2016-02-09 23:12:00 +00:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-15 23:03:42 +00:00
|
|
|
// 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).
|
2014-10-31 18:54:05 +00:00
|
|
|
//
|
|
|
|
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
|
|
// found in the LICENSE file. See the AUTHORS file for names of contributors.
|
|
|
|
|
|
|
|
#include "db/version_builder.h"
|
|
|
|
|
|
|
|
#include <algorithm>
|
2015-08-11 19:19:56 +00:00
|
|
|
#include <atomic>
|
2019-09-20 19:00:55 +00:00
|
|
|
#include <cinttypes>
|
2016-12-16 19:17:26 +00:00
|
|
|
#include <functional>
|
2017-08-24 23:05:16 +00:00
|
|
|
#include <map>
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
#include <memory>
|
2014-10-31 18:54:05 +00:00
|
|
|
#include <set>
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
#include <sstream>
|
2015-08-11 19:19:56 +00:00
|
|
|
#include <thread>
|
2014-12-10 04:33:43 +00:00
|
|
|
#include <unordered_map>
|
|
|
|
#include <unordered_set>
|
2015-09-02 17:25:20 +00:00
|
|
|
#include <utility>
|
2014-10-31 18:54:05 +00:00
|
|
|
#include <vector>
|
|
|
|
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
#include "cache/cache_reservation_manager.h"
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
#include "db/blob/blob_file_meta.h"
|
2014-10-31 18:54:05 +00:00
|
|
|
#include "db/dbformat.h"
|
2015-09-02 17:25:20 +00:00
|
|
|
#include "db/internal_stats.h"
|
2014-10-31 18:54:05 +00:00
|
|
|
#include "db/table_cache.h"
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
#include "db/version_edit.h"
|
2024-08-13 04:09:37 +00:00
|
|
|
#include "db/version_edit_handler.h"
|
2014-10-31 18:54:05 +00:00
|
|
|
#include "db/version_set.h"
|
2017-02-06 22:43:55 +00:00
|
|
|
#include "port/port.h"
|
2014-10-31 18:54:05 +00:00
|
|
|
#include "table/table_reader.h"
|
2019-08-29 21:06:07 +00:00
|
|
|
#include "util/string_util.h"
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
class VersionBuilder::Rep {
|
2024-08-13 04:09:37 +00:00
|
|
|
class NewestFirstBySeqNo {
|
|
|
|
public:
|
|
|
|
bool operator()(const FileMetaData* lhs, const FileMetaData* rhs) const {
|
|
|
|
assert(lhs);
|
|
|
|
assert(rhs);
|
|
|
|
|
|
|
|
if (lhs->fd.largest_seqno != rhs->fd.largest_seqno) {
|
|
|
|
return lhs->fd.largest_seqno > rhs->fd.largest_seqno;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (lhs->fd.smallest_seqno != rhs->fd.smallest_seqno) {
|
|
|
|
return lhs->fd.smallest_seqno > rhs->fd.smallest_seqno;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Break ties by file number
|
|
|
|
return lhs->fd.GetNumber() > rhs->fd.GetNumber();
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
class NewestFirstByEpochNumber {
|
|
|
|
private:
|
|
|
|
inline static const NewestFirstBySeqNo seqno_cmp;
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
public:
|
|
|
|
bool operator()(const FileMetaData* lhs, const FileMetaData* rhs) const {
|
|
|
|
assert(lhs);
|
|
|
|
assert(rhs);
|
2014-10-31 18:54:05 +00:00
|
|
|
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
if (lhs->epoch_number != rhs->epoch_number) {
|
|
|
|
return lhs->epoch_number > rhs->epoch_number;
|
|
|
|
} else {
|
|
|
|
return seqno_cmp(lhs, rhs);
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
};
|
2021-11-03 18:51:22 +00:00
|
|
|
class BySmallestKey {
|
|
|
|
public:
|
|
|
|
explicit BySmallestKey(const InternalKeyComparator* cmp) : cmp_(cmp) {}
|
|
|
|
|
|
|
|
bool operator()(const FileMetaData* lhs, const FileMetaData* rhs) const {
|
|
|
|
assert(lhs);
|
|
|
|
assert(rhs);
|
|
|
|
assert(cmp_);
|
|
|
|
|
|
|
|
const int r = cmp_->Compare(lhs->smallest, rhs->smallest);
|
|
|
|
if (r != 0) {
|
|
|
|
return (r < 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Break ties by file number
|
|
|
|
return (lhs->fd.GetNumber() < rhs->fd.GetNumber());
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
const InternalKeyComparator* cmp_;
|
|
|
|
};
|
|
|
|
|
2014-10-31 18:54:05 +00:00
|
|
|
struct LevelState {
|
2014-12-10 04:33:43 +00:00
|
|
|
std::unordered_set<uint64_t> deleted_files;
|
|
|
|
// Map from file number to file meta data.
|
|
|
|
std::unordered_map<uint64_t, FileMetaData*> added_files;
|
2014-10-31 18:54:05 +00:00
|
|
|
};
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
// A class that represents the accumulated changes (like additional garbage or
|
|
|
|
// newly linked/unlinked SST files) for a given blob file after applying a
|
|
|
|
// series of VersionEdits.
|
2020-05-19 16:57:49 +00:00
|
|
|
class BlobFileMetaDataDelta {
|
|
|
|
public:
|
|
|
|
bool IsEmpty() const {
|
2021-10-30 00:45:35 +00:00
|
|
|
return !additional_garbage_count_ && !additional_garbage_bytes_ &&
|
|
|
|
newly_linked_ssts_.empty() && newly_unlinked_ssts_.empty();
|
2020-05-19 16:57:49 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
uint64_t GetAdditionalGarbageCount() const {
|
|
|
|
return additional_garbage_count_;
|
|
|
|
}
|
|
|
|
|
|
|
|
uint64_t GetAdditionalGarbageBytes() const {
|
|
|
|
return additional_garbage_bytes_;
|
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
const std::unordered_set<uint64_t>& GetNewlyLinkedSsts() const {
|
|
|
|
return newly_linked_ssts_;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::unordered_set<uint64_t>& GetNewlyUnlinkedSsts() const {
|
|
|
|
return newly_unlinked_ssts_;
|
|
|
|
}
|
|
|
|
|
2020-05-19 16:57:49 +00:00
|
|
|
void AddGarbage(uint64_t count, uint64_t bytes) {
|
|
|
|
additional_garbage_count_ += count;
|
|
|
|
additional_garbage_bytes_ += bytes;
|
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
void LinkSst(uint64_t sst_file_number) {
|
|
|
|
assert(newly_linked_ssts_.find(sst_file_number) ==
|
|
|
|
newly_linked_ssts_.end());
|
|
|
|
|
|
|
|
// Reconcile with newly unlinked SSTs on the fly. (Note: an SST can be
|
|
|
|
// linked to and unlinked from the same blob file in the case of a trivial
|
|
|
|
// move.)
|
|
|
|
auto it = newly_unlinked_ssts_.find(sst_file_number);
|
|
|
|
|
|
|
|
if (it != newly_unlinked_ssts_.end()) {
|
|
|
|
newly_unlinked_ssts_.erase(it);
|
|
|
|
} else {
|
|
|
|
newly_linked_ssts_.emplace(sst_file_number);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void UnlinkSst(uint64_t sst_file_number) {
|
|
|
|
assert(newly_unlinked_ssts_.find(sst_file_number) ==
|
|
|
|
newly_unlinked_ssts_.end());
|
|
|
|
|
|
|
|
// Reconcile with newly linked SSTs on the fly. (Note: an SST can be
|
|
|
|
// linked to and unlinked from the same blob file in the case of a trivial
|
|
|
|
// move.)
|
|
|
|
auto it = newly_linked_ssts_.find(sst_file_number);
|
|
|
|
|
|
|
|
if (it != newly_linked_ssts_.end()) {
|
|
|
|
newly_linked_ssts_.erase(it);
|
|
|
|
} else {
|
|
|
|
newly_unlinked_ssts_.emplace(sst_file_number);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-19 16:57:49 +00:00
|
|
|
private:
|
|
|
|
uint64_t additional_garbage_count_ = 0;
|
|
|
|
uint64_t additional_garbage_bytes_ = 0;
|
2020-06-12 16:52:14 +00:00
|
|
|
std::unordered_set<uint64_t> newly_linked_ssts_;
|
|
|
|
std::unordered_set<uint64_t> newly_unlinked_ssts_;
|
2020-05-19 16:57:49 +00:00
|
|
|
};
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
// A class that represents the state of a blob file after applying a series of
|
|
|
|
// VersionEdits. In addition to the resulting state, it also contains the
|
|
|
|
// delta (see BlobFileMetaDataDelta above). The resulting state can be used to
|
|
|
|
// identify obsolete blob files, while the delta makes it possible to
|
|
|
|
// efficiently detect trivial moves.
|
|
|
|
class MutableBlobFileMetaData {
|
|
|
|
public:
|
|
|
|
// To be used for brand new blob files
|
|
|
|
explicit MutableBlobFileMetaData(
|
|
|
|
std::shared_ptr<SharedBlobFileMetaData>&& shared_meta)
|
|
|
|
: shared_meta_(std::move(shared_meta)) {}
|
|
|
|
|
|
|
|
// To be used for pre-existing blob files
|
|
|
|
explicit MutableBlobFileMetaData(
|
|
|
|
const std::shared_ptr<BlobFileMetaData>& meta)
|
|
|
|
: shared_meta_(meta->GetSharedMeta()),
|
|
|
|
linked_ssts_(meta->GetLinkedSsts()),
|
|
|
|
garbage_blob_count_(meta->GetGarbageBlobCount()),
|
|
|
|
garbage_blob_bytes_(meta->GetGarbageBlobBytes()) {}
|
|
|
|
|
|
|
|
const std::shared_ptr<SharedBlobFileMetaData>& GetSharedMeta() const {
|
|
|
|
return shared_meta_;
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
uint64_t GetBlobFileNumber() const {
|
|
|
|
assert(shared_meta_);
|
|
|
|
return shared_meta_->GetBlobFileNumber();
|
|
|
|
}
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
bool HasDelta() const { return !delta_.IsEmpty(); }
|
|
|
|
|
|
|
|
const std::unordered_set<uint64_t>& GetLinkedSsts() const {
|
|
|
|
return linked_ssts_;
|
|
|
|
}
|
|
|
|
|
|
|
|
uint64_t GetGarbageBlobCount() const { return garbage_blob_count_; }
|
|
|
|
|
|
|
|
uint64_t GetGarbageBlobBytes() const { return garbage_blob_bytes_; }
|
|
|
|
|
2021-11-01 19:31:13 +00:00
|
|
|
bool AddGarbage(uint64_t count, uint64_t bytes) {
|
|
|
|
assert(shared_meta_);
|
|
|
|
|
|
|
|
if (garbage_blob_count_ + count > shared_meta_->GetTotalBlobCount() ||
|
|
|
|
garbage_blob_bytes_ + bytes > shared_meta_->GetTotalBlobBytes()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
delta_.AddGarbage(count, bytes);
|
|
|
|
|
|
|
|
garbage_blob_count_ += count;
|
|
|
|
garbage_blob_bytes_ += bytes;
|
2021-11-01 19:31:13 +00:00
|
|
|
|
|
|
|
return true;
|
2021-10-30 00:45:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void LinkSst(uint64_t sst_file_number) {
|
|
|
|
delta_.LinkSst(sst_file_number);
|
|
|
|
|
|
|
|
assert(linked_ssts_.find(sst_file_number) == linked_ssts_.end());
|
|
|
|
linked_ssts_.emplace(sst_file_number);
|
|
|
|
}
|
|
|
|
|
|
|
|
void UnlinkSst(uint64_t sst_file_number) {
|
|
|
|
delta_.UnlinkSst(sst_file_number);
|
|
|
|
|
|
|
|
assert(linked_ssts_.find(sst_file_number) != linked_ssts_.end());
|
|
|
|
linked_ssts_.erase(sst_file_number);
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
std::shared_ptr<SharedBlobFileMetaData> shared_meta_;
|
|
|
|
// Accumulated changes
|
|
|
|
BlobFileMetaDataDelta delta_;
|
|
|
|
// Resulting state after applying the changes
|
|
|
|
BlobFileMetaData::LinkedSsts linked_ssts_;
|
|
|
|
uint64_t garbage_blob_count_ = 0;
|
|
|
|
uint64_t garbage_blob_bytes_ = 0;
|
|
|
|
};
|
|
|
|
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 22:47:08 +00:00
|
|
|
const FileOptions& file_options_;
|
2020-04-30 18:23:32 +00:00
|
|
|
const ImmutableCFOptions* const ioptions_;
|
2014-10-31 18:54:05 +00:00
|
|
|
TableCache* table_cache_;
|
|
|
|
VersionStorageInfo* base_vstorage_;
|
2020-04-30 18:23:32 +00:00
|
|
|
VersionSet* version_set_;
|
2017-08-24 23:05:16 +00:00
|
|
|
int num_levels_;
|
2014-10-31 18:54:05 +00:00
|
|
|
LevelState* levels_;
|
2020-06-03 18:21:16 +00:00
|
|
|
// Store sizes of levels larger than num_levels_. We do this instead of
|
2017-08-24 23:05:16 +00:00
|
|
|
// storing them in levels_ to avoid regression in case there are no files
|
|
|
|
// on invalid levels. The version is not consistent if in the end the files
|
|
|
|
// on invalid levels don't cancel out.
|
2020-06-03 18:21:16 +00:00
|
|
|
std::unordered_map<int, size_t> invalid_level_sizes_;
|
2017-08-24 23:05:16 +00:00
|
|
|
// Whether there are invalid new files or invalid deletion on levels larger
|
|
|
|
// than num_levels_.
|
|
|
|
bool has_invalid_levels_;
|
2020-06-03 18:21:16 +00:00
|
|
|
// Current levels of table files affected by additions/deletions.
|
|
|
|
std::unordered_map<uint64_t, int> table_file_levels_;
|
2022-06-21 18:56:53 +00:00
|
|
|
// Current compact cursors that should be changed after the last compaction
|
|
|
|
std::unordered_map<int, InternalKey> updated_compact_cursors_;
|
2024-08-21 17:09:10 +00:00
|
|
|
const std::shared_ptr<const NewestFirstByEpochNumber>
|
|
|
|
level_zero_cmp_by_epochno_;
|
|
|
|
const std::shared_ptr<const NewestFirstBySeqNo> level_zero_cmp_by_seqno_;
|
|
|
|
const std::shared_ptr<const BySmallestKey> level_nonzero_cmp_;
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
// Mutable metadata objects for all blob files affected by the series of
|
|
|
|
// version edits.
