet?) considered a
production integration for open source users.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9546
Test Plan:
CircleCI tests updated so that a couple of them use folly.
Most internal unit & stress/crash tests updated to use Meta-internal latest folly.
(Note: they should probably use buck but they currently use Makefile.)
Example performance improvement: when filter partitions are pinned in cache,
they are tracked by PartitionedFilterBlockReader::filter_map_ and we can build
a test that exercises that heavily. Build DB with
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters
```
and test with (simultaneous runs with & without folly, ~20 times each to see
convergence)
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench_folly -readonly -use_existing_db -benchmarks=readrandom -num=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters -duration=40 -pin_l0_filter_and_index_blocks_in_cache
```
Average ops/s no folly: 26229.2
Average ops/s with folly: 26853.3 (+2.4%)
Reviewed By: ajkr
Differential Revision: D34181736
Pulled By: pdillinger
fbshipit-source-id: ffa6ad5104c2880321d8a1aa7187e00ab0d02e94
3 years ago
|
|
|
EXPECT_EQ(ConstexprFloorLog2(vm1), i - 1);
|
|
|
|
EXPECT_EQ(FloorLog2(everyOtherBit & vm1), (i - 1) & ~1);
|
Meta-internal folly integration with F14FastMap (#9546)
Summary:
Especially after updating to C++17, I don't see a compelling case for
*requiring* any folly components in RocksDB. I was able to purge the existing
hard dependencies, and it can be quite difficult to strip out non-trivial components
from folly for use in RocksDB. (The prospect of doing that on F14 has changed
my mind on the best approach here.)
But this change creates an optional integration where we can plug in
components from folly at compile time, starting here with F14FastMap to replace
std::unordered_map when possible (probably no public APIs for example). I have
replaced the biggest CPU users of std::unordered_map with compile-time
pluggable UnorderedMap which will use F14FastMap when USE_FOLLY is set.
USE_FOLLY is always set in the Meta-internal buck build, and a simulation of
that is in the Makefile for public CI testing. A full folly build is not needed, but
checking out the full folly repo is much simpler for getting the dependency,
and anything else we might want to optionally integrate in the future.
Some picky details:
* I don't think the distributed mutex stuff is actually used, so it was easy to remove.
* I implemented an alternative to `folly::constexpr_log2` (which is much easier
in C++17 than C++11) so that I could pull out the hard dependencies on
`ConstexprMath.h`
* I had to add noexcept move constructors/operators to some types to make
F14's complainUnlessNothrowMoveAndDestroy check happy, and I added a
macro to make that easier in some common cases.
* Updated Meta-internal buck build to use folly F14Map (always)
No updates to HISTORY.md nor INSTALL.md as this is not (yet?) considered a
production integration for open source users.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9546
Test Plan:
CircleCI tests updated so that a couple of them use folly.
Most internal unit & stress/crash tests updated to use Meta-internal latest folly.
(Note: they should probably use buck but they currently use Makefile.)
Example performance improvement: when filter partitions are pinned in cache,
they are tracked by PartitionedFilterBlockReader::filter_map_ and we can build
a test that exercises that heavily. Build DB with
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters
```
and test with (simultaneous runs with & without folly, ~20 times each to see
convergence)
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench_folly -readonly -use_existing_db -benchmarks=readrandom -num=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters -duration=40 -pin_l0_filter_and_index_blocks_in_cache
