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

1259 lines
49 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
#include "db/compaction/compaction_iterator.h"
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
#include <iterator>
#include <limits>
#include "db/blob/blob_fetcher.h"
#include "db/blob/blob_file_builder.h"
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
#include "db/blob/blob_index.h"
#include "db/blob/prefetch_buffer_collection.h"
#include "db/snapshot_checker.h"
#include "logging/logging.h"
#include "port/likely.h"
#include "rocksdb/listener.h"
#include "table/internal_iterator.h"
#include "test_util/sync_point.h"
namespace ROCKSDB_NAMESPACE {
CompactionIterator::CompactionIterator(
InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
SequenceNumber earliest_write_conflict_snapshot,
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
Env* env, bool report_detailed_time, bool expect_valid_internal_key,
CompactionRangeDelAggregator* range_del_agg,
BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
bool enforce_single_del_contracts,
const std::atomic<bool>& manual_compaction_canceled,
const Compaction* compaction, const CompactionFilter* compaction_filter,
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
7 years ago
const std::atomic<bool>* shutting_down,
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
const std::shared_ptr<Logger> info_log,
const std::string* full_history_ts_low)
: CompactionIterator(
input, cmp, merge_helper, last_sequence, snapshots,
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
report_detailed_time, expect_valid_internal_key, range_del_agg,
blob_file_builder, allow_data_in_errors, enforce_single_del_contracts,
manual_compaction_canceled,
std::unique_ptr<CompactionProxy>(
compaction ? new RealCompaction(compaction) : nullptr),
compaction_filter, shutting_down, info_log, full_history_ts_low) {}
CompactionIterator::CompactionIterator(
InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
SequenceNumber /*last_sequence*/, std::vector<SequenceNumber>* snapshots,
SequenceNumber earliest_write_conflict_snapshot,
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
Env* env, bool report_detailed_time, bool expect_valid_internal_key,
CompactionRangeDelAggregator* range_del_agg,
BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
bool enforce_single_del_contracts,
const std::atomic<bool>& manual_compaction_canceled,
std::unique_ptr<CompactionProxy> compaction,
const CompactionFilter* compaction_filter,
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
7 years ago
const std::atomic<bool>* shutting_down,
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
const std::shared_ptr<Logger> info_log,
const std::string* full_history_ts_low)
: input_(input, cmp,
!compaction || compaction->DoesInputReferenceBlobFiles()),
cmp_(cmp),
merge_helper_(merge_helper),
snapshots_(snapshots),
earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
job_snapshot_(job_snapshot),
snapshot_checker_(snapshot_checker),
env_(env),
clock_(env_->GetSystemClock().get()),
report_detailed_time_(report_detailed_time),
expect_valid_internal_key_(expect_valid_internal_key),
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
range_del_agg_(range_del_agg),
blob_file_builder_(blob_file_builder),
compaction_(std::move(compaction)),
compaction_filter_(compaction_filter),
shutting_down_(shutting_down),
manual_compaction_canceled_(manual_compaction_canceled),
info_log_(info_log),
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
allow_data_in_errors_(allow_data_in_errors),
enforce_single_del_contracts_(enforce_single_del_contracts),
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
timestamp_size_(cmp_ ? cmp_->timestamp_size() : 0),
full_history_ts_low_(full_history_ts_low),
current_user_key_sequence_(0),
current_user_key_snapshot_(0),
merge_out_iter_(merge_helper_),
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
blob_garbage_collection_cutoff_file_number_(
ComputeBlobGarbageCollectionCutoffFileNumber(compaction_.get())),
blob_fetcher_(CreateBlobFetcherIfNeeded(compaction_.get())),
prefetch_buffers_(
CreatePrefetchBufferCollectionIfNeeded(compaction_.get())),
current_key_committed_(false),
cmp_with_history_ts_low_(0),
level_(compaction_ == nullptr ? 0 : compaction_->level()) {
assert(snapshots_ != nullptr);
bottommost_level_ = compaction_ == nullptr
? false
: compaction_->bottommost_level() &&
!compaction_->allow_ingest_behind();
if (compaction_ != nullptr) {
level_ptrs_ = std::vector<size_t>(compaction_->number_levels(), 0);
}
if (snapshots_->size() == 0) {
// optimize for fast path if there are no snapshots
visible_at_tip_ = true;
earliest_snapshot_iter_ = snapshots_->end();
earliest_snapshot_ = kMaxSequenceNumber;
latest_snapshot_ = 0;
} else {
visible_at_tip_ = false;
earliest_snapshot_iter_ = snapshots_->begin();
earliest_snapshot_ = snapshots_->at(0);
latest_snapshot_ = snapshots_->back();
}
#ifndef NDEBUG
// findEarliestVisibleSnapshot assumes this ordering.
for (size_t i = 1; i < snapshots_->size(); ++i) {
assert(snapshots_->at(i - 1) < snapshots_->at(i));
}
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
assert(timestamp_size_ == 0 || !full_history_ts_low_ ||
timestamp_size_ == full_history_ts_low_->size());
#endif
input_.SetPinnedItersMgr(&pinned_iters_mgr_);
TEST_SYNC_POINT_CALLBACK("CompactionIterator:AfterInit", compaction_.get());
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
8 years ago
}
CompactionIterator::~CompactionIterator() {
// input_ Iterator lifetime is longer than pinned_iters_mgr_ lifetime
input_.SetPinnedItersMgr(nullptr);
}
void CompactionIterator::ResetRecordCounts() {
iter_stats_.num_record_drop_user = 0;
iter_stats_.num_record_drop_hidden = 0;
iter_stats_.num_record_drop_obsolete = 0;
iter_stats_.num_record_drop_range_del = 0;
iter_stats_.num_range_del_drop_obsolete = 0;
iter_stats_.num_optimized_del_drop_obsolete = 0;
}
void CompactionIterator::SeekToFirst() {
NextFromInput();
PrepareOutput();
}
void CompactionIterator::Next() {
// If there is a merge output, return it before continuing to process the
// input.
if (merge_out_iter_.Valid()) {
merge_out_iter_.Next();
// Check if we returned all records of the merge output.
if (merge_out_iter_.Valid()) {
key_ = merge_out_iter_.key();
value_ = merge_out_iter_.value();
Status s = ParseInternalKey(key_, &ikey_, allow_data_in_errors_);
// MergeUntil stops when it encounters a corrupt key and does not
// include them in the result, so we expect the keys here to be valid.
if (!s.ok()) {
ROCKS_LOG_FATAL(
info_log_, "Invalid ikey %s in compaction. %s",
allow_data_in_errors_ ? key_.ToString(true).c_str() : "hidden",
s.getState());
assert(false);
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// Keep current_key_ in sync.
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
key_ = current_key_.GetInternalKey();
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
ikey_.user_key = current_key_.GetUserKey();
valid_ = true;
} else {
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
8 years ago
// We consumed all pinned merge operands, release pinned iterators
pinned_iters_mgr_.ReleasePinnedData();
// MergeHelper moves the iterator to the first record after the merged
// records, so even though we reached the end of the merge output, we do
// not want to advance the iterator.
NextFromInput();
}
} else {
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// Only advance the input iterator if there is no merge output and the
// iterator is not already at the next record.
if (!at_next_) {
AdvanceInputIter();
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
}
NextFromInput();
}
if (valid_) {
// Record that we've outputted a record for the current key.
has_outputted_key_ = true;
}
PrepareOutput();
}
bool CompactionIterator::InvokeFilterIfNeeded(bool* need_skip,
Slice* skip_until) {
// TODO: support compaction filter for wide-column entities
if (!compaction_filter_ ||
(ikey_.type != kTypeValue && ikey_.type != kTypeBlobIndex)) {
return true;
}
bool error = false;
// If the user has specified a compaction filter and the sequence
// number is greater than any external snapshot, then invoke the
// filter. If the return value of the compaction filter is true,
// replace the entry with a deletion marker.
