Summary:
Partitioned metadata blocks were introduced back in 2017 to deal more gracefully with large DBs where RAM is relatively scarce and some data might be much colder than other data. The feature allows metadata blocks to compete for memory in the block cache against data blocks while alleviating tail latencies and thrash conditions that can arise with large metadata blocks (sometimes megabytes each) that can arise with large SST files. In general, the cost to partitioned metadata is more CPU in accesses (especially for filters where more binary search is needed before hashing can be used) and a bit more memory fragmentation and related overheads.
However the feature has always had a subtle limitation with a subtle effect on performance: index partitions and filter partitions must be cut at the same time, regardless of which wins the space race (hahaha) to metadata_block_size. Commonly filters will be a few times larger than indexes, so index partitions will be under-sized compared to filter (and data) blocks. While this does affect fragmentation and related overheads a bit, I suspect the bigger impact on performance is in the block cache. The coupling of the partition cuts would be defensible if the binary search done to find the filter block was used (on filter hit) to short-circuit binary search to an index partition, but that optimization has not been developed.
Consider two metadata blocks, an under-sized one and a normal-sized one, covering proportional sections of the key space with the same density of read queries. The under-sized one will be more prone to eviction from block cache because it is used less often. This is unfair because of its despite its proportionally smaller cost of keeping in block cache, and most of the cost of a miss to re-load it (random IO) is not proportional to the size (similar latency etc. up to ~32KB).
## This change
Adds a new table option decouple_partitioned_filters allows filter blocks and index blocks to be cut independently. To make this work, the partitioned filter block builder needs to know about the previous key, to generate an appropriate separator for the partition index. In most cases, BlockBasedTableBuilder already has easy access to the previous key to provide to the filter block builder.
This change includes refactoring to pass that previous key to the filter builder when available, with the filter building caching the previous key itself when unavailable, such as during compression dictionary training and some unit tests. Access to the previous key eliminates the need to track the previous prefix, which results in a small SST construction CPU win in prefix filtering cases, regardless of coupling, and possibly a small regression for some non-prefix cases, regardless of coupling, but still overall improvement especially with https://github.com/facebook/rocksdb/issues/12931.
Suggested follow-up:
* Update confusing use of "last key" to refer to "previous key"
* Expand unit test coverage with parallel compression and dictionary training
* Consider an option or enhancement to alleviate under-sized metadata blocks "at the end" of an SST file due to no coordination or awareness of when files are cut.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12939
Test Plan:
unit tests updated. Also did some unit test runs with "hard wired" usage of parallel compression and dictionary training code paths to ensure they were working. Also ran blackbox_crash_test for a while with the new feature.
## SST write performance (CPU)
Using the same testing setup as in https://github.com/facebook/rocksdb/issues/12931 but with -decouple_partitioned_filters=1 in the "after" configuration, which benchmarking shows makes almost no difference in terms of SST write CPU. "After" vs. "before" this PR
```
-partition_index_and_filters=0 -prefix_size=0 -whole_key_filtering=1
923691 vs. 924851 (-0.13%)
-partition_index_and_filters=0 -prefix_size=8 -whole_key_filtering=0
921398 vs. 922973 (-0.17%)
-partition_index_and_filters=0 -prefix_size=8 -whole_key_filtering=1
902259 vs. 908756 (-0.71%)
-partition_index_and_filters=1 -prefix_size=8 -whole_key_filtering=0
917932 vs. 916901 (+0.60%)
-partition_index_and_filters=1 -prefix_size=8 -whole_key_filtering=0
912755 vs. 907298 (+0.60%)
-partition_index_and_filters=1 -prefix_size=8 -whole_key_filtering=1
899754 vs. 892433 (+0.82%)
```
I think this is a pretty good trade, especially in attracting more movement toward partitioned configurations.
## Read performance
Let's see how decoupling affects read performance across various degrees of memory constraint. To simplify LSM structure, we're using FIFO compaction. Since decoupling will overall increase metadata block size, we control for this somewhat with an extra "before" configuration with larger metadata block size setting (8k instead of 4k). Basic setup:
```
(for CS in 0300 1200; do TEST_TMPDIR=/dev/shm/rocksdb1 ./db_bench -benchmarks=fillrandom,flush,readrandom,block_cache_entry_stats -num=5000000 -duration=30 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=10 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters=1 -statistics=1 -cache_size=${CS}000000 -metadata_block_size=4096 -decouple_partitioned_filters=1 2>&1 | tee results-$CS; done)
```
And read ops/s results:
```CSV
Cache size MB,After/decoupled/4k,Before/4k,Before/8k
3,15593,15158,12826
6,16295,16693,14134
10,20427,20813,18459
20,27035,26836,27384
30,33250,31810,33846
60,35518,32585,35329
100,36612,31805,35292
300,35780,31492,35481
1000,34145,31551,35411
1100,35219,31380,34302
1200,35060,31037,34322
```
If you graph this with log scale on the X axis (internal link: https://pxl.cl/5qKRc), you see that the decoupled/4k configuration is essentially the best of both the before/4k and before/8k configurations: handles really tight memory closer to the old 4k configuration and handles generous memory closer to the old 8k configuration.