```
Average ops/s no folly: 26229.2
Average ops/s with folly: 26853.3 (+2.4%)
Reviewed By: ajkr
Differential Revision: D34181736
Pulled By: pdillinger
fbshipit-source-id: ffa6ad5104c2880321d8a1aa7187e00ab0d02e94
3 years ago
|
|
|
EXPECT_EQ(ConstexprFloorLog2(everyOtherBit & vm1), (i - 1) & ~1);
|
|
|
|
}
|
|
|
|
|
|
|
|
// CountTrailingZeroBits
|
|
|
|
if (v != 0) {
|
|
|
|
EXPECT_EQ(CountTrailingZeroBits(v), i);
|
|
|
|
}
|
|
|
|
if (vm1 != 0) {
|
|
|
|
EXPECT_EQ(CountTrailingZeroBits(vm1), 0);
|
|
|
|
}
|
|
|
|
if (i < int{8 * sizeof(T)} - 1) {
|
|
|
|
EXPECT_EQ(CountTrailingZeroBits(~vm1 & everyOtherBit), (i + 1) & ~1);
|
|
|
|
}
|
|
|
|
|
|
|
|
// BitsSetToOne
|
|
|
|
EXPECT_EQ(BitsSetToOne(v), 1);
|
|
|
|
EXPECT_EQ(BitsSetToOne(vm1), i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(vm1 & everyOtherBit), (i + 1) / 2);
|
|
|
|
|
|
|
|
// BitParity
|
|
|
|
EXPECT_EQ(BitParity(v), 1);
|
|
|
|
EXPECT_EQ(BitParity(vm1), i & 1);
|
|
|
|
EXPECT_EQ(BitParity(vm1 & everyOtherBit), ((i + 1) / 2) & 1);
|
|
|
|
|
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
|
|
|
// EndianSwapValue
|
|
|
|
T ev = T{1} << (((sizeof(T) - 1 - (i / 8)) * 8) + i % 8);
|
|
|
|
EXPECT_EQ(EndianSwapValue(v), ev);
|
|
|
|
|
|
|
|
// ReverseBits
|
|
|
|
EXPECT_EQ(ReverseBits(v), static_cast<T>(T{1} << (8 * sizeof(T) - 1 - i)));
|
|
|
|
#ifdef HAVE_UINT128_EXTENSION // Uses multiplication
|
|
|
|
if (std::is_unsigned<T>::value) { // Technical UB on signed type
|
|
|
|
T rv = T{1} << (8 * sizeof(T) - 1 - i);
|
|
|
|
EXPECT_EQ(ReverseBits(vm1), static_cast<T>(rv * ~T{1}));
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
vm1 = (vm1 << 1) | 1;
|
|
|
|
}
|
Meta-internal folly integration with F14FastMap (#9546)
Summary:
Especially after updating to C++17, I don't see a compelling case for
*requiring* any folly components in RocksDB. I was able to purge the existing
hard dependencies, and it can be quite difficult to strip out non-trivial components
from folly for use in RocksDB. (The prospect of doing that on F14 has changed
my mind on the best approach here.)
But this change creates an optional integration where we can plug in
components from folly at compile time, starting here with F14FastMap to replace
std::unordered_map when possible (probably no public APIs for example). I have
replaced the biggest CPU users of std::unordered_map with compile-time
pluggable UnorderedMap which will use F14FastMap when USE_FOLLY is set.
USE_FOLLY is always set in the Meta-internal buck build, and a simulation of
that is in the Makefile for public CI testing. A full folly build is not needed, but
checking out the full folly repo is much simpler for getting the dependency,
and anything else we might want to optionally integrate in the future.
Some picky details:
* I don't think the distributed mutex stuff is actually used, so it was easy to remove.
* I implemented an alternative to `folly::constexpr_log2` (which is much easier
in C++17 than C++11) so that I could pull out the hard dependencies on
`ConstexprMath.h`
* I had to add noexcept move constructors/operators to some types to make
F14's complainUnlessNothrowMoveAndDestroy check happy, and I added a
macro to make that easier in some common cases.
* Updated Meta-internal buck build to use folly F14Map (always)
No updates to HISTORY.md nor INSTALL.md as this is not (yet?) considered a
production integration for open source users.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9546
Test Plan:
CircleCI tests updated so that a couple of them use folly.
Most internal unit & stress/crash tests updated to use Meta-internal latest folly.
(Note: they should probably use buck but they currently use Makefile.)