CompactionFilter::Decision filter = CompactionFilter::Decision::kUndetermined;
compaction_filter_value_.clear();
compaction_filter_skip_until_.Clear();
CompactionFilter::ValueType value_type =
ikey_.type == kTypeValue ? CompactionFilter::ValueType::kValue
: CompactionFilter::ValueType::kBlobIndex;
// Hack: pass internal key to BlobIndexCompactionFilter since it needs
// to get sequence number.
assert(compaction_filter_);
Slice& filter_key =
(ikey_.type == kTypeValue ||
!compaction_filter_->IsStackedBlobDbInternalCompactionFilter())
? ikey_.user_key
: key_;
{
StopWatchNano timer(clock_, report_detailed_time_);
if (kTypeBlobIndex == ikey_.type) {
blob_value_.Reset();
filter = compaction_filter_->FilterBlobByKey(
level_, filter_key, &compaction_filter_value_,
compaction_filter_skip_until_.rep());
if (CompactionFilter::Decision::kUndetermined == filter &&
!compaction_filter_->IsStackedBlobDbInternalCompactionFilter()) {
if (compaction_ == nullptr) {
status_ =
Status::Corruption("Unexpected blob index outside of compaction");
valid_ = false;
return false;
}
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::InvokeFilterIfNeeded::TamperWithBlobIndex",
&value_);
// For integrated BlobDB impl, CompactionIterator reads blob value.
// For Stacked BlobDB impl, the corresponding CompactionFilter's
// FilterV2 method should read the blob value.
BlobIndex blob_index;
Status s = blob_index.DecodeFrom(value_);
if (!s.ok()) {
status_ = s;
valid_ = false;
return false;
}
FilePrefetchBuffer* prefetch_buffer =
prefetch_buffers_ ? prefetch_buffers_->GetOrCreatePrefetchBuffer(
blob_index.file_number())
: nullptr;
uint64_t bytes_read = 0;
assert(blob_fetcher_);
s = blob_fetcher_->FetchBlob(ikey_.user_key, blob_index,
prefetch_buffer, &blob_value_,
&bytes_read);
if (!s.ok()) {
status_ = s;
valid_ = false;
return false;
}
++iter_stats_.num_blobs_read;
iter_stats_.total_blob_bytes_read += bytes_read;
value_type = CompactionFilter::ValueType::kValue;
}
}
if (CompactionFilter::Decision::kUndetermined == filter) {
filter = compaction_filter_->FilterV2(
level_, filter_key, value_type,
blob_value_.empty() ? value_ : blob_value_, &compaction_filter_value_,
compaction_filter_skip_until_.rep());
}
iter_stats_.total_filter_time +=
env_ != nullptr && report_detailed_time_ ? timer.ElapsedNanos() : 0;
}
if (CompactionFilter::Decision::kUndetermined == filter) {
// Should not reach here, since FilterV2 should never return kUndetermined.
status_ =
Status::NotSupported("FilterV2() should never return kUndetermined");
valid_ = false;
return false;
}
if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil &&
cmp_->Compare(*compaction_filter_skip_until_.rep(), ikey_.user_key) <=
0) {
// Can't skip to a key smaller than the current one.
// Keep the key as per FilterV2 documentation.
filter = CompactionFilter::Decision::kKeep;
}
if (filter == CompactionFilter::Decision::kRemove) {
// convert the current key to a delete; key_ is pointing into
// current_key_ at this point, so updating current_key_ updates key()
ikey_.type = kTypeDeletion;
current_key_.UpdateInternalKey(ikey_.sequence, kTypeDeletion);
// no value associated with delete
value_.clear();
iter_stats_.num_record_drop_user++;
} else if (filter == CompactionFilter::Decision::kPurge) {
// convert the current key to a single delete; key_ is pointing into
// current_key_ at this point, so updating current_key_ updates key()
ikey_.type = kTypeSingleDeletion;
current_key_.UpdateInternalKey(ikey_.sequence, kTypeSingleDeletion);
// no value associated with single delete
value_.clear();
iter_stats_.num_record_drop_user++;
} else if (filter == CompactionFilter::Decision::kChangeValue) {
if (ikey_.type == kTypeBlobIndex) {
// value transfer from blob file to inlined data
ikey_.type = kTypeValue;
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
}
value_ = compaction_filter_value_;
} else if (filter == CompactionFilter::Decision::kRemoveAndSkipUntil) {
*need_skip = true;
compaction_filter_skip_until_.ConvertFromUserKey(kMaxSequenceNumber,
kValueTypeForSeek);
*skip_until = compaction_filter_skip_until_.Encode();
} else if (filter == CompactionFilter::Decision::kChangeBlobIndex) {
// Only the StackableDB-based BlobDB impl's compaction filter should return
// kChangeBlobIndex. Decision about rewriting blob and changing blob index
// in the integrated BlobDB impl is made in subsequent call to
// PrepareOutput() and its callees.
if (!compaction_filter_->IsStackedBlobDbInternalCompactionFilter()) {
status_ = Status::NotSupported(
"Only stacked BlobDB's internal compaction filter can return "
"kChangeBlobIndex.");
valid_ = false;
return false;
}
if (ikey_.type == kTypeValue) {
// value transfer from inlined data to blob file
ikey_.type = kTypeBlobIndex;
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
}
value_ = compaction_filter_value_;
} else if (filter == CompactionFilter::Decision::kIOError) {
if (!compaction_filter_->IsStackedBlobDbInternalCompactionFilter()) {
status_ = Status::NotSupported(
"CompactionFilter for integrated BlobDB should not return kIOError");
valid_ = false;
return false;
}
status_ = Status::IOError("Failed to access blob during compaction filter");
error = true;
}
return !error;
}
void CompactionIterator::NextFromInput() {
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
at_next_ = false;
valid_ = false;
while (!valid_ && input_.Valid() && !IsPausingManualCompaction() &&
!IsShuttingDown()) {
key_ = input_.key();
value_ = input_.value();
iter_stats_.num_input_records++;
Status pik_status = ParseInternalKey(key_, &ikey_, allow_data_in_errors_);
if (!pik_status.ok()) {
iter_stats_.num_input_corrupt_records++;
// If `expect_valid_internal_key_` is false, return the corrupted key
// and let the caller decide what to do with it.
if (expect_valid_internal_key_) {
status_ = pik_status;
return;
}
key_ = current_key_.SetInternalKey(key_);
has_current_user_key_ = false;
current_user_key_sequence_ = kMaxSequenceNumber;
current_user_key_snapshot_ = 0;
valid_ = true;
break;
}
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
TEST_SYNC_POINT_CALLBACK("CompactionIterator:ProcessKV", &ikey_);
// Update input statistics
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
if (ikey_.type == kTypeDeletion || ikey_.type == kTypeSingleDeletion ||
ikey_.type == kTypeDeletionWithTimestamp) {
iter_stats_.num_input_deletion_records++;
}
iter_stats_.total_input_raw_key_bytes += key_.size();
iter_stats_.total_input_raw_value_bytes += value_.size();
// If need_skip is true, we should seek the input iterator
// to internal key skip_until and continue from there.
bool need_skip = false;
// Points either into compaction_filter_skip_until_ or into
// merge_helper_->compaction_filter_skip_until_.