Reviewed By: jowlyzhang
Differential Revision: D61376772
Pulled By: pdillinger
fbshipit-source-id: fc2af2aee44290e2d9620f79651a30640799e01f
Summary:
This is in part a refactoring / simplification to set up for "decoupled" partitioned filters and in part to fix an intentional regression for a correctness fix in https://github.com/facebook/rocksdb/issues/12872. Basically, we are taking out some complexity of the filter block builders, and pushing part of it (simultaneous de-duplication of prefixes and whole keys) into the filter bits builders, where it is more efficient by operating on hashes (rather than copied keys).
Previously, the FullFilterBlockBuilder had a somewhat fragile and confusing set of conditions under which it would keep a copy of the most recent prefix and most recent whole key, along with some other state that is essentially redundant. Now we just track (always) the previous prefix in the PartitionedFilterBlockBuilder, to deal with the boundary prefix Seek filtering problem. (Btw, the next PR will optimize this away since BlockBasedTableReader already tracks the previous key.) And to deal with the problem of de-duplicating both whole keys and prefixes going into a single filter, we add a new function to FilterBitsBuilder that has that extra de-duplication capabilty, which is relatively efficient because we only have to cache an extra 64-bit hash, not a copied key or prefix. (The API of this new function is somewhat awkward to avoid a small CPU regression in some cases.)
Also previously, there was awkward logic split between FullFilterBlockBuilder and PartitionedFilterBlockBuilder to deal with some things specific to partitioning. And confusing names like Add vs. AddKey. FullFilterBlockBuilder is much cleaner and simplified now.
The splitting of PartitionedFilterBlockBuilder::MaybeCutAFilterBlock into DecideCutAFilterBlock and CutAFilterBlock is to address what would have been a slight performance regression in some cases. The split allows for more intruction-level parallelism by reducing unnecessary control dependencies.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12931
Test Plan:
existing tests (with some minor updates)
Also manually ported over the pre-broken regression test described in
https://github.com/facebook/rocksdb/issues/12870 and ran it (passed).
Performance:
Here we validate that an entire series of recent related PRs are a net improvement in aggregate. "Before" is with these PRs reverted: https://github.com/facebook/rocksdb/issues/12872#12911https://github.com/facebook/rocksdb/issues/12874#12867https://github.com/facebook/rocksdb/issues/12903#12904. "After" includes this PR (and all
of those, with base revision 16c21af). Simultaneous test script designed to maximally depend on SST construction efficiency:
```
for PF in 0 1; do for PS in 0 8; do for WK in 0 1; do [ "$PS" == "$WK" ] || (for I in `seq 1 20`; do TEST_TMPDIR=/dev/shm/rocksdb2 ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -memtablerep=vector -allow_concurrent_memtable_write=0 -bloom_bits=10 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -partition_index_and_filters=$PF -prefix_size=$PS -whole_key_filtering=$WK 2>&1 | grep micros/op; done) | awk '{ t += $5; c++; print } END { print 1.0 * t / c }'; echo "Was -partition_index_and_filters=$PF -prefix_size=$PS -whole_key_filtering=$WK"; done; done; done) | tee results
```
Showing average ops/sec of "after" vs. "before"
```
-partition_index_and_filters=0 -prefix_size=0 -whole_key_filtering=1
935586 vs. 928176 (+0.79%)
-partition_index_and_filters=0 -prefix_size=8 -whole_key_filtering=0
930171 vs. 926801 (+0.36%)
-partition_index_and_filters=0 -prefix_size=8 -whole_key_filtering=1
910727 vs. 894397 (+1.8%)
-partition_index_and_filters=1 -prefix_size=0 -whole_key_filtering=1
929795 vs. 922007 (+0.84%)
-partition_index_and_filters=1 -prefix_size=8 -whole_key_filtering=0
921924 vs. 917285 (+0.51%)
-partition_index_and_filters=1 -prefix_size=8 -whole_key_filtering=1
903393 vs. 887340 (+1.8%)
```
As one would predict, the most improvement is seen in cases where we have optimized away copying the whole key.
Reviewed By: jowlyzhang
Differential Revision: D61138271
Pulled By: pdillinger
fbshipit-source-id: 427cef0b1465017b45d0a507bfa7720fa20af043
Summary:
`VersionEditHandlerPointInTime` is tracking found files, missing files, intermediate files in order to decide to build a `Version` on negative edge trigger (transition from valid to invalid) without applying the current `VersionEdit`. However, applying `VersionEdit` and check completeness of a `Version` are specialization of `VersionBuilder`. More importantly, when we augment best efforts recovery to recover not just complete point in time Version but also a prefix of seqno for a point in time Version, such checks need to be duplicated in `VersionEditHandlerPointInTime` and `VersionBuilder`.