|
|
|
|
std::map<uint64_t, MutableBlobFileMetaData> mutable_blob_file_metas_;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
std::shared_ptr<CacheReservationManager> file_metadata_cache_res_mgr_;
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
ColumnFamilyData* cfd_;
|
|
|
|
VersionEditHandler* version_edit_handler_;
|
|
|
|
bool track_found_and_missing_files_;
|
2024-08-17 00:18:54 +00:00
|
|
|
// If false, only a complete Version with all files consisting it found is
|
|
|
|
// considered valid. If true, besides complete Version, if the Version is
|
|
|
|
// never edited in an atomic group, an incomplete Version with only a suffix
|
|
|
|
// of L0 files missing is also considered valid.
|
|
|
|
bool allow_incomplete_valid_version_;
|
2024-08-13 04:09:37 +00:00
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
// These are only tracked if `track_found_and_missing_files_` is enabled.
|
|
|
|
|
|
|
|
// The SST files that are found (blob files not included yet).
|
2024-08-13 04:09:37 +00:00
|
|
|
std::unordered_set<uint64_t> found_files_;
|
2024-08-17 00:18:54 +00:00
|
|
|
// Missing SST files for L0
|
|
|
|
std::unordered_set<uint64_t> l0_missing_files_;
|
|
|
|
// Missing SST files for non L0 levels
|
|
|
|
std::unordered_set<uint64_t> non_l0_missing_files_;
|
|
|
|
// Intermediate SST files (blob files not included yet)
|
2024-08-13 04:09:37 +00:00
|
|
|
std::vector<std::string> intermediate_files_;
|
2024-08-17 00:18:54 +00:00
|
|
|
// The highest file number for all the missing blob files, useful to check
|
|
|
|
// if a complete Version is available.
|
2024-08-13 04:09:37 +00:00
|
|
|
uint64_t missing_blob_files_high_ = kInvalidBlobFileNumber;
|
2024-08-17 00:18:54 +00:00
|
|
|
// Missing blob files, useful to check if only the missing L0 files'
|
|
|
|
// associated blob files are missing.
|
|
|
|
std::unordered_set<uint64_t> missing_blob_files_;
|
|
|
|
// True if all files consisting the Version can be found. Or if
|
|
|
|
// `allow_incomplete_valid_version_` is true and the version history is not
|
|
|
|
// ever edited in an atomic group, this will be true if only a
|
|
|
|
// suffix of L0 SST files and their associated blob files are missing.
|
|
|
|
bool valid_version_available_;
|
|
|
|
// True if version is ever edited in an atomic group.
|
|
|
|
bool edited_in_atomic_group_;
|
|
|
|
|
|
|
|
// Flag to indicate if the Version is updated since last validity check. If no
|
|
|
|
// `Apply` call is made between a `Rep`'s construction and a
|
|
|
|
// `ValidVersionAvailable` check or between two `ValidVersionAvailable` calls.
|
|
|
|
// This flag will be true to indicate the cached validity value can be
|
|
|
|
// directly used without a recheck.
|
|
|
|
bool version_updated_since_last_check_;
|
|
|
|
|
|
|
|
// End of fields that are only tracked when `track_found_and_missing_files_`
|
|
|
|
// is enabled.
|
2024-08-13 04:09:37 +00:00
|
|
|
|
2014-10-31 18:54:05 +00:00
|
|
|
public:
|
2020-04-30 18:23:32 +00:00
|
|
|
Rep(const FileOptions& file_options, const ImmutableCFOptions* ioptions,
|
|
|
|
TableCache* table_cache, VersionStorageInfo* base_vstorage,
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
VersionSet* version_set,
|
2024-08-13 04:09:37 +00:00
|
|
|
std::shared_ptr<CacheReservationManager> file_metadata_cache_res_mgr,
|
|
|
|
ColumnFamilyData* cfd, VersionEditHandler* version_edit_handler,
|
2024-08-17 00:18:54 +00:00
|
|
|
bool track_found_and_missing_files, bool allow_incomplete_valid_version)
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 22:47:08 +00:00
|
|
|
: file_options_(file_options),
|
2020-04-30 18:23:32 +00:00
|
|
|
ioptions_(ioptions),
|
2014-10-31 18:54:05 +00:00
|
|
|
table_cache_(table_cache),
|
2017-08-24 23:05:16 +00:00
|
|
|
base_vstorage_(base_vstorage),
|
2020-04-30 18:23:32 +00:00
|
|
|
version_set_(version_set),
|
2017-08-24 23:05:16 +00:00
|
|
|
num_levels_(base_vstorage->num_levels()),
|
2021-11-03 18:51:22 +00:00
|
|
|
has_invalid_levels_(false),
|
2024-08-13 04:09:37 +00:00
|
|
|
level_zero_cmp_by_epochno_(
|
|
|
|
std::make_shared<NewestFirstByEpochNumber>()),
|
|
|
|
level_zero_cmp_by_seqno_(std::make_shared<NewestFirstBySeqNo>()),
|
|
|
|
level_nonzero_cmp_(std::make_shared<BySmallestKey>(
|
|
|
|
base_vstorage_->InternalComparator())),
|
|
|
|
file_metadata_cache_res_mgr_(file_metadata_cache_res_mgr),
|
|
|
|
cfd_(cfd),
|
|
|
|
version_edit_handler_(version_edit_handler),
|
2024-08-17 00:18:54 +00:00
|
|
|
track_found_and_missing_files_(track_found_and_missing_files),
|
|
|
|
allow_incomplete_valid_version_(allow_incomplete_valid_version) {
|
2020-04-30 18:23:32 +00:00
|
|
|
assert(ioptions_);
|
|
|
|
|
2017-08-24 23:05:16 +00:00
|
|
|
levels_ = new LevelState[num_levels_];
|
2024-08-13 04:09:37 +00:00
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
assert(cfd_);
|
|
|
|
assert(version_edit_handler_);
|
2024-08-17 00:18:54 +00:00
|
|
|
// `track_found_and_missing_files_` mode used by VersionEditHandlerPIT
|
|
|
|
// assumes the initial base version is valid. For best efforts recovery,
|
|
|
|
// base will be empty. For manifest tailing usage like secondary instance,
|
|
|
|
// they do not allow incomplete version, so the base version in subsequent
|
|
|
|
// catch up attempts should be valid too.
|
|
|
|
valid_version_available_ = true;
|
|
|
|
edited_in_atomic_group_ = false;
|
|
|
|
version_updated_since_last_check_ = false;
|
2024-08-13 04:09:37 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Rep(const Rep& other)
|
|
|
|
: file_options_(other.file_options_),
|
|
|
|
ioptions_(other.ioptions_),
|
|
|
|
table_cache_(other.table_cache_),
|
|
|
|
base_vstorage_(other.base_vstorage_),
|
|
|
|
version_set_(other.version_set_),
|
|
|
|
num_levels_(other.num_levels_),
|
|
|
|
invalid_level_sizes_(other.invalid_level_sizes_),
|
|
|
|
has_invalid_levels_(other.has_invalid_levels_),
|
|
|
|
table_file_levels_(other.table_file_levels_),
|
|
|
|
updated_compact_cursors_(other.updated_compact_cursors_),
|
|
|
|
level_zero_cmp_by_epochno_(other.level_zero_cmp_by_epochno_),
|
|
|
|
level_zero_cmp_by_seqno_(other.level_zero_cmp_by_seqno_),
|
|
|
|
level_nonzero_cmp_(other.level_nonzero_cmp_),
|
|
|
|
mutable_blob_file_metas_(other.mutable_blob_file_metas_),
|
|
|
|
file_metadata_cache_res_mgr_(other.file_metadata_cache_res_mgr_),
|
|
|
|
cfd_(other.cfd_),
|
|
|
|
version_edit_handler_(other.version_edit_handler_),
|
|
|
|
track_found_and_missing_files_(other.track_found_and_missing_files_),
|
2024-08-17 00:18:54 +00:00
|
|
|
allow_incomplete_valid_version_(other.allow_incomplete_valid_version_),
|
2024-08-13 04:09:37 +00:00
|
|
|
found_files_(other.found_files_),
|
2024-08-17 00:18:54 +00:00
|
|
|
l0_missing_files_(other.l0_missing_files_),
|
|
|
|
non_l0_missing_files_(other.non_l0_missing_files_),
|
2024-08-13 04:09:37 +00:00
|
|
|
intermediate_files_(other.intermediate_files_),
|
2024-08-17 00:18:54 +00:00
|
|
|
missing_blob_files_high_(other.missing_blob_files_high_),
|
|
|
|
missing_blob_files_(other.missing_blob_files_),
|
|
|
|
valid_version_available_(other.valid_version_available_),
|
|
|
|
edited_in_atomic_group_(other.edited_in_atomic_group_),
|
|
|
|
version_updated_since_last_check_(
|
|
|
|
other.version_updated_since_last_check_) {
|
2024-08-13 04:09:37 +00:00
|
|
|
assert(ioptions_);
|
|
|
|
levels_ = new LevelState[num_levels_];
|
|
|
|
for (int level = 0; level < num_levels_; level++) {
|
|
|
|
levels_[level] = other.levels_[level];
|
|
|
|
const auto& added = levels_[level].added_files;
|
|
|
|
for (auto& pair : added) {
|
|
|
|
RefFile(pair.second);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
assert(cfd_);
|
|
|
|
assert(version_edit_handler_);
|
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
~Rep() {
|
2017-08-24 23:05:16 +00:00
|
|
|
for (int level = 0; level < num_levels_; level++) {
|
2014-12-10 04:33:43 +00:00
|
|
|
const auto& added = levels_[level].added_files;
|
|
|
|
for (auto& pair : added) {
|
2015-01-07 18:29:21 +00:00
|
|
|
UnrefFile(pair.second);
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
delete[] levels_;
|
|
|
|
}
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
void RefFile(FileMetaData* f) {
|
|
|
|
assert(f);
|
|
|
|
assert(f->refs > 0);
|
|
|
|
f->refs++;
|
|
|
|
}
|
|
|
|
|
2015-01-07 18:29:21 +00:00
|
|
|
void UnrefFile(FileMetaData* f) {
|
|
|
|
f->refs--;
|
|
|
|
if (f->refs <= 0) {
|
|
|
|
if (f->table_reader_handle) {
|
|
|
|
assert(table_cache_ != nullptr);
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
// NOTE: have to release in raw cache interface to avoid using a
|
|
|
|
// TypedHandle for FileMetaData::table_reader_handle
|
|
|
|
table_cache_->get_cache().get()->Release(f->table_reader_handle);
|
2015-01-07 18:29:21 +00:00
|
|
|
f->table_reader_handle = nullptr;
|
|
|
|
}
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
|
|
|
|
if (file_metadata_cache_res_mgr_) {
|
|
|
|
Status s = file_metadata_cache_res_mgr_->UpdateCacheReservation(
|
|
|
|
f->ApproximateMemoryUsage(), false /* increase */);
|
|
|
|
s.PermitUncheckedError();
|
|
|
|
}
|
2015-01-07 18:29:21 +00:00
|
|
|
delete f;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
// Mapping used for checking the consistency of links between SST files and
|
|
|
|
// blob files. It is built using the forward links (table file -> blob file),
|
|
|
|
// and is subsequently compared with the inverse mapping stored in the
|
|
|
|
// BlobFileMetaData objects.
|
2020-06-12 16:52:14 +00:00
|
|
|
using ExpectedLinkedSsts =
|
|
|
|
std::unordered_map<uint64_t, BlobFileMetaData::LinkedSsts>;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
static void UpdateExpectedLinkedSsts(
|
|
|
|
uint64_t table_file_number, uint64_t blob_file_number,
|
|
|
|
ExpectedLinkedSsts* expected_linked_ssts) {
|
|
|
|
assert(expected_linked_ssts);
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
|
|
|
if (blob_file_number == kInvalidBlobFileNumber) {
|
2020-06-12 16:52:14 +00:00
|
|
|
return;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
(*expected_linked_ssts)[blob_file_number].emplace(table_file_number);
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
template <typename Checker>
|
2021-11-03 18:51:22 +00:00
|
|
|
Status CheckConsistencyDetailsForLevel(
|
|
|
|
const VersionStorageInfo* vstorage, int level, Checker checker,
|
|
|
|
const std::string& sync_point,
|
|
|
|
ExpectedLinkedSsts* expected_linked_ssts) const {
|
|
|
|
#ifdef NDEBUG
|
|
|
|
(void)sync_point;
|
|
|
|
#endif
|
2020-06-12 16:52:14 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
assert(vstorage);
|
|
|
|
assert(level >= 0 && level < num_levels_);
|
|
|
|
assert(expected_linked_ssts);
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
const auto& level_files = vstorage->LevelFiles(level);
|
|
|
|
|
|
|
|
if (level_files.empty()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(level_files[0]);
|
|
|
|
UpdateExpectedLinkedSsts(level_files[0]->fd.GetNumber(),
|
|
|
|
level_files[0]->oldest_blob_file_number,
|
|
|
|
expected_linked_ssts);
|
|
|
|
|
|
|
|
for (size_t i = 1; i < level_files.size(); ++i) {
|
|
|
|
assert(level_files[i]);
|
|
|
|
UpdateExpectedLinkedSsts(level_files[i]->fd.GetNumber(),
|
|
|
|
level_files[i]->oldest_blob_file_number,
|
|
|
|
expected_linked_ssts);
|
|
|
|
|
|
|
|
auto lhs = level_files[i - 1];
|
|
|
|
auto rhs = level_files[i];
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2019-08-29 21:06:07 +00:00
|
|
|
#ifndef NDEBUG
|
2021-11-03 18:51:22 +00:00
|
|
|
auto pair = std::make_pair(&lhs, &rhs);
|
|
|
|
TEST_SYNC_POINT_CALLBACK(sync_point, &pair);
|
2019-08-29 21:06:07 +00:00
|
|
|
#endif
|
2021-11-03 18:51:22 +00:00
|
|
|
|
|
|
|
const Status s = checker(lhs, rhs);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Make sure table files are sorted correctly and that the links between
|
|
|
|
// table files and blob files are consistent.