Slice skip_until;
Enable backward iterator for keys with user-defined timestamp (#8035) Summary: This PR does the following: - Enable backward iteration for keys with user-defined timestamp. Note that merge, single delete, range delete are not supported yet. - Introduces a new helper API `Comparator::EqualWithoutTimestamp()`. - Fix a typo in `SetTimestamp()`. - Add/update unit tests Run db_bench (built with DEBUG_LEVEL=0) to demonstrate that no overhead is introduced for CPU-intensive workloads with a lot of `Prev()`. Also provided results of iterating keys with timestamps. 1. Disable timestamp, run: ``` ./db_bench -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Results: > Baseline > - seekrandom [AVG 6 runs] : 96115 ops/sec; 53.2 MB/sec > - seekrandom [MEDIAN 6 runs] : 98075 ops/sec; 54.2 MB/sec > > This PR > - seekrandom [AVG 6 runs] : 95521 ops/sec; 52.8 MB/sec > - seekrandom [MEDIAN 6 runs] : 96338 ops/sec; 53.3 MB/sec 2. Enable timestamp, run: ``` ./db_bench -user_timestamp_size=8 -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Result: > Baseline: not supported > > This PR > - seekrandom [AVG 6 runs] : 90514 ops/sec; 50.1 MB/sec > - seekrandom [MEDIAN 6 runs] : 90834 ops/sec; 50.2 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/8035 Reviewed By: ltamasi Differential Revision: D26926668 Pulled By: riversand963 fbshipit-source-id: 95330cc2242397c03e09d29e5417dfb0adc98ef5
4 years ago
bool user_key_equal_without_ts = false;
int cmp_ts = 0;
if (has_current_user_key_) {
Enable backward iterator for keys with user-defined timestamp (#8035) Summary: This PR does the following: - Enable backward iteration for keys with user-defined timestamp. Note that merge, single delete, range delete are not supported yet. - Introduces a new helper API `Comparator::EqualWithoutTimestamp()`. - Fix a typo in `SetTimestamp()`. - Add/update unit tests Run db_bench (built with DEBUG_LEVEL=0) to demonstrate that no overhead is introduced for CPU-intensive workloads with a lot of `Prev()`. Also provided results of iterating keys with timestamps. 1. Disable timestamp, run: ``` ./db_bench -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Results: > Baseline > - seekrandom [AVG 6 runs] : 96115 ops/sec; 53.2 MB/sec > - seekrandom [MEDIAN 6 runs] : 98075 ops/sec; 54.2 MB/sec > > This PR > - seekrandom [AVG 6 runs] : 95521 ops/sec; 52.8 MB/sec > - seekrandom [MEDIAN 6 runs] : 96338 ops/sec; 53.3 MB/sec 2. Enable timestamp, run: ``` ./db_bench -user_timestamp_size=8 -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Result: > Baseline: not supported > > This PR > - seekrandom [AVG 6 runs] : 90514 ops/sec; 50.1 MB/sec > - seekrandom [MEDIAN 6 runs] : 90834 ops/sec; 50.2 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/8035 Reviewed By: ltamasi Differential Revision: D26926668 Pulled By: riversand963 fbshipit-source-id: 95330cc2242397c03e09d29e5417dfb0adc98ef5
4 years ago
user_key_equal_without_ts =
cmp_->EqualWithoutTimestamp(ikey_.user_key, current_user_key_);
// if timestamp_size_ > 0, then curr_ts_ has been initialized by a
// previous key.
cmp_ts = timestamp_size_ ? cmp_->CompareTimestamp(
ExtractTimestampFromUserKey(
ikey_.user_key, timestamp_size_),
curr_ts_)
: 0;
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// Check whether the user key changed. After this if statement current_key_
// is a copy of the current input key (maybe converted to a delete by the
// compaction filter). ikey_.user_key is pointing to the copy.
Enable backward iterator for keys with user-defined timestamp (#8035) Summary: This PR does the following: - Enable backward iteration for keys with user-defined timestamp. Note that merge, single delete, range delete are not supported yet. - Introduces a new helper API `Comparator::EqualWithoutTimestamp()`. - Fix a typo in `SetTimestamp()`. - Add/update unit tests Run db_bench (built with DEBUG_LEVEL=0) to demonstrate that no overhead is introduced for CPU-intensive workloads with a lot of `Prev()`. Also provided results of iterating keys with timestamps. 1. Disable timestamp, run: ``` ./db_bench -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Results: > Baseline > - seekrandom [AVG 6 runs] : 96115 ops/sec; 53.2 MB/sec > - seekrandom [MEDIAN 6 runs] : 98075 ops/sec; 54.2 MB/sec > > This PR > - seekrandom [AVG 6 runs] : 95521 ops/sec; 52.8 MB/sec > - seekrandom [MEDIAN 6 runs] : 96338 ops/sec; 53.3 MB/sec 2. Enable timestamp, run: ``` ./db_bench -user_timestamp_size=8 -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Result: > Baseline: not supported > > This PR > - seekrandom [AVG 6 runs] : 90514 ops/sec; 50.1 MB/sec > - seekrandom [MEDIAN 6 runs] : 90834 ops/sec; 50.2 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/8035 Reviewed By: ltamasi Differential Revision: D26926668 Pulled By: riversand963 fbshipit-source-id: 95330cc2242397c03e09d29e5417dfb0adc98ef5
4 years ago
if (!has_current_user_key_ || !user_key_equal_without_ts || cmp_ts != 0) {
// First occurrence of this user key
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
// Copy key for output
key_ = current_key_.SetInternalKey(key_, &ikey_);
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
int prev_cmp_with_ts_low =
!full_history_ts_low_ ? 0
: curr_ts_.empty()
? 0
: cmp_->CompareTimestamp(curr_ts_, *full_history_ts_low_);
// If timestamp_size_ > 0, then copy from ikey_ to curr_ts_ for the use
// in next iteration to compare with the timestamp of next key.
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
UpdateTimestampAndCompareWithFullHistoryLow();
// If
// (1) !has_current_user_key_, OR
// (2) timestamp is disabled, OR
// (3) all history will be preserved, OR
// (4) user key (excluding timestamp) is different from previous key, OR
// (5) timestamp is NO older than *full_history_ts_low_, OR
// (6) timestamp is the largest one older than full_history_ts_low_,
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
// then current_user_key_ must be treated as a different user key.
// This means, if a user key (excluding ts) is the same as the previous
// user key, and its ts is older than *full_history_ts_low_, then we
// consider this key for GC, e.g. it may be dropped if certain conditions
// match.
if (!has_current_user_key_ || !timestamp_size_ || !full_history_ts_low_ ||
!user_key_equal_without_ts || cmp_with_history_ts_low_ >= 0 ||
prev_cmp_with_ts_low >= 0) {
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
// Initialize for future comparison for rule (A) and etc.
current_user_key_sequence_ = kMaxSequenceNumber;
current_user_key_snapshot_ = 0;
has_current_user_key_ = true;
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
current_user_key_ = ikey_.user_key;
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
has_outputted_key_ = false;
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
last_key_seq_zeroed_ = false;
current_key_committed_ = KeyCommitted(ikey_.sequence);
// Apply the compaction filter to the first committed version of the user
// key.
if (current_key_committed_ &&
!InvokeFilterIfNeeded(&need_skip, &skip_until)) {
break;
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
} else {
// Update the current key to reflect the new sequence number/type without
// copying the user key.