Example performance improvement: when filter partitions are pinned in cache,
they are tracked by PartitionedFilterBlockReader::filter_map_ and we can build
a test that exercises that heavily. Build DB with
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters
```
and test with (simultaneous runs with & without folly, ~20 times each to see
convergence)
```
TEST_TMPDIR=/dev/shm/rocksdb ./db_bench_folly -readonly -use_existing_db -benchmarks=readrandom -num=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters -duration=40 -pin_l0_filter_and_index_blocks_in_cache
```
Average ops/s no folly: 26229.2
Average ops/s with folly: 26853.3 (+2.4%)
Reviewed By: ajkr
Differential Revision: D34181736
Pulled By: pdillinger
fbshipit-source-id: ffa6ad5104c2880321d8a1aa7187e00ab0d02e94
3 years ago
|
|
|
|
|
|
|
EXPECT_EQ(ConstexprFloorLog2(T{1}), 0);
|
|
|
|
EXPECT_EQ(ConstexprFloorLog2(T{2}), 1);
|
|
|
|
EXPECT_EQ(ConstexprFloorLog2(T{3}), 1);
|
|
|
|
EXPECT_EQ(ConstexprFloorLog2(T{42}), 5);
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(MathTest, BitOps) {
|
|
|
|
test_BitOps<uint32_t>();
|
|
|
|
test_BitOps<uint64_t>();
|
|
|
|
test_BitOps<uint16_t>();
|
|
|
|
test_BitOps<uint8_t>();
|
|
|
|
test_BitOps<unsigned char>();
|
|
|
|
test_BitOps<unsigned short>();
|
|
|
|
test_BitOps<unsigned int>();
|
|
|
|
test_BitOps<unsigned long>();
|
|
|
|
test_BitOps<unsigned long long>();
|
|
|
|
test_BitOps<char>();
|
|
|
|
test_BitOps<size_t>();
|
|
|
|
test_BitOps<int32_t>();
|
|
|
|
test_BitOps<int64_t>();
|
|
|
|
test_BitOps<int16_t>();
|
|
|
|
test_BitOps<int8_t>();
|
|
|
|
test_BitOps<signed char>();
|
|
|
|
test_BitOps<short>();
|
|
|
|
test_BitOps<int>();
|
|
|
|
test_BitOps<long>();
|
|
|
|
test_BitOps<long long>();
|
|
|
|
test_BitOps<ptrdiff_t>();
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(MathTest, BitOps128) { test_BitOps<Unsigned128>(); }
|
|
|
|
|
|
|
|
TEST(MathTest, Math128) {
|
|
|
|
const Unsigned128 sixteenHexOnes = 0x1111111111111111U;
|
|
|
|
const Unsigned128 thirtyHexOnes = (sixteenHexOnes << 56) | sixteenHexOnes;
|
|
|
|
const Unsigned128 sixteenHexTwos = 0x2222222222222222U;
|
|
|
|
const Unsigned128 thirtyHexTwos = (sixteenHexTwos << 56) | sixteenHexTwos;
|
|
|
|
|
|
|
|
// v will slide from all hex ones to all hex twos
|
|
|
|
Unsigned128 v = thirtyHexOnes;
|
|
|
|
for (int i = 0; i <= 30; ++i) {
|
|
|
|
// Test bitwise operations
|
|
|
|
EXPECT_EQ(BitsSetToOne(v), 30);
|
|
|
|
EXPECT_EQ(BitsSetToOne(~v), 128 - 30);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v & thirtyHexOnes), 30 - i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v | thirtyHexOnes), 30 + i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v ^ thirtyHexOnes), 2 * i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v & thirtyHexTwos), i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v | thirtyHexTwos), 60 - i);
|
|
|
|
EXPECT_EQ(BitsSetToOne(v ^ thirtyHexTwos), 60 - 2 * i);
|
|
|
|
|
|
|
|
// Test comparisons
|
|
|
|
EXPECT_EQ(v == thirtyHexOnes, i == 0);
|
|
|
|
EXPECT_EQ(v == thirtyHexTwos, i == 30);
|
|
|
|
EXPECT_EQ(v > thirtyHexOnes, i > 0);
|
|
|
|
EXPECT_EQ(v > thirtyHexTwos, false);
|
|
|
|
EXPECT_EQ(v >= thirtyHexOnes, true);
|
|
|
|