|
|
|
|
Status CheckConsistencyDetails(const VersionStorageInfo* vstorage) const {
|
|
|
|
assert(vstorage);
|
|
|
|
|
|
|
|
ExpectedLinkedSsts expected_linked_ssts;
|
|
|
|
|
|
|
|
if (num_levels_ > 0) {
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
const InternalKeyComparator* const icmp = vstorage->InternalComparator();
|
|
|
|
EpochNumberRequirement epoch_number_requirement =
|
|
|
|
vstorage->GetEpochNumberRequirement();
|
|
|
|
assert(icmp);
|
2021-11-03 18:51:22 +00:00
|
|
|
// Check L0
|
|
|
|
{
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
auto l0_checker = [this, epoch_number_requirement, icmp](
|
|
|
|
const FileMetaData* lhs,
|
|
|
|
const FileMetaData* rhs) {
|
2021-11-03 18:51:22 +00:00
|
|
|
assert(lhs);
|
|
|
|
assert(rhs);
|
|
|
|
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
if (epoch_number_requirement ==
|
|
|
|
EpochNumberRequirement::kMightMissing) {
|
2024-08-13 04:09:37 +00:00
|
|
|
if (!level_zero_cmp_by_seqno_->operator()(lhs, rhs)) {
|
2021-11-03 18:51:22 +00:00
|
|
|
std::ostringstream oss;
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
oss << "L0 files are not sorted properly: files #"
|
|
|
|
<< lhs->fd.GetNumber() << " with seqnos (largest, smallest) "
|
|
|
|
<< lhs->fd.largest_seqno << " , " << lhs->fd.smallest_seqno
|
|
|
|
<< ", #" << rhs->fd.GetNumber()
|
|
|
|
<< " with seqnos (largest, smallest) "
|
|
|
|
<< rhs->fd.largest_seqno << " , " << rhs->fd.smallest_seqno;
|
2021-11-03 18:51:22 +00:00
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
2016-10-21 00:05:32 +00:00
|
|
|
}
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
} else if (epoch_number_requirement ==
|
|
|
|
EpochNumberRequirement::kMustPresent) {
|
|
|
|
if (lhs->epoch_number == rhs->epoch_number) {
|
|
|
|
bool range_overlapped =
|
|
|
|
icmp->Compare(lhs->smallest, rhs->largest) <= 0 &&
|
|
|
|
icmp->Compare(lhs->largest, rhs->smallest) >= 0;
|
|
|
|
|
|
|
|
if (range_overlapped) {
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "L0 files of same epoch number but overlapping range #"
|
|
|
|
<< lhs->fd.GetNumber()
|
|
|
|
<< " , smallest key: " << lhs->smallest.DebugString(true)
|
|
|
|
<< " , largest key: " << lhs->largest.DebugString(true)
|
|
|
|
<< " , epoch number: " << lhs->epoch_number << " vs. file #"
|
|
|
|
<< rhs->fd.GetNumber()
|
|
|
|
<< " , smallest key: " << rhs->smallest.DebugString(true)
|
|
|
|
<< " , largest key: " << rhs->largest.DebugString(true)
|
|
|
|
<< " , epoch number: " << rhs->epoch_number;
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
}
|
2021-11-03 18:51:22 +00:00
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
if (!level_zero_cmp_by_epochno_->operator()(lhs, rhs)) {
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "L0 files are not sorted properly: files #"
|
|
|
|
<< lhs->fd.GetNumber() << " with epoch number "
|
|
|
|
<< lhs->epoch_number << ", #" << rhs->fd.GetNumber()
|
|
|
|
<< " with epoch number " << rhs->epoch_number;
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
2016-10-08 00:21:45 +00:00
|
|
|
}
|
2021-11-03 18:51:22 +00:00
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
};
|
|
|
|
|
|
|
|
const Status s = CheckConsistencyDetailsForLevel(
|
|
|
|
vstorage, /* level */ 0, l0_checker,
|
|
|
|
"VersionBuilder::CheckConsistency0", &expected_linked_ssts);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check L1 and up
|
|
|
|
|
|
|
|
for (int level = 1; level < num_levels_; ++level) {
|
|
|
|
auto checker = [this, level, icmp](const FileMetaData* lhs,
|
|
|
|
const FileMetaData* rhs) {
|
|
|
|
assert(lhs);
|
|
|
|
assert(rhs);
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
if (!level_nonzero_cmp_->operator()(lhs, rhs)) {
|
2021-11-03 18:51:22 +00:00
|
|
|
std::ostringstream oss;
|
|
|
|
oss << 'L' << level << " files are not sorted properly: files #"
|
|
|
|
<< lhs->fd.GetNumber() << ", #" << rhs->fd.GetNumber();
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
2016-10-08 00:21:45 +00:00
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
// Make sure there is no overlap in level
|
|
|
|
if (icmp->Compare(lhs->largest, rhs->smallest) >= 0) {
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << 'L' << level << " has overlapping ranges: file #"
|
|
|
|
<< lhs->fd.GetNumber()
|
|
|
|
<< " largest key: " << lhs->largest.DebugString(true)
|
|
|
|
<< " vs. file #" << rhs->fd.GetNumber()
|
|
|
|
<< " smallest key: " << rhs->smallest.DebugString(true);
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
2021-11-03 18:51:22 +00:00
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
};
|
|
|
|
|
|
|
|
const Status s = CheckConsistencyDetailsForLevel(
|
|
|
|
vstorage, level, checker, "VersionBuilder::CheckConsistency1",
|
|
|
|
&expected_linked_ssts);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
// Make sure that all blob files in the version have non-garbage data and
|
|
|
|
// the links between them and the table files are consistent.
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
const auto& blob_files = vstorage->GetBlobFiles();
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
for (const auto& blob_file_meta : blob_files) {
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
assert(blob_file_meta);
|
|
|
|
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
const uint64_t blob_file_number = blob_file_meta->GetBlobFileNumber();
|
|
|
|
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
if (blob_file_meta->GetGarbageBlobCount() >=
|
|
|
|
blob_file_meta->GetTotalBlobCount()) {
|
|
|
|
std::ostringstream oss;
|
2020-06-12 16:52:14 +00:00
|
|
|
oss << "Blob file #" << blob_file_number
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
<< " consists entirely of garbage";
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
2020-06-12 16:52:14 +00:00
|
|
|
|
|
|
|
if (blob_file_meta->GetLinkedSsts() !=
|
|
|
|
expected_linked_ssts[blob_file_number]) {
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Links are inconsistent between table files and blob file #"
|
|
|
|
<< blob_file_number;
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
|
2020-05-04 21:15:55 +00:00
|
|
|
Status ret_s;
|
|
|
|
TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistencyBeforeReturn",
|
|
|
|
&ret_s);
|
|
|
|
return ret_s;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
Status CheckConsistency(const VersionStorageInfo* vstorage) const {
|
|
|
|
assert(vstorage);
|
|
|
|
|
2020-09-30 18:56:16 +00:00
|
|
|
// Always run consistency checks in debug build
|
|
|
|
#ifdef NDEBUG
|
|
|
|
if (!vstorage->force_consistency_checks()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
Status s = CheckConsistencyDetails(vstorage);
|
|
|
|
if (s.IsCorruption() && s.getState()) {
|
|
|
|
// Make it clear the error is due to force_consistency_checks = 1 or
|
|
|
|
// debug build
|
|
|
|
#ifdef NDEBUG
|
|
|
|
auto prefix = "force_consistency_checks";
|
|
|
|
#else
|
|
|
|
auto prefix = "force_consistency_checks(DEBUG)";
|
|
|
|
#endif
|
|
|
|
s = Status::Corruption(prefix, s.getState());
|
|
|
|
} else {
|
|
|
|
// was only expecting corruption with message, or OK
|
|
|
|
assert(s.ok());
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
bool CheckConsistencyForNumLevels() const {
|
2017-08-24 23:05:16 +00:00
|
|
|
// Make sure there are no files on or beyond num_levels().
|
|
|
|
if (has_invalid_levels_) {
|
|
|
|
return false;
|
|
|
|
}
|
2020-06-03 18:21:16 +00:00
|
|
|
|
|
|
|
for (const auto& pair : invalid_level_sizes_) {
|
|
|
|
const size_t level_size = pair.second;
|
|
|
|
if (level_size != 0) {
|
2017-08-24 23:05:16 +00:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
2020-06-03 18:21:16 +00:00
|
|
|
|
2017-08-24 23:05:16 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
bool IsBlobFileInVersion(uint64_t blob_file_number) const {
|
|
|
|
auto mutable_it = mutable_blob_file_metas_.find(blob_file_number);
|
|
|
|
if (mutable_it != mutable_blob_file_metas_.end()) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(base_vstorage_);
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
const auto meta = base_vstorage_->GetBlobFileMetaData(blob_file_number);
|
2021-10-30 00:45:35 +00:00
|
|
|
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
return !!meta;
|
2021-10-30 00:45:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
MutableBlobFileMetaData* GetOrCreateMutableBlobFileMetaData(
|
|
|
|
uint64_t blob_file_number) {
|
|
|
|
auto mutable_it = mutable_blob_file_metas_.find(blob_file_number);
|
|
|
|
if (mutable_it != mutable_blob_file_metas_.end()) {
|
|
|
|
return &mutable_it->second;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(base_vstorage_);
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
const auto meta = base_vstorage_->GetBlobFileMetaData(blob_file_number);
|
2021-10-30 00:45:35 +00:00
|
|
|
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
if (meta) {
|
2021-10-30 00:45:35 +00:00
|
|
|
mutable_it = mutable_blob_file_metas_
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
.emplace(blob_file_number, MutableBlobFileMetaData(meta))
|
2021-10-30 00:45:35 +00:00
|
|
|
.first;
|
|
|
|
return &mutable_it->second;
|
|
|
|
}
|
|
|
|
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
Status ApplyBlobFileAddition(const BlobFileAddition& blob_file_addition) {
|
|
|
|
const uint64_t blob_file_number = blob_file_addition.GetBlobFileNumber();
|
|
|
|
|
2020-05-19 16:57:49 +00:00
|
|
|
if (IsBlobFileInVersion(blob_file_number)) {
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Blob file #" << blob_file_number << " already added";
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
|
2020-04-30 18:23:32 +00:00
|
|
|
// Note: we use C++11 for now but in C++14, this could be done in a more
|
|
|
|
// elegant way using generalized lambda capture.
|
|
|
|
VersionSet* const vs = version_set_;
|
|
|
|
const ImmutableCFOptions* const ioptions = ioptions_;
|
|
|
|
|
|
|
|
auto deleter = [vs, ioptions](SharedBlobFileMetaData* shared_meta) {
|
|
|
|
if (vs) {
|
|
|
|
assert(ioptions);
|
|
|
|
assert(!ioptions->cf_paths.empty());
|
|
|
|
assert(shared_meta);
|
|
|
|
|
|
|
|
vs->AddObsoleteBlobFile(shared_meta->GetBlobFileNumber(),
|
|
|
|
ioptions->cf_paths.front().path);
|
|
|
|
}
|
|
|
|
|
|
|
|
delete shared_meta;
|
|
|
|
};
|
|
|
|
|
2020-04-23 20:41:32 +00:00
|
|
|
auto shared_meta = SharedBlobFileMetaData::Create(
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
blob_file_number, blob_file_addition.GetTotalBlobCount(),
|
|
|
|
blob_file_addition.GetTotalBlobBytes(),
|
|
|
|
blob_file_addition.GetChecksumMethod(),
|
2020-04-30 18:23:32 +00:00
|
|
|
blob_file_addition.GetChecksumValue(), deleter);
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
mutable_blob_file_metas_.emplace(
|
|
|
|
blob_file_number, MutableBlobFileMetaData(std::move(shared_meta)));
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
Status s;
|
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
assert(version_edit_handler_);
|
|
|
|
s = version_edit_handler_->VerifyBlobFile(cfd_, blob_file_number,
|
|
|
|
blob_file_addition);
|
|
|
|
if (s.IsPathNotFound() || s.IsNotFound() || s.IsCorruption()) {
|
|
|
|
missing_blob_files_high_ =
|
|
|
|
std::max(missing_blob_files_high_, blob_file_number);
|
2024-08-17 00:18:54 +00:00
|
|
|
missing_blob_files_.insert(blob_file_number);
|
2024-08-13 04:09:37 +00:00
|
|
|
s = Status::OK();
|
|
|
|
} else if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
Status ApplyBlobFileGarbage(const BlobFileGarbage& blob_file_garbage) {
|
|
|
|
const uint64_t blob_file_number = blob_file_garbage.GetBlobFileNumber();
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
MutableBlobFileMetaData* const mutable_meta =
|
|
|
|
GetOrCreateMutableBlobFileMetaData(blob_file_number);
|
|
|
|
|
|
|
|
if (!mutable_meta) {
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Blob file #" << blob_file_number << " not found";
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
|
2021-11-01 19:31:13 +00:00
|
|
|
if (!mutable_meta->AddGarbage(blob_file_garbage.GetGarbageBlobCount(),
|
|
|
|
blob_file_garbage.GetGarbageBlobBytes())) {
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Garbage overflow for blob file #" << blob_file_number;
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
int GetCurrentLevelForTableFile(uint64_t file_number) const {
|
|
|
|
auto it = table_file_levels_.find(file_number);
|
|
|
|
if (it != table_file_levels_.end()) {
|
|
|
|
return it->second;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(base_vstorage_);
|
|
|
|
return base_vstorage_->GetFileLocation(file_number).GetLevel();
|
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
uint64_t GetOldestBlobFileNumberForTableFile(int level,
|
|
|
|
uint64_t file_number) const {
|
|
|
|
assert(level < num_levels_);
|
|
|
|
|
|
|
|
const auto& added_files = levels_[level].added_files;
|
|
|
|
|
|
|
|
auto it = added_files.find(file_number);
|
|
|
|
if (it != added_files.end()) {
|
|
|
|
const FileMetaData* const meta = it->second;
|
|
|
|
assert(meta);
|
|
|
|
|
|
|
|
return meta->oldest_blob_file_number;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(base_vstorage_);
|
|
|
|
const FileMetaData* const meta =
|
|
|
|
base_vstorage_->GetFileMetaDataByNumber(file_number);
|
|
|
|
assert(meta);
|
|
|
|
|
|
|
|
return meta->oldest_blob_file_number;
|
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
Status ApplyFileDeletion(int level, uint64_t file_number) {
|
|
|
|
assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
|
|
|
|
|
|
|
|
const int current_level = GetCurrentLevelForTableFile(file_number);
|
|
|
|
|
|
|
|
if (level != current_level) {
|
|
|
|
if (level >= num_levels_) {
|
|
|
|
has_invalid_levels_ = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Cannot delete table file #" << file_number << " from level "
|
|
|
|
<< level << " since it is ";
|
|
|
|
if (current_level ==
|
|
|
|
VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
|
|
|
|
oss << "not in the LSM tree";
|
|
|
|
} else {
|
|
|
|
oss << "on level " << current_level;
|
|
|
|
}
|
|
|
|
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
|
|
|
|
if (level >= num_levels_) {
|
|
|
|
assert(invalid_level_sizes_[level] > 0);
|
|
|
|
--invalid_level_sizes_[level];
|
|
|
|
|
|
|
|
table_file_levels_[file_number] =
|
|
|
|
VersionStorageInfo::FileLocation::Invalid().GetLevel();
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
const uint64_t blob_file_number =
|
|
|
|
GetOldestBlobFileNumberForTableFile(level, file_number);
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
if (blob_file_number != kInvalidBlobFileNumber) {
|
|
|
|
MutableBlobFileMetaData* const mutable_meta =
|
|
|
|
GetOrCreateMutableBlobFileMetaData(blob_file_number);
|
|
|
|
if (mutable_meta) {
|
|
|
|
mutable_meta->UnlinkSst(file_number);
|
|
|
|
}
|
2020-06-12 16:52:14 +00:00
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
auto& level_state = levels_[level];
|
|
|
|
|
|
|
|
auto& add_files = level_state.added_files;
|
|
|
|
auto add_it = add_files.find(file_number);
|
|
|
|
if (add_it != add_files.end()) {
|
|
|
|
UnrefFile(add_it->second);
|
|
|
|
add_files.erase(add_it);
|
|
|
|
}
|
|
|
|
|
2020-06-12 01:29:51 +00:00
|
|
|
auto& del_files = level_state.deleted_files;
|
|
|
|
assert(del_files.find(file_number) == del_files.end());
|
|
|
|
del_files.emplace(file_number);
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
table_file_levels_[file_number] =
|
|
|
|
VersionStorageInfo::FileLocation::Invalid().GetLevel();
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
assert(version_edit_handler_);
|
2024-08-17 00:18:54 +00:00
|
|
|
if (l0_missing_files_.find(file_number) != l0_missing_files_.end()) {
|
|
|
|
l0_missing_files_.erase(file_number);
|
|
|
|
} else if (non_l0_missing_files_.find(file_number) !=
|
|
|
|
non_l0_missing_files_.end()) {
|
|
|
|
non_l0_missing_files_.erase(file_number);
|
2024-08-13 04:09:37 +00:00
|
|
|
} else {
|
2024-08-17 00:18:54 +00:00
|
|
|
auto fiter = found_files_.find(file_number);
|
2024-08-13 04:09:37 +00:00
|
|
|
// Only mark new files added during this catchup attempt for deletion.