// TODO(rven): Compaction filter does not process keys in this path
// Need to have the compaction filter process multiple versions
// if we have versions on both sides of a snapshot
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
key_ = current_key_.GetInternalKey();
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
ikey_.user_key = current_key_.GetUserKey();
// Note that newer version of a key is ordered before older versions. If a
// newer version of a key is committed, so as the older version. No need
// to query snapshot_checker_ in that case.
if (UNLIKELY(!current_key_committed_)) {
assert(snapshot_checker_ != nullptr);
current_key_committed_ = KeyCommitted(ikey_.sequence);
// Apply the compaction filter to the first committed version of the
// user key.
if (current_key_committed_ &&
!InvokeFilterIfNeeded(&need_skip, &skip_until)) {
break;
}
}
}
if (UNLIKELY(!current_key_committed_)) {
assert(snapshot_checker_ != nullptr);
valid_ = true;
break;
}
// If there are no snapshots, then this kv affect visibility at tip.
// Otherwise, search though all existing snapshots to find the earliest
// snapshot that is affected by this kv.
SequenceNumber last_sequence = current_user_key_sequence_;
current_user_key_sequence_ = ikey_.sequence;
SequenceNumber last_snapshot = current_user_key_snapshot_;
SequenceNumber prev_snapshot = 0; // 0 means no previous snapshot
current_user_key_snapshot_ =
visible_at_tip_
? earliest_snapshot_
: findEarliestVisibleSnapshot(ikey_.sequence, &prev_snapshot);
if (need_skip) {
// This case is handled below.
} else if (clear_and_output_next_key_) {
// In the previous iteration we encountered a single delete that we could
// not compact out. We will keep this Put, but can drop it's data.
// (See Optimization 3, below.)
if (ikey_.type != kTypeValue && ikey_.type != kTypeBlobIndex &&
ikey_.type != kTypeWideColumnEntity) {
ROCKS_LOG_FATAL(info_log_, "Unexpected key %s for compaction output",
ikey_.DebugString(allow_data_in_errors_, true).c_str());
assert(false);
}
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
if (current_user_key_snapshot_ < last_snapshot) {
ROCKS_LOG_FATAL(info_log_,
"key %s, current_user_key_snapshot_ (%" PRIu64
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
") < last_snapshot (%" PRIu64 ")",
ikey_.DebugString(allow_data_in_errors_, true).c_str(),
current_user_key_snapshot_, last_snapshot);
assert(false);
}
if (ikey_.type == kTypeBlobIndex || ikey_.type == kTypeWideColumnEntity) {
ikey_.type = kTypeValue;
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
}
value_.clear();
valid_ = true;
clear_and_output_next_key_ = false;
} else if (ikey_.type == kTypeSingleDeletion) {
// We can compact out a SingleDelete if:
// 1) We encounter the corresponding PUT -OR- we know that this key
// doesn't appear past this output level
// =AND=
// 2) We've already returned a record in this snapshot -OR-
// there are no earlier earliest_write_conflict_snapshot.
//
// A note about 2) above:
// we try to determine whether there is any earlier write conflict
// checking snapshot by calling DefinitelyInSnapshot() with seq and
// earliest_write_conflict_snapshot as arguments. For write-prepared
// and write-unprepared transactions, if earliest_write_conflict_snapshot
// is evicted from WritePreparedTxnDB::commit_cache, then
// DefinitelyInSnapshot(seq, earliest_write_conflict_snapshot) returns
// false, even if the seq is actually visible within
// earliest_write_conflict_snapshot. Consequently, CompactionIterator
// may try to zero out its sequence number, thus hitting assertion error
// in debug mode or cause incorrect DBIter return result.
// We observe that earliest_write_conflict_snapshot >= earliest_snapshot,
// and the seq zeroing logic depends on
// DefinitelyInSnapshot(seq, earliest_snapshot). Therefore, if we cannot
// determine whether seq is **definitely** in
// earliest_write_conflict_snapshot, then we can additionally check if
// seq is definitely in earliest_snapshot. If the latter holds, then the
// former holds too.
//
// Rule 1 is needed for SingleDelete correctness. Rule 2 is needed to
// allow Transactions to do write-conflict checking (if we compacted away
// all keys, then we wouldn't know that a write happened in this
// snapshot). If there is no earlier snapshot, then we know that there
// are no active transactions that need to know about any writes.
//
// Optimization 3:
// If we encounter a SingleDelete followed by a PUT and Rule 2 is NOT
// true, then we must output a SingleDelete. In this case, we will decide
// to also output the PUT. While we are compacting less by outputting the
// PUT now, hopefully this will lead to better compaction in the future
// when Rule 2 is later true (Ie, We are hoping we can later compact out
// both the SingleDelete and the Put, while we couldn't if we only
// outputted the SingleDelete now).
// In this case, we can save space by removing the PUT's value as it will
// never be read.
//
// Deletes and Merges are not supported on the same key that has a
// SingleDelete as it is not possible to correctly do any partial
// compaction of such a combination of operations. The result of mixing
// those operations for a given key is documented as being undefined. So
// we can choose how to handle such a combinations of operations. We will
// try to compact out as much as we can in these cases.
// We will report counts on these anomalous cases.
//
// Note: If timestamp is enabled, then record will be eligible for
// deletion, only if, along with above conditions (Rule 1 and Rule 2)
// full_history_ts_low_ is specified and timestamp for that key is less
// than *full_history_ts_low_. If it's not eligible for deletion, then we
// will output the SingleDelete. For Optimization 3 also, if
// full_history_ts_low_ is specified and timestamp for the key is less
// than *full_history_ts_low_ then only optimization will be applied.
// The easiest way to process a SingleDelete during iteration is to peek
// ahead at the next key.
const bool is_timestamp_eligible_for_gc =
(timestamp_size_ == 0 ||
(full_history_ts_low_ && cmp_with_history_ts_low_ < 0));
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
ParsedInternalKey next_ikey;
AdvanceInputIter();
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// Check whether the next key exists, is not corrupt, and is the same key
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// as the single delete.
if (input_.Valid() &&
ParseInternalKey(input_.key(), &next_ikey, allow_data_in_errors_)
.ok() &&
cmp_->EqualWithoutTimestamp(ikey_.user_key, next_ikey.user_key)) {
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
#ifndef NDEBUG
const Compaction* c =
compaction_ ? compaction_->real_compaction() : nullptr;
#endif
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::NextFromInput:SingleDelete:1",
const_cast<Compaction*>(c));
if (last_key_seq_zeroed_) {
++iter_stats_.num_record_drop_hidden;
++iter_stats_.num_record_drop_obsolete;
assert(bottommost_level_);
AdvanceInputIter();
} else if (prev_snapshot == 0 ||
DefinitelyNotInSnapshot(next_ikey.sequence, prev_snapshot)) {
// Check whether the next key belongs to the same snapshot as the
// SingleDelete.
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::NextFromInput:SingleDelete:2", nullptr);
if (next_ikey.type == kTypeSingleDeletion) {
// We encountered two SingleDeletes for same key in a row. This
Fix assertion error during compaction with write-prepared txn enabled (#9105) Summary: Pull Request resolved: https://github.com/facebook/rocksdb/pull/9105 The user contract of SingleDelete is that: a SingleDelete can only be issued to a key that exists and has NOT been updated. For example, application can insert one key `key`, and uses a SingleDelete to delete it in the future. The `key` cannot be updated or removed using Delete. In reality, especially when write-prepared transaction is being used, things can get tricky. For example, a prepared transaction already writes `key` to the memtable after a successful Prepare(). Afterwards, should the transaction rollback, it will insert a Delete into the memtable to cancel out the prior Put. Consider the following sequence of operations. ``` // operation sequence 1 Begin txn Put(key) Prepare() Flush() Rollback txn Flush() ``` There will be two SSTs resulting from above. One of the contains a PUT, while the second one contains a Delete. It is also known that releasing a snapshot can lead to an L0 containing only a SD for a particular key. Consider the following operations following the above block. ``` // operation sequence 2 db->Put(key) db->SingleDelete(key) Flush() ``` The operation sequence 2 can result in an L0 with only the SD. Should there be a snapshot for conflict checking created before operation sequence 1, then an attempt to compact the db may hit the assertion failure below, because ikey_.type is Delete (from a rollback). ``` else if (clear_and_output_next_key_) { assert(ikey_.type == kTypeValue || ikey_.type == kTypeBlobIndex); } ``` To fix the assertion failure, we can skip the SingleDelete if we detect an earlier Delete in the same snapshot interval. Reviewed By: ltamasi Differential Revision: D32056848 fbshipit-source-id: 23620a91e28562d91c45cf7e95f414b54b729748
3 years ago
// could be due to unexpected user input. If write-(un)prepared
// transaction is used, this could also be due to releasing an old
// snapshot between a Put and its matching SingleDelete.