To avoid this, this refactor move all the file tracking functionality in `VersionEditHandlerPointInTime` into `VersionBuilder`. To continue to let `VersionEditHandlerPIT` do the edge trigger check and build a `Version` before applying the current `VersionEdit`, a suite of APIs to supporting creating a save point and its associated functions are added in `VersionBuilder` to achieve this.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12928
Test Plan: Existing tests
Reviewed By: anand1976
Differential Revision: D61171320
Pulled By: jowlyzhang
fbshipit-source-id: 604f66f8b1e3a3e13da59d8ba357c74e8a366dbc
Summary:
Add a couple of ticker stats for corruption retry count and successful retries. This PR also eliminates an extra read attempt when there's a checksum mismatch in a block read from the prefetch buffer.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12923
Test Plan: Update existing tests
Reviewed By: jowlyzhang
Differential Revision: D61024687
Pulled By: anand1976
fbshipit-source-id: 3a08403580ab244000e0d480b7ee0f5a03d76b06
Summary:
this testcase set syncpoint function which reference this test case heap variable "enable_per_key_placement_" and this sync point function will be triggered by another testcase, so asan will report asan heap use after free error
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12908
Reviewed By: hx235
Differential Revision: D60973363
Pulled By: cbi42
fbshipit-source-id: df4f488f51e7741784d5a92fc0a5fc538c5d5b1a
Summary:
Corruption status returned by `GetFromTable()` could be overwritten here: b6c3495a71/db/version_set.cc (L2614)
This PR fixes this issue by setting `*(s->found_final_value) = true;` in SaveValue. Also makes the handling of the return value of `GetFromTable()` more robust and added asserts in a couple places.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12842
Test Plan: Updated an existing unit test to cover MultiGet. It fails the assertion here before this PR: b6c3495a71/db/version_set.cc (L2601)
Reviewed By: anand1976
Differential Revision: D59498203
Pulled By: cbi42
fbshipit-source-id: 1f071c1b2c5b66fb71264b547a9e670d1cf592f0
Summary:
I was investigating a crash test failure with "Corruption: SST file is ahead of WALs" which I haven't reproduced, but I did reproduce a data loss issue on recovery which I suspect could be the same root problem. The problem is already somewhat known (see https://github.com/facebook/rocksdb/issues/12403 and https://github.com/facebook/rocksdb/issues/12639) where it's only safe to recovery multiple recycled WAL files with trailing old data if the sequence numbers between them are adjacent (to ensure we didn't lose anything in the corrupt/obsolete WAL tail).
However, aside from disableWAL=true, there are features like external file ingestion that can increment the sequence numbers without writing to the WAL. It is simply unsustainable to worry about this kind of feature interaction limiting where we can consume sequence numbers. It is very hard to test and audit as well. For reliable crash recovery of recycled WALs, we need a better way of detecting that we didn't drop data from one WAL to the next.
Until then, let's disable WAL recycling in the crash test, to help stabilize it.
Ideas for follow-up to fix the underlying problem:
(a) With recycling, we could always sync the WAL before opening the next one. HOWEVER, this potentially very large sync could cause a big hiccup in writes (vs. O(1) sized manifest sync).
(a1) The WAL sync could ensure it is truncated to size, or
(a2) By requiring track_and_verify_wals_in_manifest, we could assume that the last synced size in the manifest is the final usable size of the WAL. (It might also be worth avoiding truncating recycled WALs.)
(b) Add a new mechanism to record and verify the final size of a WAL without requiring a sync.
(b1) By requiring track_and_verify_wals_in_manifest, this could be new WAL metadata recorded in the manifest (at the time of switching WALs). Note that new fields of WalMetadata are not forward-compatible, but a new kind of manifest record (next to WalAddition, WalDeletion; e.g. WalCompletion) is IIRC forward-compatible.
(b2) A new kind of WAL header entry (not forward compatible, unfortunately) could record the final size of the previous WAL.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12918
Test Plan: Added disabled reproducer for non-linear data loss on recovery
Reviewed By: hx235
Differential Revision: D60917527
Pulled By: pdillinger
fbshipit-source-id: 3663d79aec81851f5cf41669f84a712bb4563fd7
Summary:
Ahead of a "decoupled" variant of partitioned filters, refactoring this unit test file to make it easier to incorporate that new variant.