|
|
|
|
// These files were never installed in VersionStorageInfo.
|
|
|
|
// Already referenced files that are deleted by a VersionEdit will
|
|
|
|
// be added to the VersionStorageInfo's obsolete files when the old
|
|
|
|
// version is dereferenced.
|
|
|
|
if (fiter != found_files_.end()) {
|
|
|
|
assert(!ioptions_->cf_paths.empty());
|
|
|
|
intermediate_files_.emplace_back(
|
|
|
|
MakeTableFileName(ioptions_->cf_paths[0].path, file_number));
|
|
|
|
found_files_.erase(fiter);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status ApplyFileAddition(int level, const FileMetaData& meta) {
|
|
|
|
assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
|
|
|
|
|
|
|
|
const uint64_t file_number = meta.fd.GetNumber();
|
|
|
|
|
|
|
|
const int current_level = GetCurrentLevelForTableFile(file_number);
|
|
|
|
|
|
|
|
if (current_level !=
|
|
|
|
VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
|
|
|
|
if (level >= num_levels_) {
|
|
|
|
has_invalid_levels_ = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Cannot add table file #" << file_number << " to level " << level
|
|
|
|
<< " since it is already in the LSM tree on level " << current_level;
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
|
|
|
|
if (level >= num_levels_) {
|
|
|
|
++invalid_level_sizes_[level];
|
|
|
|
table_file_levels_[file_number] = level;
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
auto& level_state = levels_[level];
|
|
|
|
|
|
|
|
auto& del_files = level_state.deleted_files;
|
|
|
|
auto del_it = del_files.find(file_number);
|
|
|
|
if (del_it != del_files.end()) {
|
|
|
|
del_files.erase(del_it);
|
|
|
|
}
|
|
|
|
|
2020-06-12 01:29:51 +00:00
|
|
|
FileMetaData* const f = new FileMetaData(meta);
|
|
|
|
f->refs = 1;
|
|
|
|
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
if (file_metadata_cache_res_mgr_) {
|
|
|
|
Status s = file_metadata_cache_res_mgr_->UpdateCacheReservation(
|
|
|
|
f->ApproximateMemoryUsage(), true /* increase */);
|
|
|
|
if (!s.ok()) {
|
|
|
|
delete f;
|
|
|
|
s = Status::MemoryLimit(
|
|
|
|
"Can't allocate " +
|
|
|
|
kCacheEntryRoleToCamelString[static_cast<std::uint32_t>(
|
|
|
|
CacheEntryRole::kFileMetadata)] +
|
|
|
|
" due to exceeding the memory limit "
|
|
|
|
"based on "
|
|
|
|
"cache capacity");
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-06-12 01:29:51 +00:00
|
|
|
auto& add_files = level_state.added_files;
|
|
|
|
assert(add_files.find(file_number) == add_files.end());
|
|
|
|
add_files.emplace(file_number, f);
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
const uint64_t blob_file_number = f->oldest_blob_file_number;
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
if (blob_file_number != kInvalidBlobFileNumber) {
|
|
|
|
MutableBlobFileMetaData* const mutable_meta =
|
|
|
|
GetOrCreateMutableBlobFileMetaData(blob_file_number);
|
|
|
|
if (mutable_meta) {
|
|
|
|
mutable_meta->LinkSst(file_number);
|
|
|
|
}
|
2020-06-12 16:52:14 +00:00
|
|
|
}
|
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
table_file_levels_[file_number] = level;
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
Status s;
|
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
assert(version_edit_handler_);
|
|
|
|
assert(!ioptions_->cf_paths.empty());
|
|
|
|
const std::string fpath =
|
|
|
|
MakeTableFileName(ioptions_->cf_paths[0].path, file_number);
|
|
|
|
s = version_edit_handler_->VerifyFile(cfd_, fpath, level, meta);
|
|
|
|
if (s.IsPathNotFound() || s.IsNotFound() || s.IsCorruption()) {
|
2024-08-17 00:18:54 +00:00
|
|
|
if (0 == level) {
|
|
|
|
l0_missing_files_.insert(file_number);
|
|
|
|
} else {
|
|
|
|
non_l0_missing_files_.insert(file_number);
|
|
|
|
}
|
2024-08-13 04:09:37 +00:00
|
|
|
if (s.IsCorruption()) {
|
|
|
|
found_files_.insert(file_number);
|
|
|
|
}
|
|
|
|
s = Status::OK();
|
|
|
|
} else if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
} else {
|
|
|
|
found_files_.insert(file_number);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
2020-06-03 18:21:16 +00:00
|
|
|
}
|
|
|
|
|
2022-06-21 18:56:53 +00:00
|
|
|
Status ApplyCompactCursors(int level,
|
|
|
|
const InternalKey& smallest_uncompacted_key) {
|
|
|
|
if (level < 0) {
|
|
|
|
std::ostringstream oss;
|
|
|
|
oss << "Cannot add compact cursor (" << level << ","
|
|
|
|
<< smallest_uncompacted_key.Encode().ToString()
|
|
|
|
<< " due to invalid level (level = " << level << ")";
|
|
|
|
return Status::Corruption("VersionBuilder", oss.str());
|
|
|
|
}
|
|
|
|
if (level < num_levels_) {
|
|
|
|
// Omit levels (>= num_levels_) when re-open with shrinking num_levels_
|
|
|
|
updated_compact_cursors_[level] = smallest_uncompacted_key;
|
|
|
|
}
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2014-10-31 18:54:05 +00:00
|
|
|
// Apply all of the edits in *edit to the current state.
|
2021-11-03 18:51:22 +00:00
|
|
|
Status Apply(const VersionEdit* edit) {
|
2024-08-17 00:18:54 +00:00
|
|
|
bool version_updated = false;
|
2020-06-03 18:21:16 +00:00
|
|
|
{
|
|
|
|
const Status s = CheckConsistency(base_vstorage_);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
2019-08-29 21:06:07 +00:00
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
// Note: we process the blob file related changes first because the
|
|
|
|
// table file addition/deletion logic depends on the blob files
|
|
|
|
// already being there.
|
|
|
|
|
|
|
|
// Add new blob files
|
|
|
|
for (const auto& blob_file_addition : edit->GetBlobFileAdditions()) {
|
|
|
|
const Status s = ApplyBlobFileAddition(blob_file_addition);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
2024-08-17 00:18:54 +00:00
|
|
|
version_updated = true;
|
2020-06-12 16:52:14 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Increase the amount of garbage for blob files affected by GC
|
|
|
|
for (const auto& blob_file_garbage : edit->GetBlobFileGarbages()) {
|
|
|
|
const Status s = ApplyBlobFileGarbage(blob_file_garbage);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
2024-08-17 00:18:54 +00:00
|
|
|
version_updated = true;
|
2020-06-12 16:52:14 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Delete table files
|
2020-06-03 18:21:16 +00:00
|
|
|
for (const auto& deleted_file : edit->GetDeletedFiles()) {
|
|
|
|
const int level = deleted_file.first;
|
|
|
|
const uint64_t file_number = deleted_file.second;
|
2017-08-24 23:05:16 +00:00
|
|
|
|
2020-06-03 18:21:16 +00:00
|
|
|
const Status s = ApplyFileDeletion(level, file_number);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
2015-01-07 18:29:21 +00:00
|
|
|
}
|
2024-08-17 00:18:54 +00:00
|
|
|
version_updated = true;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
|
2020-06-12 16:52:14 +00:00
|
|
|
// Add new table files
|
2014-10-31 18:54:05 +00:00
|
|
|
for (const auto& new_file : edit->GetNewFiles()) {
|
|
|
|
const int level = new_file.first;
|
2020-06-03 18:21:16 +00:00
|
|
|
const FileMetaData& meta = new_file.second;
|
|
|
|
|
|
|
|
const Status s = ApplyFileAddition(level, meta);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
2017-08-24 23:05:16 +00:00
|
|
|
}
|
2024-08-17 00:18:54 +00:00
|
|
|
version_updated = true;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2022-06-21 18:56:53 +00:00
|
|
|
// Populate compact cursors for round-robin compaction, leave
|
|
|
|
// the cursor to be empty to indicate it is invalid
|
|
|
|
for (const auto& cursor : edit->GetCompactCursors()) {
|
|
|
|
const int level = cursor.first;
|
|
|
|
const InternalKey smallest_uncompacted_key = cursor.second;
|
|
|
|
const Status s = ApplyCompactCursors(level, smallest_uncompacted_key);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
2024-08-17 00:18:54 +00:00
|
|
|
|
|
|
|
if (track_found_and_missing_files_ && version_updated) {
|
|
|
|
version_updated_since_last_check_ = true;
|
|
|
|
if (!edited_in_atomic_group_ && edit->IsInAtomicGroup()) {
|
|
|
|
edited_in_atomic_group_ = true;
|
|
|
|
}
|
|
|
|
}
|
2020-06-12 16:52:14 +00:00
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
// Helper function template for merging the blob file metadata from the base
|
|
|
|
// version with the mutable metadata representing the state after applying the
|
|
|
|
// edits. The function objects process_base and process_mutable are
|
|
|
|
// respectively called to handle a base version object when there is no
|
|
|
|
// matching mutable object, and a mutable object when there is no matching
|
|
|
|
// base version object. process_both is called to perform the merge when a
|
|
|
|
// given blob file appears both in the base version and the mutable list. The
|
|
|
|
// helper stops processing objects if a function object returns false. Blob
|
|
|
|
// files with a file number below first_blob_file are not processed.
|
|
|
|
template <typename ProcessBase, typename ProcessMutable, typename ProcessBoth>
|
|
|
|
void MergeBlobFileMetas(uint64_t first_blob_file, ProcessBase process_base,
|
|
|
|
ProcessMutable process_mutable,
|
|
|
|
ProcessBoth process_both) const {
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
assert(base_vstorage_);
|
Clean up blob files based on the linked SST set (#7001)
Summary:
The earlier `VersionBuilder` code only cleaned up blob files that were
marked as entirely consisting of garbage using `VersionEdits` with
`BlobFileGarbage`. This covers the cases when table files go through
regular compaction, where we iterate through the KVs and thus have an
opportunity to calculate the amount of garbage (that is, most cases).
However, it does not help when table files are simply dropped (e.g. deletion
compactions or the `DeleteFile` API). To deal with such cases, the patch
adds logic that cleans up all blob files at the head of the list until the first
one with linked SSTs is found. (As an example, let's assume we have blob files
with numbers 1..10, and the first one with any linked SSTs is number 8.
This means that SSTs in the `Version` only rely on blob files with numbers >= 8,
and thus 1..7 are no longer needed.)
The code change itself is pretty small; however, changing the logic like this
necessitated changes to some tests that have been added recently (namely
to the ones that use blob files in isolation, i.e. without any table files referring
to them). Some of these cases were fixed by bypassing `VersionBuilder` altogether
in order to keep the tests simple (which actually makes them more proper unit tests
as well), while the `VersionBuilder` unit tests were fixed by adding dummy table
files to the test cases as needed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7001
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D22119474
Pulled By: ltamasi
fbshipit-source-id: c6547141355667d4291d9661d6518eb741e7b54a
2020-06-30 22:30:01 +00:00
|
|
|
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
auto base_it = base_vstorage_->GetBlobFileMetaDataLB(first_blob_file);
|
|
|
|
const auto base_it_end = base_vstorage_->GetBlobFiles().end();
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
auto mutable_it = mutable_blob_file_metas_.lower_bound(first_blob_file);
|
2021-10-30 00:45:35 +00:00
|
|
|
const auto mutable_it_end = mutable_blob_file_metas_.end();
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
while (base_it != base_it_end && mutable_it != mutable_it_end) {
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
const auto& base_meta = *base_it;
|
|
|
|
assert(base_meta);
|
|
|
|
|
|
|
|
const uint64_t base_blob_file_number = base_meta->GetBlobFileNumber();
|
2021-10-30 00:45:35 +00:00
|
|
|
const uint64_t mutable_blob_file_number = mutable_it->first;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
if (base_blob_file_number < mutable_blob_file_number) {
|
2021-11-05 23:49:46 +00:00
|
|
|
if (!process_base(base_meta)) {
|
|
|
|
return;
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
|
|
|
++base_it;
|
2021-10-30 00:45:35 +00:00
|
|
|
} else if (mutable_blob_file_number < base_blob_file_number) {
|
|
|
|
const auto& mutable_meta = mutable_it->second;
|
2020-09-09 17:23:52 +00:00
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
if (!process_mutable(mutable_meta)) {
|
|
|
|
return;
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
++mutable_it;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
} else {
|
2021-10-30 00:45:35 +00:00
|
|
|
assert(base_blob_file_number == mutable_blob_file_number);
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
const auto& mutable_meta = mutable_it->second;
|
2020-05-19 16:57:49 +00:00
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
if (!process_both(base_meta, mutable_meta)) {
|
|
|
|
return;
|
2021-10-30 00:45:35 +00:00
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
|
|
|
++base_it;
|
2021-10-30 00:45:35 +00:00
|
|
|
++mutable_it;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
while (base_it != base_it_end) {
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
const auto& base_meta = *base_it;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
if (!process_base(base_meta)) {
|
|
|
|
return;
|
|
|
|
}
|
Clean up blob files based on the linked SST set (#7001)
Summary:
The earlier `VersionBuilder` code only cleaned up blob files that were
marked as entirely consisting of garbage using `VersionEdits` with
`BlobFileGarbage`. This covers the cases when table files go through
regular compaction, where we iterate through the KVs and thus have an
opportunity to calculate the amount of garbage (that is, most cases).