// Skip the first SingleDelete and let the next iteration decide
// how to handle the second SingleDelete.
// First SingleDelete has been skipped since we already called
// input_.Next().
++iter_stats_.num_record_drop_obsolete;
++iter_stats_.num_single_del_mismatch;
} else if (next_ikey.type == kTypeDeletion) {
std::ostringstream oss;
oss << "Found SD and type: " << static_cast<int>(next_ikey.type)
<< " on the same key, violating the contract "
"of SingleDelete. Check your application to make sure the "
"application does not mix SingleDelete and Delete for "
"the same key. If you are using "
"write-prepared/write-unprepared transactions, and use "
"SingleDelete to delete certain keys, then make sure "
"TransactionDBOptions::rollback_deletion_type_callback is "
"configured properly. Mixing SD and DEL can lead to "
"undefined behaviors";
++iter_stats_.num_record_drop_obsolete;
++iter_stats_.num_single_del_mismatch;
if (enforce_single_del_contracts_) {
ROCKS_LOG_ERROR(info_log_, "%s", oss.str().c_str());
valid_ = false;
status_ = Status::Corruption(oss.str());
return;
}
ROCKS_LOG_WARN(info_log_, "%s", oss.str().c_str());
} else if (!is_timestamp_eligible_for_gc) {
// We cannot drop the SingleDelete as timestamp is enabled, and
// timestamp of this key is greater than or equal to
// *full_history_ts_low_. We will output the SingleDelete.
valid_ = true;
} else if (has_outputted_key_ ||
DefinitelyInSnapshot(ikey_.sequence,
earliest_write_conflict_snapshot_) ||
(earliest_snapshot_ < earliest_write_conflict_snapshot_ &&
DefinitelyInSnapshot(ikey_.sequence,
earliest_snapshot_))) {
// Found a matching value, we can drop the single delete and the
// value. It is safe to drop both records since we've already
// outputted a key in this snapshot, or there is no earlier
// snapshot (Rule 2 above).
// Note: it doesn't matter whether the second key is a Put or if it
// is an unexpected Merge or Delete. We will compact it out
// either way. We will maintain counts of how many mismatches
// happened
if (next_ikey.type != kTypeValue &&
next_ikey.type != kTypeBlobIndex &&
next_ikey.type != kTypeWideColumnEntity) {
++iter_stats_.num_single_del_mismatch;
}
++iter_stats_.num_record_drop_hidden;
++iter_stats_.num_record_drop_obsolete;
// Already called input_.Next() once. Call it a second time to
// skip past the second key.
AdvanceInputIter();
} else {
// Found a matching value, but we cannot drop both keys since
// there is an earlier snapshot and we need to leave behind a record
// to know that a write happened in this snapshot (Rule 2 above).
// Clear the value and output the SingleDelete. (The value will be
// outputted on the next iteration.)
// Setting valid_ to true will output the current SingleDelete
valid_ = true;
// Set up the Put to be outputted in the next iteration.
// (Optimization 3).
clear_and_output_next_key_ = true;
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::NextFromInput:KeepSDForWW",
/*arg=*/nullptr);
}
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
} else {
// We hit the next snapshot without hitting a put, so the iterator
// returns the single delete.
valid_ = true;
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::NextFromInput:SingleDelete:3",
const_cast<Compaction*>(c));
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
}
} else {
// We are at the end of the input, could not parse the next key, or hit
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
// a different key. The iterator returns the single delete if the key
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
// possibly exists beyond the current output level. We set
// has_current_user_key to false so that if the iterator is at the next
// key, we do not compare it again against the previous key at the next
// iteration. If the next key is corrupt, we return before the
// comparison, so the value of has_current_user_key does not matter.
has_current_user_key_ = false;
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
if (compaction_ != nullptr &&
DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
&level_ptrs_) &&
is_timestamp_eligible_for_gc) {
// Key doesn't exist outside of this range.
// Can compact out this SingleDelete.
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
++iter_stats_.num_record_drop_obsolete;
++iter_stats_.num_single_del_fallthru;
if (!bottommost_level_) {
++iter_stats_.num_optimized_del_drop_obsolete;
}
} else if (last_key_seq_zeroed_) {
// Skip.
++iter_stats_.num_record_drop_hidden;
++iter_stats_.num_record_drop_obsolete;
assert(bottommost_level_);
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
} else {
// Output SingleDelete
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
9 years ago
valid_ = true;
}
}
if (valid_) {
at_next_ = true;
}
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
} else if (last_snapshot == current_user_key_snapshot_ ||
(last_snapshot > 0 &&
last_snapshot < current_user_key_snapshot_)) {
// If the earliest snapshot is which this key is visible in
// is the same as the visibility of a previous instance of the
// same key, then this kv is not visible in any snapshot.
// Hidden by an newer entry for same user key
//
// Note: Dropping this key will not affect TransactionDB write-conflict
// checking since there has already been a record returned for this key
// in this snapshot.
if (last_sequence < current_user_key_sequence_) {
ROCKS_LOG_FATAL(info_log_,
"key %s, last_sequence (%" PRIu64
") < current_user_key_sequence_ (%" PRIu64 ")",
ikey_.DebugString(allow_data_in_errors_, true).c_str(),
last_sequence, current_user_key_sequence_);
assert(false);
}
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
++iter_stats_.num_record_drop_hidden; // rule (A)
AdvanceInputIter();
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
} else if (compaction_ != nullptr &&
(ikey_.type == kTypeDeletion ||
(ikey_.type == kTypeDeletionWithTimestamp &&
cmp_with_history_ts_low_ < 0)) &&
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
&level_ptrs_)) {
// TODO(noetzli): This is the only place where we use compaction_
// (besides the constructor). We should probably get rid of this
// dependency and find a way to do similar filtering during flushes.
//
// For this user key:
// (1) there is no data in higher levels
// (2) data in lower levels will have larger sequence numbers
// (3) data in layers that are being compacted here and have
// smaller sequence numbers will be dropped in the next
// few iterations of this loop (by rule (A) above).
// Therefore this deletion marker is obsolete and can be dropped.
//
// Note: Dropping this Delete will not affect TransactionDB
// write-conflict checking since it is earlier than any snapshot.
//
// It seems that we can also drop deletion later than earliest snapshot
// given that:
// (1) The deletion is earlier than earliest_write_conflict_snapshot, and
// (2) No value exist earlier than the deletion.
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
//
// Note also that a deletion marker of type kTypeDeletionWithTimestamp
// will be treated as a different user key unless the timestamp is older
// than *full_history_ts_low_.