* bool test param to new enum class FilterPartitioning
* Some cases of iterating over that bool to new parameterized test
* Combine some common functionality for configuring parameterized options
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12911
Test Plan: no production changes, and no intentional changes to scope or conditions of tests
Differential Revision: D60701287
fbshipit-source-id: 3497e3230e29a4f62c934bcb75693965a2df41d8
Summary:
In normal use cases, meta info like column family's timestamp size is tracked at the transaction layer, so it's not necessary and even detrimental to track such info inside the internal WriteBatch because it may let anti-patterns like bypassing Transaction write APIs and directly write to its internal WriteBatch like this:
9d0a754dc9/storage/rocksdb/ha_rocksdb.cc (L4949-L4950)
Setting this option to true will keep aforementioned use case continue to work before it's refactored out. This option is only for this purpose and it will be gradually deprecated after aforementioned MyRocks use case are refactored.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12864
Test Plan: Added unit tests
Reviewed By: cbi42
Differential Revision: D60194094
Pulled By: jowlyzhang
fbshipit-source-id: 64a98822167e99aa7e4fa2a60085d44a5deaa45c
Summary:
https://github.com/facebook/rocksdb/issues/12891 updated this deletion rate in the test to be much higher, which makes the test flaky. The rate is being intentionally set to very low to maximize the retention of a ".log.trash" file after DB closes. This PR just change it back.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12915
Reviewed By: ltamasi
Differential Revision: D60776312
Pulled By: jowlyzhang
fbshipit-source-id: d193557a042c65816fcc337cceb09905e042e9f6
Summary:
Make `DestroyDB` slowly delete files if it's configured and enabled via `SstFileManager`.
It's currently not available mainly because of DeleteScheduler's logic related to tracked total_size_ and total_trash_size_. These accounting and logic should not be applied to `DestroyDB`. This PR adds a `DeleteUnaccountedDBFile` util for this purpose which deletes files without accounting it. This util also supports assigning a file to a specified trash bucket so that user can later wait for a specific trash bucket to be empty. For `DestroyDB`, files with more than 1 hard links will be deleted immediately.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12891
Test Plan: Added unit tests, existing tests.
Reviewed By: anand1976
Differential Revision: D60300220
Pulled By: jowlyzhang
fbshipit-source-id: 8b18109a177a3a9532f6dc2e40e08310c08ca3c7
Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12910
There is currently a call to `GetBGError()` in `DBImpl::WriteImplWALOnly()` where the DB mutex is (incorrectly) not held, leading to a data race. Technically, we could acquire the mutex here but instead, the patch removes the affected check altogether, since the same check is already performed (in a thread-safe manner) in the subsequent call to `PreprocessWrite()`.
Reviewed By: cbi42
Differential Revision: D60682008
fbshipit-source-id: 54b67975dcf57d67c068cac71e8ada09a1793ec5
Summary:
In leveled compaction, we pick intra-L0 compaction instead of L0->Lbase whenever L0 size is small. When L0 files contain many deletions, it makes more sense to compact then down instead of accumulating tombstones in L0. This PR uses compensated_file_size when computing L0 size for determining intra-L0 compaction. Also scale down the limit on total L0 size further to be more cautious about accumulating data in L0.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12878
Test Plan: updated unit test.
Reviewed By: hx235
Differential Revision: D59932421
Pulled By: cbi42
fbshipit-source-id: 9de973ac51eb7df81b38b8c68110072b1aa06321
Summary:
Crash test encountered this failure:
```file ingestion error: Corruption: properties unsorted under specified IngestExternalFileOptions: move_files: 0, verify_checksums_before_ingest: 1, verify_checksums_readahead_size: 1048576 (Empty string or missing field indicates default option or value is used```
Further inspection showed out of order table properties in an external file created by `SstFileWriter` for ingestion, and the file is likely created like this because it passed the initial checksum check. This change added some assertions to check invariant at the properties creation and collecting side.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12898
Test Plan: Existing tests
Reviewed By: hx235
Differential Revision: D60459817
Pulled By: jowlyzhang
fbshipit-source-id: 91474943d2f9d7795f00b6031c08a13ab91e2470
Summary:
A crash test failure in log sync in DBImpl::WriteToWAL is due to a missed case in https://github.com/facebook/rocksdb/issues/12734. Just need to apply similar logic from DBImpl::SyncWalImpl to check for an already closed WAL (nullptr writer). This is extremely rare because it only comes from failed Sync on a closed WAL.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12899
Test Plan: watch crash test
Reviewed By: cbi42
Differential Revision: D60481652
Pulled By: pdillinger
fbshipit-source-id: 4a176bb6a53dcf077f88344710a110c2f946c386
Summary:
By reusing an object that owns a vector. The vector allocation/sizing was substantial in a CPU profile.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12893
Test Plan: existing tests
Reviewed By: jowlyzhang
Differential Revision: D60405139
Pulled By: pdillinger
fbshipit-source-id: 8bfbc07cd9b4829f2ac9015e90f2b4eba61fd984
Summary:
**Context/Summary:**
We discovered the following false positive in our crash test lately:
(1) PUT() writes k/v to WAL but fails in `ApplyWALToManifest()`. The k/v is in the WAL
(2) Current stress test logic will rollback the expected state of such k/v since PUT() fails
(3) If the DB crashes before recovery finishes and reopens, the WAL will be replayed and the k/v is in the DB while the expected state have been roll-backed.
We decided to leave those expected state to be pending until the loop-write of the same key succeeds.