EXPECT_EQ(v >= thirtyHexTwos, i == 30);
|
|
|
|
EXPECT_EQ(v < thirtyHexOnes, false);
|
|
|
|
EXPECT_EQ(v < thirtyHexTwos, i < 30);
|
|
|
|
EXPECT_EQ(v <= thirtyHexOnes, i == 0);
|
|
|
|
EXPECT_EQ(v <= thirtyHexTwos, true);
|
|
|
|
|
|
|
|
// Update v, clearing upper-most byte
|
|
|
|
v = ((v << 12) >> 8) | 0x2;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int i = 0; i < 128; ++i) {
|
|
|
|
// Test shifts
|
|
|
|
Unsigned128 sl = thirtyHexOnes << i;
|
|
|
|
Unsigned128 sr = thirtyHexOnes >> i;
|
|
|
|
EXPECT_EQ(BitsSetToOne(sl), std::min(30, 32 - i / 4));
|
|
|
|
EXPECT_EQ(BitsSetToOne(sr), std::max(0, 30 - (i + 3) / 4));
|
|
|
|
EXPECT_EQ(BitsSetToOne(sl & sr), i % 2 ? 0 : std::max(0, 30 - i / 2));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test 64x64->128 multiply
|
|
|
|
Unsigned128 product =
|
|
|
|
Multiply64to128(0x1111111111111111U, 0x2222222222222222U);
|
|
|
|
EXPECT_EQ(Lower64of128(product), 2295594818061633090U);
|
|
|
|
EXPECT_EQ(Upper64of128(product), 163971058432973792U);
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(MathTest, Coding128) {
|
|
|
|
const char *in = "_1234567890123456";
|
|
|
|
// Note: in + 1 is likely unaligned
|
|
|
|
Unsigned128 decoded = DecodeFixed128(in + 1);
|
|
|
|
EXPECT_EQ(Lower64of128(decoded), 0x3837363534333231U);
|
|
|
|
EXPECT_EQ(Upper64of128(decoded), 0x3635343332313039U);
|
|
|
|
char out[18];
|
|
|
|
out[0] = '_';
|
|
|
|
EncodeFixed128(out + 1, decoded);
|
|
|
|
out[17] = '\0';
|
|
|
|
EXPECT_EQ(std::string(in), std::string(out));
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(MathTest, CodingGeneric) {
|
|
|
|
const char *in = "_1234567890123456";
|
|
|
|
// Decode
|
|
|
|
// Note: in + 1 is likely unaligned
|
|
|
|
Unsigned128 decoded128 = DecodeFixedGeneric<Unsigned128>(in + 1);
|
|
|
|
EXPECT_EQ(Lower64of128(decoded128), 0x3837363534333231U);
|
|
|
|
EXPECT_EQ(Upper64of128(decoded128), 0x3635343332313039U);
|
|
|
|
|
|
|
|
uint64_t decoded64 = DecodeFixedGeneric<uint64_t>(in + 1);
|
|
|
|
EXPECT_EQ(decoded64, 0x3837363534333231U);
|
|
|
|
|
|
|
|
uint32_t decoded32 = DecodeFixedGeneric<uint32_t>(in + 1);
|
|
|
|
EXPECT_EQ(decoded32, 0x34333231U);
|
|
|
|
|
|
|
|
uint16_t decoded16 = DecodeFixedGeneric<uint16_t>(in + 1);
|
|
|
|
EXPECT_EQ(decoded16, 0x3231U);
|
|
|
|
|
|
|
|
// Encode
|
|
|
|
char out[18];
|
|
|
|
out[0] = '_';
|
|
|
|
memset(out + 1, '\0', 17);
|
|
|
|
EncodeFixedGeneric(out + 1, decoded128);
|
|
|
|
EXPECT_EQ(std::string(in), std::string(out));
|
|
|
|
|
|
|
|
memset(out + 1, '\0', 9);
|
|
|
|
EncodeFixedGeneric(out + 1, decoded64);
|
|
|
|
EXPECT_EQ(std::string("_12345678"), std::string(out));
|
|
|
|
|
|
|
|
memset(out + 1, '\0', 5);
|
|
|
|
EncodeFixedGeneric(out + 1, decoded32);
|
|
|
|
EXPECT_EQ(std::string("_1234"), std::string(out));
|
|
|
|
|
|
|
|
memset(out + 1, '\0', 3);
|
|
|
|
EncodeFixedGeneric(out + 1, decoded16);
|
|
|
|
EXPECT_EQ(std::string("_12"), std::string(out));
|
|
|
|
}
|
|
|
|
|
|
|
|
int main(int argc, char** argv) {
|
|
|
|
fprintf(stderr, "NPHash64 id: %x\n",
|
|
|
|
static_cast<int>(ROCKSDB_NAMESPACE::GetSliceNPHash64("RocksDB")));
|
|
|
|
::testing::InitGoogleTest(&argc, argv);
|
|
|
|
|
|
|
|
return RUN_ALL_TESTS();
|
|
|
|
}
|