However, it does not help when table files are simply dropped (e.g. deletion
compactions or the `DeleteFile` API). To deal with such cases, the patch
adds logic that cleans up all blob files at the head of the list until the first
one with linked SSTs is found. (As an example, let's assume we have blob files
with numbers 1..10, and the first one with any linked SSTs is number 8.
This means that SSTs in the `Version` only rely on blob files with numbers >= 8,
and thus 1..7 are no longer needed.)
The code change itself is pretty small; however, changing the logic like this
necessitated changes to some tests that have been added recently (namely
to the ones that use blob files in isolation, i.e. without any table files referring
to them). Some of these cases were fixed by bypassing `VersionBuilder` altogether
in order to keep the tests simple (which actually makes them more proper unit tests
as well), while the `VersionBuilder` unit tests were fixed by adding dummy table
files to the test cases as needed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7001
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D22119474
Pulled By: ltamasi
fbshipit-source-id: c6547141355667d4291d9661d6518eb741e7b54a
2020-06-30 22:30:01 +00:00
|
|
|
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
++base_it;
|
|
|
|
}
|
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
while (mutable_it != mutable_it_end) {
|
|
|
|
const auto& mutable_meta = mutable_it->second;
|
2020-05-19 16:57:49 +00:00
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
if (!process_mutable(mutable_meta)) {
|
|
|
|
return;
|
|
|
|
}
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
|
2021-10-30 00:45:35 +00:00
|
|
|
++mutable_it;
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
// Helper function template for finding the first blob file that has linked
|
|
|
|
// SSTs.
|
|
|
|
template <typename Meta>
|
|
|
|
static bool CheckLinkedSsts(const Meta& meta,
|
|
|
|
uint64_t* min_oldest_blob_file_num) {
|
|
|
|
assert(min_oldest_blob_file_num);
|
|
|
|
|
|
|
|
if (!meta.GetLinkedSsts().empty()) {
|
|
|
|
assert(*min_oldest_blob_file_num == kInvalidBlobFileNumber);
|
|
|
|
|
|
|
|
*min_oldest_blob_file_num = meta.GetBlobFileNumber();
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Find the oldest blob file that has linked SSTs.
|
|
|
|
uint64_t GetMinOldestBlobFileNumber() const {
|
|
|
|
uint64_t min_oldest_blob_file_num = kInvalidBlobFileNumber;
|
|
|
|
|
|
|
|
auto process_base =
|
|
|
|
[&min_oldest_blob_file_num](
|
|
|
|
const std::shared_ptr<BlobFileMetaData>& base_meta) {
|
|
|
|
assert(base_meta);
|
|
|
|
|
|
|
|
return CheckLinkedSsts(*base_meta, &min_oldest_blob_file_num);
|
|
|
|
};
|
|
|
|
|
|
|
|
auto process_mutable = [&min_oldest_blob_file_num](
|
|
|
|
const MutableBlobFileMetaData& mutable_meta) {
|
|
|
|
return CheckLinkedSsts(mutable_meta, &min_oldest_blob_file_num);
|
|
|
|
};
|
|
|
|
|
|
|
|
auto process_both = [&min_oldest_blob_file_num](
|
|
|
|
const std::shared_ptr<BlobFileMetaData>& base_meta,
|
|
|
|
const MutableBlobFileMetaData& mutable_meta) {
|
|
|
|
#ifndef NDEBUG
|
|
|
|
assert(base_meta);
|
|
|
|
assert(base_meta->GetSharedMeta() == mutable_meta.GetSharedMeta());
|
|
|
|
#else
|
|
|
|
(void)base_meta;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// Look at mutable_meta since it supersedes *base_meta
|
|
|
|
return CheckLinkedSsts(mutable_meta, &min_oldest_blob_file_num);
|
|
|
|
};
|
|
|
|
|
|
|
|
MergeBlobFileMetas(kInvalidBlobFileNumber, process_base, process_mutable,
|
|
|
|
process_both);
|
|
|
|
|
|
|
|
return min_oldest_blob_file_num;
|
|
|
|
}
|
|
|
|
|
|
|
|
static std::shared_ptr<BlobFileMetaData> CreateBlobFileMetaData(
|
|
|
|
const MutableBlobFileMetaData& mutable_meta) {
|
|
|
|
return BlobFileMetaData::Create(
|
|
|
|
mutable_meta.GetSharedMeta(), mutable_meta.GetLinkedSsts(),
|
|
|
|
mutable_meta.GetGarbageBlobCount(), mutable_meta.GetGarbageBlobBytes());
|
|
|
|
}
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
bool OnlyLinkedToMissingL0Files(
|
|
|
|
const std::unordered_set<uint64_t>& linked_ssts) const {
|
|
|
|
return std::all_of(
|
|
|
|
linked_ssts.begin(), linked_ssts.end(), [&](const uint64_t& element) {
|
|
|
|
return l0_missing_files_.find(element) != l0_missing_files_.end();
|
|
|
|
});
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
// Add the blob file specified by meta to *vstorage if it is determined to
|
|
|
|
// contain valid data (blobs).
|
|
|
|
template <typename Meta>
|
2024-08-17 00:18:54 +00:00
|
|
|
void AddBlobFileIfNeeded(VersionStorageInfo* vstorage, Meta&& meta,
|
|
|
|
uint64_t blob_file_number) const {
|
2021-11-05 23:49:46 +00:00
|
|
|
assert(vstorage);
|
|
|
|
assert(meta);
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
const auto& linked_ssts = meta->GetLinkedSsts();
|
|
|
|
if (track_found_and_missing_files_) {
|
|
|
|
if (missing_blob_files_.find(blob_file_number) !=
|
|
|
|
missing_blob_files_.end()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
// Leave the empty case for the below blob garbage collection logic.
|
|
|
|
if (!linked_ssts.empty() && OnlyLinkedToMissingL0Files(linked_ssts)) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (linked_ssts.empty() &&
|
2021-11-05 23:49:46 +00:00
|
|
|
meta->GetGarbageBlobCount() >= meta->GetTotalBlobCount()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
vstorage->AddBlobFile(std::forward<Meta>(meta));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Merge the blob file metadata from the base version with the changes (edits)
|
|
|
|
// applied, and save the result into *vstorage.
|
|
|
|
void SaveBlobFilesTo(VersionStorageInfo* vstorage) const {
|
|
|
|
assert(vstorage);
|
2024-08-17 00:18:54 +00:00
|
|
|
assert(!track_found_and_missing_files_ || valid_version_available_);
|
2021-11-05 23:49:46 +00:00
|
|
|
|
Use a sorted vector instead of a map to store blob file metadata (#9526)
Summary:
The patch replaces `std::map` with a sorted `std::vector` for
`VersionStorageInfo::blob_files_` and preallocates the space
for the `vector` before saving the `BlobFileMetaData` into the
new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`.
These changes reduce the time the DB mutex is held while
saving new `Version`s, and using a sorted `vector` also makes
lookups faster thanks to better memory locality.
In addition, the patch introduces helper methods
`VersionStorageInfo::GetBlobFileMetaData` and
`VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by
clients to perform lookups in the `vector`, and does some general
cleanup in the parts of code where blob file metadata are used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526
Test Plan:
Ran `make check` and the crash test script for a while.
Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced:
```
numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value>
```
Final statistics before the patch:
```
Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s
Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s
```
With the patch:
```
Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s
Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s
```
Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs.
Reviewed By: riversand963
Differential Revision: D34082728
Pulled By: ltamasi
fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 20:35:39 +00:00
|
|
|
assert(base_vstorage_);
|
|
|
|
vstorage->ReserveBlob(base_vstorage_->GetBlobFiles().size() +
|
|
|
|
mutable_blob_file_metas_.size());
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
const uint64_t oldest_blob_file_with_linked_ssts =
|
|
|
|
GetMinOldestBlobFileNumber();
|
|
|
|
|
2024-01-16 19:15:23 +00:00
|
|
|
// If there are no blob files with linked SSTs, meaning that there are no
|
|
|
|
// valid blob files
|
|
|
|
if (oldest_blob_file_with_linked_ssts == kInvalidBlobFileNumber) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
auto process_base =
|
2024-08-17 00:18:54 +00:00
|
|
|
[this, vstorage](const std::shared_ptr<BlobFileMetaData>& base_meta) {
|
2021-11-05 23:49:46 +00:00
|
|
|
assert(base_meta);
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
AddBlobFileIfNeeded(vstorage, base_meta,
|
|
|
|
base_meta->GetBlobFileNumber());
|
2021-11-05 23:49:46 +00:00
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
auto process_mutable =
|
2024-08-17 00:18:54 +00:00
|
|
|
[this, vstorage](const MutableBlobFileMetaData& mutable_meta) {
|
|
|
|
AddBlobFileIfNeeded(vstorage, CreateBlobFileMetaData(mutable_meta),
|
|
|
|
mutable_meta.GetBlobFileNumber());
|
2021-11-05 23:49:46 +00:00
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
auto process_both = [this, vstorage](
|
2021-11-05 23:49:46 +00:00
|
|
|
const std::shared_ptr<BlobFileMetaData>& base_meta,
|
|
|
|
const MutableBlobFileMetaData& mutable_meta) {
|
|
|
|
assert(base_meta);
|
|
|
|
assert(base_meta->GetSharedMeta() == mutable_meta.GetSharedMeta());
|
|
|
|
|
|
|
|
if (!mutable_meta.HasDelta()) {
|
|
|
|
assert(base_meta->GetGarbageBlobCount() ==
|
|
|
|
mutable_meta.GetGarbageBlobCount());
|
|
|
|
assert(base_meta->GetGarbageBlobBytes() ==
|
|
|
|
mutable_meta.GetGarbageBlobBytes());
|
|
|
|
assert(base_meta->GetLinkedSsts() == mutable_meta.GetLinkedSsts());
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
AddBlobFileIfNeeded(vstorage, base_meta,
|
|
|
|
base_meta->GetBlobFileNumber());
|
2021-11-05 23:49:46 +00:00
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
AddBlobFileIfNeeded(vstorage, CreateBlobFileMetaData(mutable_meta),
|
|
|
|
mutable_meta.GetBlobFileNumber());
|
2021-11-05 23:49:46 +00:00
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
MergeBlobFileMetas(oldest_blob_file_with_linked_ssts, process_base,
|
|
|
|
process_mutable, process_both);
|
|
|
|
}
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
void MaybeAddFile(VersionStorageInfo* vstorage, int level,
|
|
|
|
FileMetaData* f) const {
|
2021-08-27 19:10:01 +00:00
|
|
|
const uint64_t file_number = f->fd.GetNumber();
|
2024-08-17 00:18:54 +00:00
|
|
|
if (track_found_and_missing_files_ && level == 0 &&
|
|
|
|
l0_missing_files_.find(file_number) != l0_missing_files_.end()) {
|
|
|
|
return;
|
|
|
|
}
|
2019-08-29 21:06:07 +00:00
|
|
|
|
2021-08-27 19:10:01 +00:00
|
|
|
const auto& level_state = levels_[level];
|
|
|
|
|
|
|
|
const auto& del_files = level_state.deleted_files;
|
|
|
|
const auto del_it = del_files.find(file_number);
|
|
|
|
|
|
|
|
if (del_it != del_files.end()) {
|
|
|
|
// f is to-be-deleted table file
|
|
|
|
vstorage->RemoveCurrentStats(f);
|
|
|
|
} else {
|
|
|
|
const auto& add_files = level_state.added_files;
|
|
|
|
const auto add_it = add_files.find(file_number);
|
|
|
|
|
|
|
|
// Note: if the file appears both in the base version and in the added
|
|
|
|
// list, the added FileMetaData supersedes the one in the base version.
|
|
|
|
if (add_it != add_files.end() && add_it->second != f) {
|
|
|
|
vstorage->RemoveCurrentStats(f);
|
|
|
|
} else {
|
|
|
|
vstorage->AddFile(level, f);
|
|
|
|
}
|
2019-08-29 21:06:07 +00:00
|
|
|
}
|
2021-08-27 19:10:01 +00:00
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
bool ContainsCompleteVersion() const {
|
2024-08-13 04:09:37 +00:00
|
|
|
assert(track_found_and_missing_files_);
|
2024-08-17 00:18:54 +00:00
|
|
|
return l0_missing_files_.empty() && non_l0_missing_files_.empty() &&
|
2024-08-13 04:09:37 +00:00
|
|
|
(missing_blob_files_high_ == kInvalidBlobFileNumber ||
|
|
|
|
missing_blob_files_high_ < GetMinOldestBlobFileNumber());
|
|
|
|
}
|
|
|
|
|
|
|
|
bool HasMissingFiles() const {
|
|
|
|
assert(track_found_and_missing_files_);
|
2024-08-17 00:18:54 +00:00
|
|
|
return !l0_missing_files_.empty() || !non_l0_missing_files_.empty() ||
|
2024-08-13 04:09:37 +00:00
|
|
|
missing_blob_files_high_ != kInvalidBlobFileNumber;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<std::string>& GetAndClearIntermediateFiles() {
|
|
|
|
assert(track_found_and_missing_files_);
|
|
|
|
return intermediate_files_;
|
|
|
|
}
|
|
|
|
|
|
|
|
void ClearFoundFiles() {
|
|
|
|
assert(track_found_and_missing_files_);
|
|
|
|
found_files_.clear();
|
|
|
|
}
|
|
|
|
|
2021-11-05 23:49:46 +00:00
|
|
|
template <typename Cmp>
|
2021-11-03 18:51:22 +00:00
|
|
|
void SaveSSTFilesTo(VersionStorageInfo* vstorage, int level, Cmp cmp) const {
|
|
|
|
// Merge the set of added files with the set of pre-existing files.
|
|
|
|
// Drop any deleted files. Store the result in *vstorage.
|
|
|
|
const auto& base_files = base_vstorage_->LevelFiles(level);
|
|
|
|
const auto& unordered_added_files = levels_[level].added_files;
|
|
|
|
vstorage->Reserve(level, base_files.size() + unordered_added_files.size());
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
MergeUnorderdAddedFilesWithBase(
|
|
|
|
base_files, unordered_added_files, cmp,
|
|
|
|
[&](FileMetaData* file) { MaybeAddFile(vstorage, level, file); });
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename Cmp, typename AddFileFunc>
|
|
|
|
void MergeUnorderdAddedFilesWithBase(
|
|
|
|
const std::vector<FileMetaData*>& base_files,
|
|
|
|
const std::unordered_map<uint64_t, FileMetaData*>& unordered_added_files,
|
|
|
|
Cmp cmp, AddFileFunc add_file_func) const {
|
2021-11-03 18:51:22 +00:00
|
|
|
// Sort added files for the level.