Bonus: Now that I realized write to manifest can also fail the write which faces the similar problem as https://github.com/facebook/rocksdb/pull/12797, I decided to disable fault injection on user write per thread (instead of globally) when tracing is needed for prefix recovery; some refactory
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12838
Test Plan:
Rehearsal CI
Run below command (varies on sync_fault_injection=1,0 to verify ExpectedState behavior) for a while to ensure crash recovery validation works fine
```
python3 tools/db_crashtest.py --simple blackbox --interval=30 --WAL_size_limit_MB=0 --WAL_ttl_seconds=0 --acquire_snapshot_one_in=10000 --adaptive_readahead=1 --adm_policy=1 --advise_random_on_open=0 --allow_concurrent_memtable_write=0 --allow_data_in_errors=True --allow_fallocate=0 --async_io=0 --auto_readahead_size=0 --avoid_flush_during_recovery=0 --avoid_flush_during_shutdown=0 --avoid_unnecessary_blocking_io=0 --backup_max_size=104857600 --backup_one_in=0 --batch_protection_bytes_per_key=0 --bgerror_resume_retry_interval=1000000 --block_align=1 --block_protection_bytes_per_key=4 --block_size=16384 --bloom_before_level=4 --bloom_bits=56.810257702625165 --bottommost_compression_type=none --bottommost_file_compaction_delay=0 --bytes_per_sync=262144 --cache_index_and_filter_blocks=1 --cache_index_and_filter_blocks_with_high_priority=1 --cache_size=8388608 --cache_type=auto_hyper_clock_cache --charge_compression_dictionary_building_buffer=1 --charge_file_metadata=1 --charge_filter_construction=1 --charge_table_reader=0 --check_multiget_consistency=0 --check_multiget_entity_consistency=1 --checkpoint_one_in=10000 --checksum_type=kxxHash --clear_column_family_one_in=0 --column_families=1 --compact_files_one_in=1000 --compact_range_one_in=1000 --compaction_pri=4 --compaction_readahead_size=1048576 --compaction_ttl=10 --compress_format_version=1 --compressed_secondary_cache_ratio=0.0 --compressed_secondary_cache_size=0 --compression_checksum=0 --compression_max_dict_buffer_bytes=0 --compression_max_dict_bytes=0 --compression_parallel_threads=1 --compression_type=none --compression_use_zstd_dict_trainer=0 --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --daily_offpeak_time_utc=04:00-08:00 --data_block_index_type=1 --db_write_buffer_size=0 --default_temperature=kWarm --default_write_temperature=kCold --delete_obsolete_files_period_micros=30000000 --delpercent=20 --delrangepercent=20 --destroy_db_initially=0 --detect_filter_construct_corruption=0 --disable_file_deletions_one_in=10000 --disable_manual_compaction_one_in=1000000 --disable_wal=0 --dump_malloc_stats=0 --enable_checksum_handoff=1 --enable_compaction_filter=0 --enable_custom_split_merge=0 --enable_do_not_compress_roles=0 --enable_index_compression=1 --enable_memtable_insert_with_hint_prefix_extractor=0 --enable_pipelined_write=0 --enable_sst_partitioner_factory=0 --enable_thread_tracking=0 --enable_write_thread_adaptive_yield=0 --error_recovery_with_no_fault_injection=1 --exclude_wal_from_write_fault_injection=0 --fail_if_options_file_error=1 --fifo_allow_compaction=0 --file_checksum_impl=crc32c --fill_cache=1 --flush_one_in=1000000 --format_version=3 --get_all_column_family_metadata_one_in=1000000 --get_current_wal_file_one_in=0 --get_live_files_apis_one_in=1000000 --get_properties_of_all_tables_one_in=1000000 --get_property_one_in=100000 --get_sorted_wal_files_one_in=0 --hard_pending_compaction_bytes_limit=274877906944 --high_pri_pool_ratio=0.5 --index_block_restart_interval=4 --index_shortening=2 --index_type=0 --ingest_external_file_one_in=0 --initial_auto_readahead_size=16384 --inplace_update_support=0 --iterpercent=10 --key_len_percent_dist=1,30,69 --key_may_exist_one_in=100 --last_level_temperature=kWarm --level_compaction_dynamic_level_bytes=1 --lock_wal_one_in=10000 --log_file_time_to_roll=60 --log_readahead_size=16777216 --long_running_snapshots=1 --low_pri_pool_ratio=0 --lowest_used_cache_tier=0 --manifest_preallocation_size=0 --manual_wal_flush_one_in=0 --mark_for_compaction_one_file_in=10 --max_auto_readahead_size=16384 --max_background_compactions=1 --max_bytes_for_level_base=67108864 --max_key=100000 --max_key_len=3 --max_log_file_size=1048576 --max_manifest_file_size=32768 --max_sequential_skip_in_iterations=1 --max_total_wal_size=0 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=10 --max_write_buffer_size_to_maintain=8388608 --memtable_insert_hint_per_batch=1 --memtable_max_range_deletions=0 --memtable_prefix_bloom_size_ratio=0.01 --memtable_protection_bytes_per_key=1 --memtable_whole_key_filtering=1 --memtablerep=skip_list --metadata_charge_policy=1 --metadata_read_fault_one_in=0 --metadata_write_fault_one_in=8 --min_write_buffer_number_to_merge=1 --mmap_read=1 --mock_direct_io=False --nooverwritepercent=1 --num_file_reads_for_auto_readahead=1 --open_files=-1 --open_metadata_read_fault_one_in=0 --open_metadata_write_fault_one_in=8 --open_read_fault_one_in=0 --open_write_fault_one_in=8 --ops_per_thread=100000000 --optimize_filters_for_hits=1 --optimize_filters_for_memory=1 --optimize_multiget_for_io=1 --paranoid_file_checks=0 --partition_filters=0 --partition_pinning=3 --pause_background_one_in=1000000 --periodic_compaction_seconds=2 --prefix_size=7 --prefixpercent=0 --prepopulate_block_cache=0 --preserve_internal_time_seconds=0 --progress_reports=0 --promote_l0_one_in=0 --read_amp_bytes_per_bit=0 --read_fault_one_in=1000 --readahead_size=524288 --readpercent=10 --recycle_log_file_num=1 --reopen=0 --report_bg_io_stats=0 --reset_stats_one_in=1000000 --sample_for_compression=0 --secondary_cache_fault_one_in=0 --set_options_one_in=0 --skip_stats_update_on_db_open=1 --snapshot_hold_ops=100000 --soft_pending_compaction_bytes_limit=68719476736 --sqfc_name=foo --sqfc_version=0 --sst_file_manager_bytes_per_sec=104857600 --sst_file_manager_bytes_per_truncate=0 --stats_dump_period_sec=10 --stats_history_buffer_size=0 --strict_bytes_per_sync=1 --subcompactions=4 --sync=1 --sync_fault_injection=0 --table_cache_numshardbits=6 --target_file_size_base=16777216 --target_file_size_multiplier=1 --test_batches_snapshots=0 --top_level_index_pinning=2 --uncache_aggressiveness=239 --universal_max_read_amp=-1 --unpartitioned_pinning=1 --use_adaptive_mutex=1 --use_adaptive_mutex_lru=1 --use_attribute_group=0 --use_delta_encoding=0 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_get_entity=0 --use_merge=0 --use_multi_cf_iterator=0 --use_multi_get_entity=0 --use_multiget=0 --use_put_entity_one_in=0 --use_sqfc_for_range_queries=1 --use_timed_put_one_in=0 --use_write_buffer_manager=0 --user_timestamp_size=0 --value_size_mult=32 --verification_only=0 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_compression=0 --verify_db_one_in=100000 --verify_file_checksums_one_in=1000000 --verify_iterator_with_expected_state_one_in=5 --verify_sst_unique_id_in_manifest=1 --wal_bytes_per_sync=0 --wal_compression=none --write_buffer_size=33554432 --write_dbid_to_manifest=0 --write_fault_one_in=8 --writepercent=40
```
Reviewed By: cbi42
Differential Revision: D59377075
Pulled By: hx235
fbshipit-source-id: 91f602fd67e2d339d378cd28b982095fd073dcb6
Summary:
This PR fix `VersionSet`'s `manifest_number_` could be pointing to an invalid number intermediately. This happens when a new manifest roll is attempted but fast failed after loading table handlers and before the new manifest file creation/writing is actually attempted.
In theory, a later manifest roll effort will overthrow this intermediate invalid in memory state. There is on harm when the DB crashes in this invalid state either. But efforts that takes a file snapshot of the DB like backup will incorrectly try to copy a non existing manifest file.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12882
Reviewed By: cbi42
Differential Revision: D60204956
Pulled By: jowlyzhang
fbshipit-source-id: effbdb124b582f879d114988af06ac63867fc549
Summary:
Something I am working on is going to expand usage of `BlockBasedTableBuilder::Rep::last_key`, but the existing code contract for `IndexBuilder::AddIndexEntry` makes that difficult because it modifies its `last_key` parameter to be the separator value recorded in the index, often something between the two boundary keys.
This change primarily changes the contract of that function and related functions to separate function inputs and outputs, without sacrificing efficiency. For efficiency, a reusable scratch string buffer is provided by the caller, which the callee can use (or not) in returning a result Slice. That should yield a performance improvement as we are reusing a buffer for keys rather than copying into a new one each time in the FindShort* functions, without any additional string copies or conditional branches.
Additional improvements in PartitionedIndexBuilder specifically:
* Reduce string copies by eliminating `sub_index_last_key_` and instead tracking the key for the next partition in a placeholder Entry.
* Simplify code and improve code quality by changing `sub_index_builder_` to unique_ptr.
* Eliminate unnecessary NewFlushBlockPolicy call/object.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12867
Test Plan: existing tests, crash test. Will validate performance along with the change this is setting up.
Reviewed By: anand1976
Differential Revision: D59793119
Pulled By: pdillinger
fbshipit-source-id: 556da75cf13b967511f84702b2713d152f536a07
Summary:
**Context/Summary:**
We recently discovered a case where write of the same key right after error recovery of a previous failed write of the same key finishes causes two same WAL entries, violating our assertion. This is because we don't advance seqno on failed write and reuse the same WAL containing the failed write for the new write if the memtable at the time is empty.