++iter_stats_.num_record_drop_obsolete;
if (!bottommost_level_) {
++iter_stats_.num_optimized_del_drop_obsolete;
}
AdvanceInputIter();
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
} else if ((ikey_.type == kTypeDeletion ||
(ikey_.type == kTypeDeletionWithTimestamp &&
cmp_with_history_ts_low_ < 0)) &&
bottommost_level_) {
// Handle the case where we have a delete key at the bottom most level
// We can skip outputting the key iff there are no subsequent puts for this
// key
assert(!compaction_ || compaction_->KeyNotExistsBeyondOutputLevel(
ikey_.user_key, &level_ptrs_));
ParsedInternalKey next_ikey;
AdvanceInputIter();
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
#ifndef NDEBUG
const Compaction* c =
compaction_ ? compaction_->real_compaction() : nullptr;
#endif
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::NextFromInput:BottommostDelete:1",
const_cast<Compaction*>(c));
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
// Skip over all versions of this key that happen to occur in the same
// snapshot range as the delete.
//
// Note that a deletion marker of type kTypeDeletionWithTimestamp will be
// considered to have a different user key unless the timestamp is older
// than *full_history_ts_low_.
while (!IsPausingManualCompaction() && !IsShuttingDown() &&
input_.Valid() &&
(ParseInternalKey(input_.key(), &next_ikey, allow_data_in_errors_)
.ok()) &&
Enable backward iterator for keys with user-defined timestamp (#8035) Summary: This PR does the following: - Enable backward iteration for keys with user-defined timestamp. Note that merge, single delete, range delete are not supported yet. - Introduces a new helper API `Comparator::EqualWithoutTimestamp()`. - Fix a typo in `SetTimestamp()`. - Add/update unit tests Run db_bench (built with DEBUG_LEVEL=0) to demonstrate that no overhead is introduced for CPU-intensive workloads with a lot of `Prev()`. Also provided results of iterating keys with timestamps. 1. Disable timestamp, run: ``` ./db_bench -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Results: > Baseline > - seekrandom [AVG 6 runs] : 96115 ops/sec; 53.2 MB/sec > - seekrandom [MEDIAN 6 runs] : 98075 ops/sec; 54.2 MB/sec > > This PR > - seekrandom [AVG 6 runs] : 95521 ops/sec; 52.8 MB/sec > - seekrandom [MEDIAN 6 runs] : 96338 ops/sec; 53.3 MB/sec 2. Enable timestamp, run: ``` ./db_bench -user_timestamp_size=8 -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Result: > Baseline: not supported > > This PR > - seekrandom [AVG 6 runs] : 90514 ops/sec; 50.1 MB/sec > - seekrandom [MEDIAN 6 runs] : 90834 ops/sec; 50.2 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/8035 Reviewed By: ltamasi Differential Revision: D26926668 Pulled By: riversand963 fbshipit-source-id: 95330cc2242397c03e09d29e5417dfb0adc98ef5
4 years ago
cmp_->EqualWithoutTimestamp(ikey_.user_key, next_ikey.user_key) &&
(prev_snapshot == 0 ||
DefinitelyNotInSnapshot(next_ikey.sequence, prev_snapshot))) {
AdvanceInputIter();
}
// If you find you still need to output a row with this key, we need to output the
// delete too
if (input_.Valid() &&
(ParseInternalKey(input_.key(), &next_ikey, allow_data_in_errors_)
.ok()) &&
Enable backward iterator for keys with user-defined timestamp (#8035) Summary: This PR does the following: - Enable backward iteration for keys with user-defined timestamp. Note that merge, single delete, range delete are not supported yet. - Introduces a new helper API `Comparator::EqualWithoutTimestamp()`. - Fix a typo in `SetTimestamp()`. - Add/update unit tests Run db_bench (built with DEBUG_LEVEL=0) to demonstrate that no overhead is introduced for CPU-intensive workloads with a lot of `Prev()`. Also provided results of iterating keys with timestamps. 1. Disable timestamp, run: ``` ./db_bench -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Results: > Baseline > - seekrandom [AVG 6 runs] : 96115 ops/sec; 53.2 MB/sec > - seekrandom [MEDIAN 6 runs] : 98075 ops/sec; 54.2 MB/sec > > This PR > - seekrandom [AVG 6 runs] : 95521 ops/sec; 52.8 MB/sec > - seekrandom [MEDIAN 6 runs] : 96338 ops/sec; 53.3 MB/sec 2. Enable timestamp, run: ``` ./db_bench -user_timestamp_size=8 -db=/dev/shm/rocksdb -disable_wal=1 -benchmarks=fillseq,seekrandom[-W1-X6] -reverse_iterator=1 -seek_nexts=5 ``` Result: > Baseline: not supported > > This PR > - seekrandom [AVG 6 runs] : 90514 ops/sec; 50.1 MB/sec > - seekrandom [MEDIAN 6 runs] : 90834 ops/sec; 50.2 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/8035 Reviewed By: ltamasi Differential Revision: D26926668 Pulled By: riversand963 fbshipit-source-id: 95330cc2242397c03e09d29e5417dfb0adc98ef5
4 years ago
cmp_->EqualWithoutTimestamp(ikey_.user_key, next_ikey.user_key)) {
valid_ = true;
at_next_ = true;
}
} else if (ikey_.type == kTypeMerge) {
if (!merge_helper_->HasOperator()) {
status_ = Status::InvalidArgument(
"merge_operator is not properly initialized.");
return;
}
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
8 years ago
pinned_iters_mgr_.StartPinning();
// We know the merge type entry is not hidden, otherwise we would
// have hit (A)
// We encapsulate the merge related state machine in a different
// object to minimize change to the existing flow.
Status s = merge_helper_->MergeUntil(
&input_, range_del_agg_, prev_snapshot, bottommost_level_,
allow_data_in_errors_, blob_fetcher_.get(), prefetch_buffers_.get(),
&iter_stats_);
merge_out_iter_.SeekToFirst();
if (!s.ok() && !s.IsMergeInProgress()) {
status_ = s;
return;
} else if (merge_out_iter_.Valid()) {
// NOTE: key, value, and ikey_ refer to old entries.
// These will be correctly set below.
key_ = merge_out_iter_.key();
value_ = merge_out_iter_.value();
pik_status = ParseInternalKey(key_, &ikey_, allow_data_in_errors_);
// MergeUntil stops when it encounters a corrupt key and does not
// include them in the result, so we expect the keys here to valid.
if (!pik_status.ok()) {
ROCKS_LOG_FATAL(
info_log_, "Invalid key %s in compaction. %s",
allow_data_in_errors_ ? key_.ToString(true).c_str() : "hidden",
pik_status.getState());
assert(false);
}
// Keep current_key_ in sync.