|
|
|
|
std::vector<FileMetaData*> added_files;
|
|
|
|
added_files.reserve(unordered_added_files.size());
|
|
|
|
for (const auto& pair : unordered_added_files) {
|
|
|
|
added_files.push_back(pair.second);
|
|
|
|
}
|
|
|
|
std::sort(added_files.begin(), added_files.end(), cmp);
|
|
|
|
|
|
|
|
auto base_iter = base_files.begin();
|
|
|
|
auto base_end = base_files.end();
|
|
|
|
auto added_iter = added_files.begin();
|
|
|
|
auto added_end = added_files.end();
|
|
|
|
while (added_iter != added_end || base_iter != base_end) {
|
|
|
|
if (base_iter == base_end ||
|
|
|
|
(added_iter != added_end && cmp(*added_iter, *base_iter))) {
|
2024-08-17 00:18:54 +00:00
|
|
|
add_file_func(*added_iter++);
|
2021-11-03 18:51:22 +00:00
|
|
|
} else {
|
2024-08-17 00:18:54 +00:00
|
|
|
add_file_func(*base_iter++);
|
2014-12-10 04:33:43 +00:00
|
|
|
}
|
2021-11-03 18:51:22 +00:00
|
|
|
}
|
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
bool PromoteEpochNumberRequirementIfNeeded(
|
|
|
|
VersionStorageInfo* vstorage) const {
|
|
|
|
if (vstorage->HasMissingEpochNumber()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int level = 0; level < num_levels_; ++level) {
|
|
|
|
for (const auto& pair : levels_[level].added_files) {
|
|
|
|
const FileMetaData* f = pair.second;
|
|
|
|
if (f->epoch_number == kUnknownEpochNumber) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
vstorage->SetEpochNumberRequirement(EpochNumberRequirement::kMustPresent);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
void SaveSSTFilesTo(VersionStorageInfo* vstorage) const {
|
|
|
|
assert(vstorage);
|
2014-12-11 23:46:01 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
if (!num_levels_) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
EpochNumberRequirement epoch_number_requirement =
|
|
|
|
vstorage->GetEpochNumberRequirement();
|
|
|
|
|
|
|
|
if (epoch_number_requirement == EpochNumberRequirement::kMightMissing) {
|
|
|
|
bool promoted = PromoteEpochNumberRequirementIfNeeded(vstorage);
|
|
|
|
if (promoted) {
|
|
|
|
epoch_number_requirement = vstorage->GetEpochNumberRequirement();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (epoch_number_requirement == EpochNumberRequirement::kMightMissing) {
|
2024-08-13 04:09:37 +00:00
|
|
|
SaveSSTFilesTo(vstorage, /* level */ 0, *level_zero_cmp_by_seqno_);
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
} else {
|
2024-08-13 04:09:37 +00:00
|
|
|
SaveSSTFilesTo(vstorage, /* level */ 0, *level_zero_cmp_by_epochno_);
|
Sort L0 files by newly introduced epoch_num (#10922)
Summary:
**Context:**
Sorting L0 files by `largest_seqno` has at least two inconvenience:
- File ingestion and compaction involving ingested files can create files of overlapping seqno range with the existing files. `force_consistency_check=true` will catch such overlap seqno range even those harmless overlap.
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n")
- insert k1@1 to memtable m1
- ingest file s1 with k2@2, ingest file s2 with k3@3
- insert k4@4 to m1
- compact files s1, s2 and result in new file s3 of seqno range [2, 3]
- flush m1 and result in new file s4 of seqno range [1, 4]. And `force_consistency_check=true` will think s4 and s3 has file reordering corruption that might cause retuning an old value of k1
- However such caught corruption is a false positive since s1, s2 will not have overlapped keys with k1 or whatever inserted into m1 before ingest file s1 by the requirement of file ingestion (otherwise the m1 will be flushed first before any of the file ingestion completes). Therefore there in fact isn't any file reordering corruption.
- Single delete can decrease a file's largest seqno and ordering by `largest_seqno` can introduce a wrong ordering hence file reordering corruption
- For example, consider the following sequence of events ("key@n" indicates key at seqno "n", Credit to ajkr for this example)
- an existing SST s1 contains only k1@1
- insert k1@2 to memtable m1
- ingest file s2 with k3@3, ingest file s3 with k4@4
- insert single delete k5@5 in m1
- flush m1 and result in new file s4 of seqno range [2, 5]
- compact s1, s2, s3 and result in new file s5 of seqno range [1, 4]
- compact s4 and result in new file s6 of seqno range [2] due to single delete
- By the last step, we have file ordering by largest seqno (">" means "newer") : s5 > s6 while s6 contains a newer version of the k1's value (i.e, k1@2) than s5, which is a real reordering corruption. While this can be caught by `force_consistency_check=true`, there isn't a good way to prevent this from happening if ordering by `largest_seqno`
Therefore, we are redesigning the sorting criteria of L0 files and avoid above inconvenience. Credit to ajkr , we now introduce `epoch_num` which describes the order of a file being flushed or ingested/imported (compaction output file will has the minimum `epoch_num` among input files'). This will avoid the above inconvenience in the following ways:
- In the first case above, there will no longer be overlap seqno range check in `force_consistency_check=true` but `epoch_number` ordering check. This will result in file ordering s1 < s2 < s4 (pre-compaction) and s3 < s4 (post-compaction) which won't trigger false positive corruption. See test class `DBCompactionTestL0FilesMisorderCorruption*` for more.
- In the second case above, this will result in file ordering s1 < s2 < s3 < s4 (pre-compacting s1, s2, s3), s5 < s4 (post-compacting s1, s2, s3), s5 < s6 (post-compacting s4), which are correct file ordering without causing any corruption.
**Summary:**
- Introduce `epoch_number` stored per `ColumnFamilyData` and sort CF's L0 files by their assigned `epoch_number` instead of `largest_seqno`.
- `epoch_number` is increased and assigned upon `VersionEdit::AddFile()` for flush (or similarly for WriteLevel0TableForRecovery) and file ingestion (except for allow_behind_true, which will always get assigned as the `kReservedEpochNumberForFileIngestedBehind`)
- Compaction output file is assigned with the minimum `epoch_number` among input files'
- Refit level: reuse refitted file's epoch_number
- Other paths needing `epoch_number` treatment:
- Import column families: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`
- Repair: reuse file's epoch_number if exists. If not, assign one based on `NewestFirstBySeqNo`.
- Assigning new epoch_number to a file and adding this file to LSM tree should be atomic. This is guaranteed by us assigning epoch_number right upon `VersionEdit::AddFile()` where this version edit will be apply to LSM tree shape right after by holding the db mutex (e.g, flush, file ingestion, import column family) or by there is only 1 ongoing edit per CF (e.g, WriteLevel0TableForRecovery, Repair).
- Assigning the minimum input epoch number to compaction output file won't misorder L0 files (even through later `Refit(target_level=0)`). It's due to for every key "k" in the input range, a legit compaction will cover a continuous epoch number range of that key. As long as we assign the key "k" the minimum input epoch number, it won't become newer or older than the versions of this key that aren't included in this compaction hence no misorder.
- Persist `epoch_number` of each file in manifest and recover `epoch_number` on db recovery
- Backward compatibility with old db without `epoch_number` support is guaranteed by assigning `epoch_number` to recovered files by `NewestFirstBySeqno` order. See `VersionStorageInfo::RecoverEpochNumbers()` for more
- Forward compatibility with manifest is guaranteed by flexibility of `NewFileCustomTag`
- Replace `force_consistent_check` on L0 with `epoch_number` and remove false positive check like case 1 with `largest_seqno` above
- Due to backward compatibility issue, we might encounter files with missing epoch number at the beginning of db recovery. We will still use old L0 sorting mechanism (`NewestFirstBySeqno`) to check/sort them till we infer their epoch number. See usages of `EpochNumberRequirement`.
- Remove fix https://github.com/facebook/rocksdb/pull/5958#issue-511150930 and their outdated tests to file reordering corruption because such fix can be replaced by this PR.
- Misc:
- update existing tests with `epoch_number` so make check will pass
- update https://github.com/facebook/rocksdb/pull/5958#issue-511150930 tests to verify corruption is fixed using `epoch_number` and cover universal/fifo compaction/CompactRange/CompactFile cases
- assert db_mutex is held for a few places before calling ColumnFamilyData::NewEpochNumber()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10922
Test Plan:
- `make check`
- New unit tests under `db/db_compaction_test.cc`, `db/db_test2.cc`, `db/version_builder_test.cc`, `db/repair_test.cc`
- Updated tests (i.e, `DBCompactionTestL0FilesMisorderCorruption*`) under https://github.com/facebook/rocksdb/pull/5958#issue-511150930
- [Ongoing] Compatibility test: manually run https://github.com/ajkr/rocksdb/commit/36a5686ec012f35a4371e409aa85c404ca1c210d (with file ingestion off for running the `.orig` binary to prevent this bug affecting upgrade/downgrade formality checking) for 1 hour on `simple black/white box`, `cf_consistency/txn/enable_ts with whitebox + test_best_efforts_recovery with blackbox`
- [Ongoing] normal db stress test
- [Ongoing] db stress test with aggressive value https://github.com/facebook/rocksdb/pull/10761
Reviewed By: ajkr
Differential Revision: D41063187
Pulled By: hx235
fbshipit-source-id: 826cb23455de7beaabe2d16c57682a82733a32a9
2022-12-13 21:29:37 +00:00
|
|
|
}
|
2021-11-03 18:51:22 +00:00
|
|
|
|
|
|
|
for (int level = 1; level < num_levels_; ++level) {
|
2024-08-13 04:09:37 +00:00
|
|
|
SaveSSTFilesTo(vstorage, level, *level_nonzero_cmp_);
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
2021-08-27 19:10:01 +00:00
|
|
|
}
|
|
|
|
|
2022-06-21 18:56:53 +00:00
|
|
|
void SaveCompactCursorsTo(VersionStorageInfo* vstorage) const {
|
|
|
|
for (auto iter = updated_compact_cursors_.begin();
|
|
|
|
iter != updated_compact_cursors_.end(); iter++) {
|
|
|
|
vstorage->AddCursorForOneLevel(iter->first, iter->second);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
bool ValidVersionAvailable() {
|
|
|
|
assert(track_found_and_missing_files_);
|
|
|
|
if (version_updated_since_last_check_) {
|
|
|
|
valid_version_available_ = ContainsCompleteVersion();
|
|
|
|
if (!valid_version_available_ && !edited_in_atomic_group_ &&
|
|
|
|
allow_incomplete_valid_version_) {
|
|
|
|
valid_version_available_ = OnlyMissingL0Suffix();
|
|
|
|
}
|
|
|
|
version_updated_since_last_check_ = false;
|
|
|
|
}
|
|
|
|
return valid_version_available_;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool OnlyMissingL0Suffix() const {
|
|
|
|
if (!non_l0_missing_files_.empty()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
assert(!(l0_missing_files_.empty() && missing_blob_files_.empty()));
|
|
|
|
|
|
|
|
if (!l0_missing_files_.empty() && !MissingL0FilesAreL0Suffix()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (!missing_blob_files_.empty() &&
|
|
|
|
!RemainingSstFilesNotMissingBlobFiles()) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check missing L0 files are a suffix of expected sorted L0 files.
|
|
|
|
bool MissingL0FilesAreL0Suffix() const {
|
|
|
|
assert(non_l0_missing_files_.empty());
|
|
|
|
assert(!l0_missing_files_.empty());
|
|
|
|
std::vector<FileMetaData*> expected_sorted_l0_files;
|
|
|
|
const auto& base_files = base_vstorage_->LevelFiles(0);
|
|
|
|
const auto& unordered_added_files = levels_[0].added_files;
|
|
|
|
expected_sorted_l0_files.reserve(base_files.size() +
|
|
|
|
unordered_added_files.size());
|
|
|
|
EpochNumberRequirement epoch_number_requirement =
|
|
|
|
base_vstorage_->GetEpochNumberRequirement();
|
|
|
|
|
|
|
|
if (epoch_number_requirement == EpochNumberRequirement::kMightMissing) {
|
|
|
|
MergeUnorderdAddedFilesWithBase(
|
|
|
|
base_files, unordered_added_files, *level_zero_cmp_by_seqno_,
|
|
|
|
[&](FileMetaData* file) {
|
|
|
|
expected_sorted_l0_files.push_back(file);
|
|
|
|
});
|
|
|
|
} else {
|
|
|
|
MergeUnorderdAddedFilesWithBase(
|
|
|
|
base_files, unordered_added_files, *level_zero_cmp_by_epochno_,
|
|
|
|
[&](FileMetaData* file) {
|
|
|
|
expected_sorted_l0_files.push_back(file);
|
|
|
|
});
|
|
|
|
}
|
|
|
|
assert(expected_sorted_l0_files.size() >= l0_missing_files_.size());
|
|
|
|
std::unordered_set<uint64_t> unaddressed_missing_files = l0_missing_files_;
|
|
|
|
for (auto iter = expected_sorted_l0_files.begin();
|
|
|
|
iter != expected_sorted_l0_files.end(); iter++) {
|
|
|
|
uint64_t file_number = (*iter)->fd.GetNumber();
|
|
|
|
if (l0_missing_files_.find(file_number) != l0_missing_files_.end()) {
|
|
|
|
assert(unaddressed_missing_files.find(file_number) !=
|
|
|
|
unaddressed_missing_files.end());
|
|
|
|
unaddressed_missing_files.erase(file_number);
|
|
|
|
} else if (!unaddressed_missing_files.empty()) {
|
|
|
|
return false;
|
|
|
|
} else {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check for each of the missing blob file missing, it either is older than
|
|
|
|
// the minimum oldest blob file required by this Version or only linked to
|
|
|
|
// the missing L0 files.
|
|
|
|
bool RemainingSstFilesNotMissingBlobFiles() const {
|
|
|
|
assert(non_l0_missing_files_.empty());
|
|
|
|
assert(!missing_blob_files_.empty());
|
|
|
|
bool no_l0_files_missing = l0_missing_files_.empty();
|
|
|
|
uint64_t min_oldest_blob_file_num = GetMinOldestBlobFileNumber();
|
|
|
|
for (const auto& missing_blob_file : missing_blob_files_) {
|
|
|
|
if (missing_blob_file < min_oldest_blob_file_num) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
auto iter = mutable_blob_file_metas_.find(missing_blob_file);
|
|
|
|
assert(iter != mutable_blob_file_metas_.end());
|
2024-08-21 17:09:10 +00:00
|
|
|
const std::unordered_set<uint64_t>& linked_ssts =
|
|
|
|
iter->second.GetLinkedSsts();
|
2024-08-17 00:18:54 +00:00
|
|
|
// TODO(yuzhangyu): In theory, if no L0 SST files ara missing, and only
|
|
|
|
// blob files exclusively linked to a L0 suffix are missing, we can
|
|
|
|
// recover to a valid point in time too. We don't recover that type of
|
|
|
|
// incomplete Version yet.
|
|
|
|
if (!linked_ssts.empty() && no_l0_files_missing) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (!OnlyLinkedToMissingL0Files(linked_ssts)) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2021-08-27 19:10:01 +00:00
|
|
|
// Save the current state in *vstorage.
|
2021-11-03 18:51:22 +00:00
|
|
|
Status SaveTo(VersionStorageInfo* vstorage) const {
|
2024-08-17 00:18:54 +00:00
|
|
|
assert(!track_found_and_missing_files_ || valid_version_available_);
|
2022-06-21 02:15:59 +00:00
|
|
|
Status s;
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
|
|
// The same check is done within Apply() so we skip it in release mode.
|
|
|
|
s = CheckConsistency(base_vstorage_);
|
2021-08-27 19:10:01 +00:00
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
2022-06-21 02:15:59 +00:00
|
|
|
#endif // NDEBUG
|
2021-08-27 19:10:01 +00:00
|
|
|
|
|
|
|
s = CheckConsistency(vstorage);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
SaveSSTFilesTo(vstorage);
|
2014-10-31 18:54:05 +00:00
|
|
|
|
Add blob files to VersionStorageInfo/VersionBuilder (#6597)
Summary:
The patch adds a couple of classes to represent metadata about
blob files: `SharedBlobFileMetaData` contains the information elements
that are immutable (once the blob file is closed), e.g. blob file number,
total number and size of blob files, checksum method/value, while
`BlobFileMetaData` contains attributes that can vary across versions like
the amount of garbage in the file. There is a single `SharedBlobFileMetaData`
for each blob file, which is jointly owned by the `BlobFileMetaData` objects
that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s
and can also be shared if the (immutable _and_ mutable) state of the blob file
is the same in two versions.