This PR reuses the flush path for an empty memtable to switch WAL and update min WAL to keep in error recovery flush
as well as updates the INFO log message for clarity.
```
2024/07/17-15:01:32.271789 327757 (Original Log Time 2024/07/17-15:01:25.942234) [/flush_job.cc:1017] [default] [JOB 2] Level-0 flush table https://github.com/facebook/rocksdb/issues/9: 0 bytes OK It's an empty SST file from a successful flush so won't be kept in the DB
2024/07/17-15:01:32.271798 327757 (Original Log Time 2024/07/17-15:01:32.269954) [/memtable_list.cc:560] [default] Level-0 commit flush result of table https://github.com/facebook/rocksdb/issues/9 started
2024/07/17-15:01:32.271802 327757 (Original Log Time 2024/07/17-15:01:32.271217) [/memtable_list.cc:760] [default] Level-0 commit flush result of table https://github.com/facebook/rocksdb/issues/9: memtable https://github.com/facebook/rocksdb/issues/1 done
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12873
Test Plan:
New UT that failed before this PR with following assertion failure (i.e, duplicate WAL entries) and passes after
```
db_wal_test: db/write_batch.cc:2254: rocksdb::Status rocksdb::{anonymous}::MemTableInserter::PutCFImpl(uint32_t, const rocksdb::Slice&, const rocksdb::Slice&, rocksdb::ValueType, RebuildTxnOp, const ProtectionInfoKVOS64*) [with RebuildTxnOp = rocksdb::{anonymous}::MemTableInserter::PutCF(uint32_t, const rocksdb::Slice&, const rocksdb::Slice&)::<lambda(rocksdb::WriteBatch*, uint32_t, const rocksdb::Slice&, const rocksdb::Slice&)>; uint32_t = unsigned int; rocksdb::ProtectionInfoKVOS64 = rocksdb::ProtectionInfoKVOS<long unsigned int>]: Assertion `seq_per_batch_' failed.
```
Reviewed By: anand1976
Differential Revision: D59884468
Pulled By: hx235
fbshipit-source-id: 5d854b719092552c69727a979f269fb7f6c39756
Summary:
InitInputTableProperties() can open and do IOs and is called under mutex_. This PR removes it from FinalizeInputInfo(). It is now called in CompactionJob::Run() and BuildCompactionJobInfo() (called in NotifyOnCompactionBegin()) without holding mutex_.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12879
Test Plan: existing unit tests. Added assert in GetInputTableProperties() to ensure that input_table_properties_ is initialized whenever it's called.
Reviewed By: hx235
Differential Revision: D59933195
Pulled By: cbi42
fbshipit-source-id: c8089e13af8567fa3ab4b94d9ec384ae98ab2ec8
Summary:
... to enable use cases like using RocksDB to merge sort data for ingestion. A new file ingestion option `IngestExternalFileOptions::allow_db_generated_files` is introduced to allows users to ingest SST files generated by live DBs instead of SstFileWriter. For now this only works if the SST files being ingested have zero as their largest sequence number AND do not overlap with any data in the DB (so we can assign seqno 0 which matches the seqno of all ingested keys).
The feature is marked the option as experimental for now.
Main changes needed to enable this:
- ignore CF id mismatch during ingestion
- ignore the missing external file version table property
Rest of the change is mostly in new unit tests.
A previous attempt is in https://github.com/facebook/rocksdb/issues/5602.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12750
Test Plan: - new unit tests
Reviewed By: ajkr, jowlyzhang
Differential Revision: D58396673
Pulled By: cbi42
fbshipit-source-id: aae513afad7b1ff5d4faa48104df5f384926bf03
Summary:
This PR fixes an assertion failure in `DBImpl::ResumeImpl` - `assert(!versions_->descriptor_log_)`. In `VersionSet`, `descriptor_log_` has a pointer to the current MANIFEST writer. When there's an error updating the manifest, `descriptor_log_` is reset, and the error recovery thread checks `io_status()` in `VersionSet` and attempts to write a new MANIFEST. If another DB manipulation happens at the same time (like external file ingestion, column family manipulation etc), it calls `LogAndApply`, which also attempts to write a new MANIFEST. The assertion in `ResumeImpl` might fail in this case since the other MANIFEST writer may have updated `descriptor_log_`. To prevent the assertion, this fix updates both `io_status_` and `descriptor_log_` while holding the DB mutex.
The other option would have been to simply remove the assert. But I think its important to have it to ensure the invariant that `io_status_` is cleared if the MANIFEST is written successfully, and this fix makes things easier to reason about.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12871
Test Plan: Existing tests and crash test
Reviewed By: hx235
Differential Revision: D59926947
Pulled By: anand1976
fbshipit-source-id: af9ad18da3e29fc62c7ec2e30e0738aa33d4e5f1
Summary:
Basically, the fix in https://github.com/facebook/rocksdb/issues/8137 was incomplete (and I missed it in the review), because if `whole_key_filtering` is false, then `last_prefix_str_` will never be set to non-empty and the fix doesn't work. Also related to https://github.com/facebook/rocksdb/issues/5835.