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
key_ = current_key_.GetInternalKey();
ikey_.user_key = current_key_.GetUserKey();
valid_ = true;
} else {
// all merge operands were filtered out. reset the user key, since the
// batch consumed by the merge operator should not shadow any keys
// coming after the merges
has_current_user_key_ = false;
pinned_iters_mgr_.ReleasePinnedData();
if (merge_helper_->FilteredUntil(&skip_until)) {
need_skip = true;
}
}
} else {
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
// 1. new user key -OR-
// 2. different snapshot stripe
bool should_delete = range_del_agg_->ShouldDelete(
key_, RangeDelPositioningMode::kForwardTraversal);
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
if (should_delete) {
++iter_stats_.num_record_drop_hidden;
++iter_stats_.num_record_drop_range_del;
AdvanceInputIter();
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
} else {
valid_ = true;
}
}
if (need_skip) {
SkipUntil(skip_until);
}
}
if (!valid_ && IsShuttingDown()) {
status_ = Status::ShutdownInProgress();
}
if (IsPausingManualCompaction()) {
status_ = Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
}
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
bool CompactionIterator::ExtractLargeValueIfNeededImpl() {
if (!blob_file_builder_) {
return false;
}
blob_index_.clear();
const Status s = blob_file_builder_->Add(user_key(), value_, &blob_index_);
if (!s.ok()) {
status_ = s;
valid_ = false;
return false;
}
if (blob_index_.empty()) {
return false;
}
value_ = blob_index_;
return true;
}
void CompactionIterator::ExtractLargeValueIfNeeded() {
assert(ikey_.type == kTypeValue);
if (!ExtractLargeValueIfNeededImpl()) {
return;
}
ikey_.type = kTypeBlobIndex;
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
}
void CompactionIterator::GarbageCollectBlobIfNeeded() {
assert(ikey_.type == kTypeBlobIndex);
if (!compaction_) {
return;
}
// GC for integrated BlobDB
if (compaction_->enable_blob_garbage_collection()) {
TEST_SYNC_POINT_CALLBACK(
"CompactionIterator::GarbageCollectBlobIfNeeded::TamperWithBlobIndex",
&value_);
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
BlobIndex blob_index;
{
const Status s = blob_index.DecodeFrom(value_);
if (!s.ok()) {
status_ = s;
valid_ = false;
return;
}
}
if (blob_index.file_number() >=
blob_garbage_collection_cutoff_file_number_) {
return;
}
FilePrefetchBuffer* prefetch_buffer =
prefetch_buffers_ ? prefetch_buffers_->GetOrCreatePrefetchBuffer(
blob_index.file_number())
: nullptr;
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
uint64_t bytes_read = 0;
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
{
assert(blob_fetcher_);
const Status s = blob_fetcher_->FetchBlob(
user_key(), blob_index, prefetch_buffer, &blob_value_, &bytes_read);
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
if (!s.ok()) {
status_ = s;
valid_ = false;
return;
}
}
++iter_stats_.num_blobs_read;
iter_stats_.total_blob_bytes_read += bytes_read;
++iter_stats_.num_blobs_relocated;
iter_stats_.total_blob_bytes_relocated += blob_index.size();
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
value_ = blob_value_;
if (ExtractLargeValueIfNeededImpl()) {
return;
}
ikey_.type = kTypeValue;
current_key_.UpdateInternalKey(ikey_.sequence, ikey_.type);
return;
}
// GC for stacked BlobDB
if (compaction_filter_ &&
compaction_filter_->IsStackedBlobDbInternalCompactionFilter()) {
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
const auto blob_decision = compaction_filter_->PrepareBlobOutput(
user_key(), value_, &compaction_filter_value_);
if (blob_decision == CompactionFilter::BlobDecision::kCorruption) {
status_ =
Status::Corruption("Corrupted blob reference encountered during GC");
valid_ = false;
return;
}
if (blob_decision == CompactionFilter::BlobDecision::kIOError) {
status_ = Status::IOError("Could not relocate blob during GC");
valid_ = false;
return;
}
if (blob_decision == CompactionFilter::BlobDecision::kChangeValue) {
value_ = compaction_filter_value_;
return;
}
}
}
void CompactionIterator::PrepareOutput() {
if (valid_) {
if (ikey_.type == kTypeValue) {
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
ExtractLargeValueIfNeeded();
} else if (ikey_.type == kTypeBlobIndex) {
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
GarbageCollectBlobIfNeeded();
}
// Zeroing out the sequence number leads to better compression.
// If this is the bottommost level (no files in lower levels)
// and the earliest snapshot is larger than this seqno
// and the userkey differs from the last userkey in compaction
// then we can squash the seqno to zero.
//
// This is safe for TransactionDB write-conflict checking since transactions
// only care about sequence number larger than any active snapshots.
//
// Can we do the same for levels above bottom level as long as
// KeyNotExistsBeyondOutputLevel() return true?
if (valid_ && compaction_ != nullptr &&
!compaction_->allow_ingest_behind() && bottommost_level_ &&
Fix bug caused by releasing snapshot(s) during compaction (#8608) Summary: In debug mode, we are seeing assertion failure as follows ``` db/compaction/compaction_iterator.cc:980: void rocksdb::CompactionIterator::PrepareOutput(): \ Assertion `ikey_.type != kTypeDeletion && ikey_.type != kTypeSingleDeletion' failed. ``` It is caused by releasing earliest snapshot during compaction between the execution of `NextFromInput()` and `PrepareOutput()`. In one case, as demonstrated in unit test `WritePreparedTransaction.ReleaseEarliestSnapshotDuringCompaction_WithSD2`, incorrect result may be returned by a following range scan if we disable assertion, as in opt compilation level: the SingleDelete marker's sequence number is zeroed out, but the preceding PUT is also outputted to the SST file after compaction. Due to the logic of DBIter, the PUT will not be skipped and will be returned by iterator in range scan. https://github.com/facebook/rocksdb/issues/8661 illustrates what happened. Fix by taking a more conservative approach: make compaction zero out sequence number only if key is in the earliest snapshot when the compaction starts. Another assertion failure is ``` Assertion `current_user_key_snapshot_ == last_snapshot' failed. ``` It's caused by releasing the snapshot between the PUT and SingleDelete during compaction. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8608 Test Plan: make check Reviewed By: jay-zhuang Differential Revision: D30145645 Pulled By: riversand963 fbshipit-source-id: 699f58e66faf70732ad53810ccef43935d3bbe81
3 years ago
DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
ikey_.type != kTypeMerge && current_key_committed_) {
if (ikey_.type == kTypeDeletion ||
(ikey_.type == kTypeSingleDeletion && timestamp_size_ == 0)) {
ROCKS_LOG_FATAL(
info_log_,
"Unexpected key %s for seq-zero optimization. "
"earliest_snapshot %" PRIu64
", earliest_write_conflict_snapshot %" PRIu64
" job_snapshot %" PRIu64
". timestamp_size: %d full_history_ts_low_ %s",
ikey_.DebugString(allow_data_in_errors_, true).c_str(),
earliest_snapshot_, earliest_write_conflict_snapshot_,
job_snapshot_, static_cast<int>(timestamp_size_),
full_history_ts_low_ != nullptr
? Slice(*full_history_ts_low_).ToString(true).c_str()
: "null");
assert(false);
}
ikey_.sequence = 0;
last_key_seq_zeroed_ = true;
TEST_SYNC_POINT_CALLBACK("CompactionIterator::PrepareOutput:ZeroingSeq",
&ikey_);
Allow compaction iterator to perform garbage collection (#7556) Summary: Add a threshold timestamp, full_history_ts_low_ of type `std::string*` to `CompactionIterator`, so that RocksDB can also perform garbage collection during compaction. * If full_history_ts_low_ is nullptr, then compaction iterator does not perform GC, preserving all timestamp history for all keys. Compaction iterator will treat user key with different timestamps as different user keys. * If full_history_ts_low_ is not nullptr, then compaction iterator performs GC. GC will look at keys older than `*full_history_ts_low_` and determine their eligibility based on factors including snapshots. Current rules of GC: * If an internal key is in the same snapshot as a previous counterpart with the same user key, and this key is eligible for GC, and the key is not single-delete or merge operand, then this key can be dropped. Note that the previous internal key cannot be a merge operand either. * If a tombstone is the most recent one in the earliest snapshot and it is eligible for GC, and keyNotExistsBeyondLevel() is true, then this tombstone can be dropped. * If a tombstone is the most recent one in a snapshot and it is eligible for GC, and the compaction is at bottommost level, then all other older internal keys of the same user key must also be eligible for GC, thus can be dropped * Single-delete, delete-range and merge are not currently supported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7556 Test Plan: make check Reviewed By: ltamasi Differential Revision: D24507728 Pulled By: riversand963 fbshipit-source-id: 3c09c7301f41eed76dfcf4d1527e68cf6e0a8bb3
4 years ago
if (!timestamp_size_) {
current_key_.UpdateInternalKey(0, ikey_.type);
} else if (full_history_ts_low_ && cmp_with_history_ts_low_ < 0) {
// We can also zero out timestamp for better compression.