In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends
`VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those
containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata
to a new `VersionStorageInfo`. Consistency checks are also extended to ensure
that table files point to blob files that are part of the `Version`, and that all blob files
that are part of any given `Version` have at least some _non_-garbage data in them.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D20656803
Pulled By: ltamasi
fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 01:48:55 +00:00
|
|
|
SaveBlobFilesTo(vstorage);
|
|
|
|
|
2022-06-21 18:56:53 +00:00
|
|
|
SaveCompactCursorsTo(vstorage);
|
|
|
|
|
2019-08-29 21:06:07 +00:00
|
|
|
s = CheckConsistency(vstorage);
|
|
|
|
return s;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
|
2024-10-17 21:13:20 +00:00
|
|
|
Status LoadTableHandlers(InternalStats* internal_stats, int max_threads,
|
|
|
|
bool prefetch_index_and_filter_in_cache,
|
|
|
|
bool is_initial_load,
|
|
|
|
const MutableCFOptions& mutable_cf_options,
|
|
|
|
size_t max_file_size_for_l0_meta_pin,
|
|
|
|
const ReadOptions& read_options) {
|
2014-10-31 18:54:05 +00:00
|
|
|
assert(table_cache_ != nullptr);
|
2024-08-17 00:18:54 +00:00
|
|
|
assert(!track_found_and_missing_files_ || valid_version_available_);
|
2018-12-29 02:00:00 +00:00
|
|
|
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
size_t table_cache_capacity =
|
|
|
|
table_cache_->get_cache().get()->GetCapacity();
|
2018-12-29 02:00:00 +00:00
|
|
|
bool always_load = (table_cache_capacity == TableCache::kInfiniteCapacity);
|
2022-05-05 20:08:21 +00:00
|
|
|
size_t max_load = std::numeric_limits<size_t>::max();
|
2018-12-29 02:00:00 +00:00
|
|
|
|
|
|
|
if (!always_load) {
|
2020-03-12 01:36:43 +00:00
|
|
|
// If it is initial loading and not set to always loading all the
|
2018-12-29 02:00:00 +00:00
|
|
|
// files, we only load up to kInitialLoadLimit files, to limit the
|
|
|
|
// time reopening the DB.
|
|
|
|
const size_t kInitialLoadLimit = 16;
|
|
|
|
size_t load_limit;
|
|
|
|
// If the table cache is not 1/4 full, we pin the table handle to
|
|
|
|
// file metadata to avoid the cache read costs when reading the file.
|
|
|
|
// The downside of pinning those files is that LRU won't be followed
|
|
|
|
// for those files. This doesn't matter much because if number of files
|
|
|
|
// of the DB excceeds table cache capacity, eventually no table reader
|
|
|
|
// will be pinned and LRU will be followed.
|
|
|
|
if (is_initial_load) {
|
|
|
|
load_limit = std::min(kInitialLoadLimit, table_cache_capacity / 4);
|
|
|
|
} else {
|
|
|
|
load_limit = table_cache_capacity / 4;
|
|
|
|
}
|
|
|
|
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
size_t table_cache_usage = table_cache_->get_cache().get()->GetUsage();
|
2018-12-29 02:00:00 +00:00
|
|
|
if (table_cache_usage >= load_limit) {
|
2019-03-26 23:41:31 +00:00
|
|
|
// TODO (yanqin) find a suitable status code.
|
|
|
|
return Status::OK();
|
2018-12-29 02:00:00 +00:00
|
|
|
} else {
|
|
|
|
max_load = load_limit - table_cache_usage;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-09-02 17:25:20 +00:00
|
|
|
// <file metadata, level>
|
|
|
|
std::vector<std::pair<FileMetaData*, int>> files_meta;
|
2019-03-26 23:41:31 +00:00
|
|
|
std::vector<Status> statuses;
|
2017-08-24 23:05:16 +00:00
|
|
|
for (int level = 0; level < num_levels_; level++) {
|
2014-12-10 04:33:43 +00:00
|
|
|
for (auto& file_meta_pair : levels_[level].added_files) {
|
|
|
|
auto* file_meta = file_meta_pair.second;
|
2024-08-17 00:18:54 +00:00
|
|
|
uint64_t file_number = file_meta->fd.GetNumber();
|
|
|
|
if (track_found_and_missing_files_ && level == 0 &&
|
|
|
|
l0_missing_files_.find(file_number) != l0_missing_files_.end()) {
|
|
|
|
continue;
|
|
|
|
}
|
2019-03-26 23:41:31 +00:00
|
|
|
// If the file has been opened before, just skip it.
|
|
|
|
if (!file_meta->table_reader_handle) {
|
|
|
|
files_meta.emplace_back(file_meta, level);
|
|
|
|
statuses.emplace_back(Status::OK());
|
|
|
|
}
|
2018-12-29 02:00:00 +00:00
|
|
|
if (files_meta.size() >= max_load) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (files_meta.size() >= max_load) {
|
|
|
|
break;
|
2015-08-11 19:19:56 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
std::atomic<size_t> next_file_meta_idx(0);
|
2018-08-21 00:31:28 +00:00
|
|
|
std::function<void()> load_handlers_func([&]() {
|
2015-08-11 19:19:56 +00:00
|
|
|
while (true) {
|
|
|
|
size_t file_idx = next_file_meta_idx.fetch_add(1);
|
|
|
|
if (file_idx >= files_meta.size()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2015-09-02 17:25:20 +00:00
|
|
|
auto* file_meta = files_meta[file_idx].first;
|
|
|
|
int level = files_meta[file_idx].second;
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
TableCache::TypedHandle* handle = nullptr;
|
2019-03-26 23:41:31 +00:00
|
|
|
statuses[file_idx] = table_cache_->FindTable(
|
2023-04-21 16:07:18 +00:00
|
|
|
read_options, file_options_,
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
*(base_vstorage_->InternalComparator()), *file_meta, &handle,
|
2024-10-17 21:13:20 +00:00
|
|
|
mutable_cf_options, false /*no_io */,
|
2018-05-21 21:33:55 +00:00
|
|
|
internal_stats->GetFileReadHist(level), false, level,
|
2021-10-07 21:57:02 +00:00
|
|
|
prefetch_index_and_filter_in_cache, max_file_size_for_l0_meta_pin,
|
2023-08-21 19:14:03 +00:00
|
|
|
file_meta->temperature);
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
if (handle != nullptr) {
|
|
|
|
file_meta->table_reader_handle = handle;
|
2014-10-31 18:54:05 +00:00
|
|
|
// Load table_reader
|
Major Cache refactoring, CPU efficiency improvement (#10975)
Summary:
This is several refactorings bundled into one to avoid having to incrementally re-modify uses of Cache several times. Overall, there are breaking changes to Cache class, and it becomes more of low-level interface for implementing caches, especially block cache. New internal APIs make using Cache cleaner than before, and more insulated from block cache evolution. Hopefully, this is the last really big block cache refactoring, because of rather effectively decoupling the implementations from the uses. This change also removes the EXPERIMENTAL designation on the SecondaryCache support in Cache. It seems reasonably mature at this point but still subject to change/evolution (as I warn in the API docs for Cache).
The high-level motivation for this refactoring is to minimize code duplication / compounding complexity in adding SecondaryCache support to HyperClockCache (in a later PR). Other benefits listed below.
* static_cast lines of code +29 -35 (net removed 6)
* reinterpret_cast lines of code +6 -32 (net removed 26)
## cache.h and secondary_cache.h
* Always use CacheItemHelper with entries instead of just a Deleter. There are several motivations / justifications:
* Simpler for implementations to deal with just one Insert and one Lookup.
* Simpler and more efficient implementation because we don't have to track which entries are using helpers and which are using deleters
* Gets rid of hack to classify cache entries by their deleter. Instead, the CacheItemHelper includes a CacheEntryRole. This simplifies a lot of code (cache_entry_roles.h almost eliminated). Fixes https://github.com/facebook/rocksdb/issues/9428.
* Makes it trivial to adjust SecondaryCache behavior based on kind of block (e.g. don't re-compress filter blocks).
* It is arguably less convenient for many direct users of Cache, but direct users of Cache are now rare with introduction of typed_cache.h (below).
* I considered and rejected an alternative approach in which we reduce customizability by assuming each secondary cache compatible value starts with a Slice referencing the uncompressed block contents (already true or mostly true), but we apparently intend to stack secondary caches. Saving an entry from a compressed secondary to a lower tier requires custom handling offered by SaveToCallback, etc.
* Make CreateCallback part of the helper and introduce CreateContext to work with it (alternative to https://github.com/facebook/rocksdb/issues/10562). This cleans up the interface while still allowing context to be provided for loading/parsing values into primary cache. This model works for async lookup in BlockBasedTable reader (reader owns a CreateContext) under the assumption that it always waits on secondary cache operations to finish. (Otherwise, the CreateContext could be destroyed while async operation depending on it continues.) This likely contributes most to the observed performance improvement because it saves an std::function backed by a heap allocation.
* Use char* for serialized data, e.g. in SaveToCallback, where void* was confusingly used. (We use `char*` for serialized byte data all over RocksDB, with many advantages over `void*`. `memcpy` etc. are legacy APIs that should not be mimicked.)
* Add a type alias Cache::ObjectPtr = void*, so that we can better indicate the intent of the void* when it is to be the object associated with a Cache entry. Related: started (but did not complete) a refactoring to move away from "value" of a cache entry toward "object" or "obj". (It is confusing to call Cache a key-value store (like DB) when it is really storing arbitrary in-memory objects, not byte strings.)
* Remove unnecessary key param from DeleterFn. This is good for efficiency in HyperClockCache, which does not directly store the cache key in memory. (Alternative to https://github.com/facebook/rocksdb/issues/10774)
* Add allocator to Cache DeleterFn. This is a kind of future-proofing change in case we get more serious about using the Cache allocator for memory tracked by the Cache. Right now, only the uncompressed block contents are allocated using the allocator, and a pointer to that allocator is saved as part of the cached object so that the deleter can use it. (See CacheAllocationPtr.) If in the future we are able to "flatten out" our Cache objects some more, it would be good not to have to track the allocator as part of each object.
* Removes legacy `ApplyToAllCacheEntries` and changes `ApplyToAllEntries` signature for Deleter->CacheItemHelper change.
## typed_cache.h
Adds various "typed" interfaces to the Cache as internal APIs, so that most uses of Cache can use simple type safe code without casting and without explicit deleters, etc. Almost all of the non-test, non-glue code uses of Cache have been migrated. (Follow-up work: CompressedSecondaryCache deserves deeper attention to migrate.) This change expands RocksDB's internal usage of metaprogramming and SFINAE (https://en.cppreference.com/w/cpp/language/sfinae).
The existing usages of Cache are divided up at a high level into these new interfaces. See updated existing uses of Cache for examples of how these are used.
* PlaceholderCacheInterface - Used for making cache reservations, with entries that have a charge but no value.
* BasicTypedCacheInterface<TValue> - Used for primary cache storage of objects of type TValue, which can be cleaned up with std::default_delete<TValue>. The role is provided by TValue::kCacheEntryRole or given in an optional template parameter.
* FullTypedCacheInterface<TValue, TCreateContext> - Used for secondary cache compatible storage of objects of type TValue. In addition to BasicTypedCacheInterface constraints, we require TValue::ContentSlice() to return persistable data. This simplifies usage for the normal case of simple secondary cache compatibility (can give you a Slice to the data already in memory). In addition to TCreateContext performing the role of Cache::CreateContext, it is also expected to provide a factory function for creating TValue.
* For each of these, there's a "Shared" version (e.g. FullTypedSharedCacheInterface) that holds a shared_ptr to the Cache, rather than assuming external ownership by holding only a raw `Cache*`.
These interfaces introduce specific handle types for each interface instantiation, so that it's easy to see what kind of object is controlled by a handle. (Ultimately, this might not be worth the extra complexity, but it seems OK so far.)
Note: I attempted to make the cache 'charge' automatically inferred from the cache object type, such as by expecting an ApproximateMemoryUsage() function, but this is not so clean because there are cases where we need to compute the charge ahead of time and don't want to re-compute it.
## block_cache.h
This header is essentially the replacement for the old block_like_traits.h. It includes various things to support block cache access with typed_cache.h for block-based table.
## block_based_table_reader.cc
Before this change, accessing the block cache here was an awkward mix of static polymorphism (template TBlocklike) and switch-case on a dynamic BlockType value. This change mostly unifies on static polymorphism, relying on minor hacks in block_cache.h to distinguish variants of Block. We still check BlockType in some places (especially for stats, which could be improved in follow-up work) but at least the BlockType is a static constant from the template parameter. (No more awkward partial redundancy between static and dynamic info.) This likely contributes to the overall performance improvement, but hasn't been tested in isolation.
The other key source of simplification here is a more unified system of creating block cache objects: for directly populating from primary cache and for promotion from secondary cache. Both use BlockCreateContext, for context and for factory functions.
## block_based_table_builder.cc, cache_dump_load_impl.cc
Before this change, warming caches was super ugly code. Both of these source files had switch statements to basically transition from the dynamic BlockType world to the static TBlocklike world. None of that mess is needed anymore as there's a new, untyped WarmInCache function that handles all the details just as promotion from SecondaryCache would. (Fixes `TODO akanksha: Dedup below code` in block_based_table_builder.cc.)
## Everything else
Mostly just updating Cache users to use new typed APIs when reasonably possible, or changed Cache APIs when not.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10975
Test Plan:
tests updated
Performance test setup similar to https://github.com/facebook/rocksdb/issues/10626 (by cache size, LRUCache when not "hyper" for HyperClockCache):
34MB 1thread base.hyper -> kops/s: 0.745 io_bytes/op: 2.52504e+06 miss_ratio: 0.140906 max_rss_mb: 76.4844
34MB 1thread new.hyper -> kops/s: 0.751 io_bytes/op: 2.5123e+06 miss_ratio: 0.140161 max_rss_mb: 79.3594
34MB 1thread base -> kops/s: 0.254 io_bytes/op: 1.36073e+07 miss_ratio: 0.918818 max_rss_mb: 45.9297
34MB 1thread new -> kops/s: 0.252 io_bytes/op: 1.36157e+07 miss_ratio: 0.918999 max_rss_mb: 44.1523
34MB 32thread base.hyper -> kops/s: 7.272 io_bytes/op: 2.88323e+06 miss_ratio: 0.162532 max_rss_mb: 516.602
34MB 32thread new.hyper -> kops/s: 7.214 io_bytes/op: 2.99046e+06 miss_ratio: 0.168818 max_rss_mb: 518.293
34MB 32thread base -> kops/s: 3.528 io_bytes/op: 1.35722e+07 miss_ratio: 0.914691 max_rss_mb: 264.926
34MB 32thread new -> kops/s: 3.604 io_bytes/op: 1.35744e+07 miss_ratio: 0.915054 max_rss_mb: 264.488
233MB 1thread base.hyper -> kops/s: 53.909 io_bytes/op: 2552.35 miss_ratio: 0.0440566 max_rss_mb: 241.984
233MB 1thread new.hyper -> kops/s: 62.792 io_bytes/op: 2549.79 miss_ratio: 0.044043 max_rss_mb: 241.922
233MB 1thread base -> kops/s: 1.197 io_bytes/op: 2.75173e+06 miss_ratio: 0.103093 max_rss_mb: 241.559
233MB 1thread new -> kops/s: 1.199 io_bytes/op: 2.73723e+06 miss_ratio: 0.10305 max_rss_mb: 240.93
233MB 32thread base.hyper -> kops/s: 1298.69 io_bytes/op: 2539.12 miss_ratio: 0.0440307 max_rss_mb: 371.418
233MB 32thread new.hyper -> kops/s: 1421.35 io_bytes/op: 2538.75 miss_ratio: 0.0440307 max_rss_mb: 347.273
233MB 32thread base -> kops/s: 9.693 io_bytes/op: 2.77304e+06 miss_ratio: 0.103745 max_rss_mb: 569.691
233MB 32thread new -> kops/s: 9.75 io_bytes/op: 2.77559e+06 miss_ratio: 0.103798 max_rss_mb: 552.82
1597MB 1thread base.hyper -> kops/s: 58.607 io_bytes/op: 1449.14 miss_ratio: 0.0249324 max_rss_mb: 1583.55
1597MB 1thread new.hyper -> kops/s: 69.6 io_bytes/op: 1434.89 miss_ratio: 0.0247167 max_rss_mb: 1584.02
1597MB 1thread base -> kops/s: 60.478 io_bytes/op: 1421.28 miss_ratio: 0.024452 max_rss_mb: 1589.45
1597MB 1thread new -> kops/s: 63.973 io_bytes/op: 1416.07 miss_ratio: 0.0243766 max_rss_mb: 1589.24
1597MB 32thread base.hyper -> kops/s: 1436.2 io_bytes/op: 1357.93 miss_ratio: 0.0235353 max_rss_mb: 1692.92
1597MB 32thread new.hyper -> kops/s: 1605.03 io_bytes/op: 1358.04 miss_ratio: 0.023538 max_rss_mb: 1702.78
1597MB 32thread base -> kops/s: 280.059 io_bytes/op: 1350.34 miss_ratio: 0.023289 max_rss_mb: 1675.36
1597MB 32thread new -> kops/s: 283.125 io_bytes/op: 1351.05 miss_ratio: 0.0232797 max_rss_mb: 1703.83
Almost uniformly improving over base revision, especially for hot paths with HyperClockCache, up to 12% higher throughput seen (1597MB, 32thread, hyper). The improvement for that is likely coming from much simplified code for providing context for secondary cache promotion (CreateCallback/CreateContext), and possibly from less branching in block_based_table_reader. And likely a small improvement from not reconstituting key for DeleterFn.
Reviewed By: anand1976
Differential Revision: D42417818
Pulled By: pdillinger
fbshipit-source-id: f86bfdd584dce27c028b151ba56818ad14f7a432
2023-01-11 22:20:40 +00:00
|
|
|
file_meta->fd.table_reader = table_cache_->get_cache().Value(handle);
|
2014-11-04 01:45:55 +00:00
|
|
|
}
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
2018-08-21 00:31:28 +00:00
|
|
|
});
|
2015-08-11 19:19:56 +00:00
|
|
|
|
2018-06-15 19:28:06 +00:00
|
|
|
std::vector<port::Thread> threads;
|
|
|
|
for (int i = 1; i < max_threads; i++) {
|
|
|
|
threads.emplace_back(load_handlers_func);
|
|
|
|
}
|
|
|
|
load_handlers_func();
|
|
|
|
for (auto& t : threads) {
|
|
|
|
t.join();
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
2020-10-03 05:09:28 +00:00
|
|
|
Status ret;
|
2019-03-26 23:41:31 +00:00
|
|
|
for (const auto& s : statuses) {
|
|
|
|
if (!s.ok()) {
|
2020-10-03 05:09:28 +00:00
|
|
|
if (ret.ok()) {
|
|
|
|
ret = s;
|
|
|
|
}
|
2019-03-26 23:41:31 +00:00
|
|
|
}
|
|
|
|
}
|
2020-10-03 05:09:28 +00:00
|
|
|
return ret;
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
|
|
|
};
|
|
|
|
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
VersionBuilder::VersionBuilder(
|
|
|
|
const FileOptions& file_options, const ImmutableCFOptions* ioptions,
|
|
|
|
TableCache* table_cache, VersionStorageInfo* base_vstorage,
|
|
|
|
VersionSet* version_set,
|
2024-08-13 04:09:37 +00:00
|
|
|
std::shared_ptr<CacheReservationManager> file_metadata_cache_res_mgr,
|
|
|
|
ColumnFamilyData* cfd, VersionEditHandler* version_edit_handler,
|
2024-08-17 00:18:54 +00:00
|
|
|
bool track_found_and_missing_files, bool allow_incomplete_valid_version)
|
2020-04-30 18:23:32 +00:00
|
|
|
: rep_(new Rep(file_options, ioptions, table_cache, base_vstorage,
|
2024-08-13 04:09:37 +00:00
|
|
|
version_set, file_metadata_cache_res_mgr, cfd,
|
2024-08-17 00:18:54 +00:00
|
|
|
version_edit_handler, track_found_and_missing_files,
|
|
|
|
allow_incomplete_valid_version)) {}
|
2017-08-24 23:05:16 +00:00
|
|
|
|
2020-03-19 17:40:31 +00:00
|
|
|
VersionBuilder::~VersionBuilder() = default;
|
2017-08-24 23:05:16 +00:00
|
|
|
|
|
|
|
bool VersionBuilder::CheckConsistencyForNumLevels() {
|
|
|
|
return rep_->CheckConsistencyForNumLevels();
|
|
|
|
}
|
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
Status VersionBuilder::Apply(const VersionEdit* edit) {
|
|
|
|
return rep_->Apply(edit);
|
|
|
|
}
|
2017-08-24 23:05:16 +00:00
|
|
|
|
2021-11-03 18:51:22 +00:00
|
|
|
Status VersionBuilder::SaveTo(VersionStorageInfo* vstorage) const {
|
2019-08-29 21:06:07 +00:00
|
|
|
return rep_->SaveTo(vstorage);
|
2014-10-31 18:54:05 +00:00
|
|
|
}
|
2017-08-24 23:05:16 +00:00
|
|
|
|
2019-03-26 23:41:31 +00:00
|
|
|
Status VersionBuilder::LoadTableHandlers(
|
|
|
|
InternalStats* internal_stats, int max_threads,
|
|
|
|
bool prefetch_index_and_filter_in_cache, bool is_initial_load,
|
2024-10-17 21:13:20 +00:00
|
|
|
const MutableCFOptions& mutable_cf_options,
|
|
|
|
size_t max_file_size_for_l0_meta_pin, const ReadOptions& read_options) {
|
|
|
|
return rep_->LoadTableHandlers(internal_stats, max_threads,
|
|
|
|
prefetch_index_and_filter_in_cache,
|
|
|
|
is_initial_load, mutable_cf_options,
|
|
|
|
max_file_size_for_l0_meta_pin, read_options);
|
2015-08-11 19:19:56 +00:00
|
|
|
}
|
2017-08-24 23:05:16 +00:00
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
void VersionBuilder::CreateOrReplaceSavePoint() {
|
|
|
|
assert(rep_);
|
|
|
|
savepoint_ = std::move(rep_);
|
|
|
|
rep_ = std::make_unique<Rep>(*savepoint_);
|
|
|
|
}
|
|
|
|
|
2024-08-17 00:18:54 +00:00
|
|
|
bool VersionBuilder::ValidVersionAvailable() {
|
|
|
|
return rep_->ValidVersionAvailable();
|
2021-04-19 18:55:20 +00:00
|
|
|
}
|
|
|
|
|
2024-08-13 04:09:37 +00:00
|
|
|
bool VersionBuilder::HasMissingFiles() const { return rep_->HasMissingFiles(); }
|
|
|
|
|
|
|
|
std::vector<std::string>& VersionBuilder::GetAndClearIntermediateFiles() {
|
|
|
|
return rep_->GetAndClearIntermediateFiles();
|
|
|
|
}
|
|
|
|
|
|
|
|
void VersionBuilder::ClearFoundFiles() { return rep_->ClearFoundFiles(); }
|
|
|
|
|
|
|
|
Status VersionBuilder::SaveSavePointTo(VersionStorageInfo* vstorage) const {
|
2024-08-17 00:18:54 +00:00
|
|
|
if (!savepoint_ || !savepoint_->ValidVersionAvailable()) {
|
2024-08-13 04:09:37 +00:00
|
|
|
return Status::InvalidArgument();
|
|
|
|
}
|
|
|
|
return savepoint_->SaveTo(vstorage);
|
|
|
|
}
|
|
|
|
|
|
|
|
Status VersionBuilder::LoadSavePointTableHandlers(
|
|
|
|
InternalStats* internal_stats, int max_threads,
|
|
|
|
bool prefetch_index_and_filter_in_cache, bool is_initial_load,
|
2024-10-17 21:13:20 +00:00
|
|
|
const MutableCFOptions& mutable_cf_options,
|
|
|
|
size_t max_file_size_for_l0_meta_pin, const ReadOptions& read_options) {
|
2024-08-17 00:18:54 +00:00
|
|
|
if (!savepoint_ || !savepoint_->ValidVersionAvailable()) {
|
2024-08-13 04:09:37 +00:00
|
|
|
return Status::InvalidArgument();
|
|
|
|
}
|
|
|
|
return savepoint_->LoadTableHandlers(
|
|
|
|
internal_stats, max_threads, prefetch_index_and_filter_in_cache,
|
2024-10-17 21:13:20 +00:00
|
|
|
is_initial_load, mutable_cf_options, max_file_size_for_l0_meta_pin,
|
|
|
|
read_options);
|
2024-08-13 04:09:37 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void VersionBuilder::ClearSavePoint() { savepoint_.reset(nullptr); }
|
|
|
|
|
2020-03-21 02:17:54 +00:00
|
|
|
BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
|
2024-08-13 04:09:37 +00:00
|
|
|
ColumnFamilyData* cfd, VersionEditHandler* version_edit_handler,
|
2024-08-17 00:18:54 +00:00
|
|
|
bool track_found_and_missing_files, bool allow_incomplete_valid_version)
|
2020-03-21 02:17:54 +00:00
|
|
|
: version_builder_(new VersionBuilder(
|
2020-04-30 18:23:32 +00:00
|
|
|
cfd->current()->version_set()->file_options(), cfd->ioptions(),
|
|
|
|
cfd->table_cache(), cfd->current()->storage_info(),
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
cfd->current()->version_set(),
|
2024-08-13 04:09:37 +00:00
|
|
|
cfd->GetFileMetadataCacheReservationManager(), cfd,
|
2024-08-17 00:18:54 +00:00
|
|
|
version_edit_handler, track_found_and_missing_files,
|
|
|
|
allow_incomplete_valid_version)),
|
2020-03-21 02:17:54 +00:00
|
|
|
version_(cfd->current()) {
|
|
|
|
version_->Ref();
|
|
|
|
}
|
|
|
|
|
|
|
|
BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
|
2024-08-13 04:09:37 +00:00
|
|
|
ColumnFamilyData* cfd, Version* v, VersionEditHandler* version_edit_handler,
|
2024-08-17 00:18:54 +00:00
|
|
|
bool track_found_and_missing_files, bool allow_incomplete_valid_version)
|
2020-03-21 02:17:54 +00:00
|
|
|
: version_builder_(new VersionBuilder(
|
2020-04-30 18:23:32 +00:00
|
|
|
cfd->current()->version_set()->file_options(), cfd->ioptions(),
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
2022-06-14 20:06:40 +00:00
|
|
|
cfd->table_cache(), v->storage_info(), v->version_set(),
|
2024-08-13 04:09:37 +00:00
|
|
|
cfd->GetFileMetadataCacheReservationManager(), cfd,
|
2024-08-17 00:18:54 +00:00
|
|
|
version_edit_handler, track_found_and_missing_files,
|
|
|
|
allow_incomplete_valid_version)),
|
2020-03-21 02:17:54 +00:00
|
|
|
version_(v) {
|
|
|
|
assert(version_ != cfd->current());
|
|
|
|
}
|
|
|
|
|
|
|
|
BaseReferencedVersionBuilder::~BaseReferencedVersionBuilder() {
|
|
|
|
version_->Unref();
|
|
|
|
}
|
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|