// For the same user key (excluding timestamp), the timestamp-based
// history can be collapsed to save some space if the timestamp is
// older than *full_history_ts_low_.
const std::string kTsMin(timestamp_size_, static_cast<char>(0));
const Slice ts_slice = kTsMin;
ikey_.SetTimestamp(ts_slice);
current_key_.UpdateInternalKey(0, ikey_.type, &ts_slice);
}
}
}
}
inline SequenceNumber CompactionIterator::findEarliestVisibleSnapshot(
SequenceNumber in, SequenceNumber* prev_snapshot) {
assert(snapshots_->size());
if (snapshots_->size() == 0) {
ROCKS_LOG_FATAL(info_log_,
"No snapshot left in findEarliestVisibleSnapshot");
}
auto snapshots_iter = std::lower_bound(
snapshots_->begin(), snapshots_->end(), in);
assert(prev_snapshot != nullptr);
if (snapshots_iter == snapshots_->begin()) {
*prev_snapshot = 0;
} else {
*prev_snapshot = *std::prev(snapshots_iter);
if (*prev_snapshot >= in) {
ROCKS_LOG_FATAL(info_log_,
"*prev_snapshot (%" PRIu64 ") >= in (%" PRIu64
") in findEarliestVisibleSnapshot",
*prev_snapshot, in);
assert(false);
}
}
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
if (snapshot_checker_ == nullptr) {
return snapshots_iter != snapshots_->end()
? *snapshots_iter : kMaxSequenceNumber;
}
bool has_released_snapshot = !released_snapshots_.empty();
for (; snapshots_iter != snapshots_->end(); ++snapshots_iter) {
auto cur = *snapshots_iter;
if (in > cur) {
ROCKS_LOG_FATAL(info_log_,
"in (%" PRIu64 ") > cur (%" PRIu64
") in findEarliestVisibleSnapshot",
in, cur);
assert(false);
}
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
// Skip if cur is in released_snapshots.
if (has_released_snapshot && released_snapshots_.count(cur) > 0) {
continue;
}
auto res = snapshot_checker_->CheckInSnapshot(in, cur);
if (res == SnapshotCheckerResult::kInSnapshot) {
return cur;
WritePrepared: fix two versions in compaction see different status for released snapshots (#4890) Summary: Fix how CompactionIterator::findEarliestVisibleSnapshots handles released snapshot. It fixing the two scenarios: Scenario 1: key1 has two values v1 and v2. There're two snapshots s1 and s2 taken after v1 and v2 are committed. Right after compaction output v2, s1 is released. Now findEarliestVisibleSnapshot may see s1 being released, and return the next snapshot, which is s2. That's larger than v2's earliest visible snapshot, which was s1. The fix: the only place we check against last snapshot and current key snapshot is when we decide whether to compact out a value if it is hidden by a later value. In the check if we see current snapshot is even larger than last snapshot, we know last snapshot is released, and we are safe to compact out current key. Scenario 2: key1 has two values v1 and v2. there are two snapshots s1 and s2 taken after v1 and v2 are committed. During compaction before we process the key, s1 is released. When compaction process v2, snapshot checker may return kSnapshotReleased, and the earliest visible snapshot for v2 become s2. When compaction process v1, snapshot checker may return kIsInSnapshot (for WritePrepared transaction, it could be because v1 is still in commit cache). The result will become inconsistent here. The fix: remember the set of released snapshots ever reported by snapshot checker, and ignore them when finding result for findEarliestVisibleSnapshot. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4890 Differential Revision: D13705538 Pulled By: maysamyabandeh fbshipit-source-id: e577f0d9ee1ff5a6035f26859e56902ecc85a5a4
6 years ago
} else if (res == SnapshotCheckerResult::kSnapshotReleased) {
released_snapshots_.insert(cur);
}
*prev_snapshot = cur;
}
return kMaxSequenceNumber;
}
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
uint64_t CompactionIterator::ComputeBlobGarbageCollectionCutoffFileNumber(
const CompactionProxy* compaction) {
if (!compaction) {
return 0;
}
if (!compaction->enable_blob_garbage_collection()) {
return 0;
}
const Version* const version = compaction->input_version();
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
assert(version);
const VersionStorageInfo* const storage_info = version->storage_info();
assert(storage_info);
const auto& blob_files = storage_info->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
3 years ago
const size_t cutoff_index = static_cast<size_t>(
compaction->blob_garbage_collection_age_cutoff() * blob_files.size());
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
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
3 years ago
if (cutoff_index >= blob_files.size()) {
return std::numeric_limits<uint64_t>::max();
}
const auto& meta = blob_files[cutoff_index];
assert(meta);
return meta->GetBlobFileNumber();
Integrated blob garbage collection: relocate blobs (#7694) Summary: The patch adds basic garbage collection support to the integrated BlobDB implementation. Valid blobs residing in the oldest blob files are relocated as they are encountered during compaction. The threshold that determines which blob files qualify is computed based on the configuration option `blob_garbage_collection_age_cutoff`, which was introduced in https://github.com/facebook/rocksdb/issues/7661 . Once a blob is retrieved for the purposes of relocation, it passes through the same logic that extracts large values to blob files in general. This means that if, for instance, the size threshold for key-value separation (`min_blob_size`) got changed or writing blob files got disabled altogether, it is possible for the value to be moved back into the LSM tree. In particular, one way to re-inline all blob values if needed would be to perform a full manual compaction with `enable_blob_files` set to `false`, `enable_blob_garbage_collection` set to `true`, and `blob_file_garbage_collection_age_cutoff` set to `1.0`. Some TODOs that I plan to address in separate PRs: 1) We'll have to measure the amount of new garbage in each blob file and log `BlobFileGarbage` entries as part of the compaction job's `VersionEdit`. (For the time being, blob files are cleaned up solely based on the `oldest_blob_file_number` relationships.) 2) When compression is used for blobs, the compression type hasn't changed, and the blob still qualifies for being written to a blob file, we can simply copy the compressed blob to the new file instead of going through decompression and compression. 3) We need to update the formula for computing write amplification to account for the amount of data read from blob files as part of GC. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7694 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D25069663 Pulled By: ltamasi fbshipit-source-id: bdfa8feb09afcf5bca3b4eba2ba72ce2f15cd06a
4 years ago
}
std::unique_ptr<BlobFetcher> CompactionIterator::CreateBlobFetcherIfNeeded(
const CompactionProxy* compaction) {
if (!compaction) {
return nullptr;
}
const Version* const version = compaction->input_version();
if (!version) {
return nullptr;
}
return std::unique_ptr<BlobFetcher>(new BlobFetcher(version, ReadOptions()));
}
std::unique_ptr<PrefetchBufferCollection>
CompactionIterator::CreatePrefetchBufferCollectionIfNeeded(
const CompactionProxy* compaction) {
if (!compaction) {
return nullptr;
}
if (!compaction->input_version()) {
return nullptr;
}
if (compaction->allow_mmap_reads()) {
return nullptr;
}
const uint64_t readahead_size = compaction->blob_compaction_readahead_size();
if (!readahead_size) {
return nullptr;
}
return std::unique_ptr<PrefetchBufferCollection>(
new PrefetchBufferCollection(readahead_size));
}
} // namespace ROCKSDB_NAMESPACE