This is intended as a safe, simple fix that will regress CPU efficiency slightly (for `whole_key_filtering=false` cases, because of extra prefix string copies during flush & compaction). An efficient fix is not possible without some substantial refactoring.
Also in this PR: new test DBBloomFilterTest.FilterNumEntriesCoalesce tests an adjacent code path that was previously untested for its effect of ensuring the number of unique prefixes and keys is tracked properly when both prefixes and whole keys are going into a filter. (Test fails when either of the two code segments checking for duplicates is disabled.) In addition, the same test would fail before the main bug fix here because the code would inappropriately add the empty string to the filter (because of unmodified `last_prefix_str_`).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12872
Test Plan: In addition to DBBloomFilterTest.FilterNumEntriesCoalesce, extended DBBloomFilterTest.SeekForPrevWithPartitionedFilters to cover the broken case. (Mostly whitespace change.)
Reviewed By: jowlyzhang
Differential Revision: D59873793
Pulled By: pdillinger
fbshipit-source-id: 2a7b7f09ca73dc188fb4dab833826ad6da7ebb11
Summary:
Context/Summary: see above, though the impact is small.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12866
Test Plan: exiting UT
Reviewed By: anand1976
Differential Revision: D59782913
Pulled By: hx235
fbshipit-source-id: ec02843645cce49466bde602035d2e61c31965b8
Summary:
The failure of `WriteCurrentStateToManifest()` in `VersionSet::ProcessManifestWrites()` was not handled properly. If it failed, `manifest_io_status` was not updated, leading to `manifest_file_number_` being updated to the newly created manifest even though its bad. This would lead to the bad manifest immediately getting deleted, and also the good manifest (referenced by `CURRENT`) getting deleted by obsolete file deletion because of `manifest_file_number_` not referencing its number.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12865
Reviewed By: hx235
Differential Revision: D59782940
Pulled By: anand1976
fbshipit-source-id: f752fb9a1c23fd3d734616e273613cbac204301b
Summary:
**Context/Summary:**
`*auto_recovery` needs to be set true in order for `OnErrorRecoveryBegin()` to be called before auto-recovery
3db030d7ee/db/event_helpers.cc (L64-L66)
Currently it's set false for auto-recovery. This PR fixes it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12860
Test Plan:
- Manual observation that it is called
- Existing UT
Reviewed By: jowlyzhang
Differential Revision: D59693315
Pulled By: hx235
fbshipit-source-id: 3f428c5b1e9818bb7697fdcd7f245d11378eb14a
Summary:
In follow-up to https://github.com/facebook/rocksdb/issues/12852:
* Use std::copy in place of copy_n for potentially overlapping buffer
* Get rid of troublesome -1 idiom from `pos_at_last_append_` and `pos_at_last_sync_`
* Small improvements to test FaultInjectionFSTest.ReadUnsyncedData
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12861
Test Plan: CI, crash test, etc.
Reviewed By: cbi42
Differential Revision: D59757484
Pulled By: pdillinger
fbshipit-source-id: c6fbdc2e97c959983184925a855cc8b0285fa23f
Summary:
Follow-up to https://github.com/facebook/rocksdb/issues/12729 and others to fix FaultInjectionTestFS handling the case where a live WAL is being appended to and synced while also being copied for checkpoint or backup, up to a known flushed (but not necessarily synced) prefix of the file. It was tricky to structure the code in a way that could handle a tricky race with Sync in another thread (see code comments, thanks Changyu) while maintaining good performance and test-ability.
For more context, see the call to FlushWAL() in DBImpl::GetLiveFilesStorageInfo().
Also, the unit test for https://github.com/facebook/rocksdb/issues/12729 was neutered by https://github.com/facebook/rocksdb/issues/12797, and this re-enables the functionality it is testing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12852
Test Plan:
unit test expanded/updated. Local runs of blackbox_crash_test.
The implementation is structured so that a multi-threaded unit test is not needed to cover at least the code lines, as the race handling is folded into "catch up after returning unsynced and then a sync."
Reviewed By: cbi42
Differential Revision: D59594045
Pulled By: pdillinger
fbshipit-source-id: 94667bb72255e2952586c53bae2c2dd384e85a50
Summary:
Create C API function for iterating over WriteBatch for custom Column Families
Adding function to C API that exposes column family specific methods to iterate over WriteBatch: put_cf, delete_cf and merge_cf. This is required when the one needs to read changes for any non-default column family. Without that functionality it is impossible to iterate over changes in WAL that are relevant to custom column families.
Fixes https://github.com/facebook/rocksdb/issues/12790
Testing:
Added WriteBatch iteration test to "columnfamilies" section of C API unit tests
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12718
Reviewed By: cbi42
Differential Revision: D59483601
Pulled By: ajkr
fbshipit-source-id: b68b900636304528a38620a8c3ad82fdce4b60cb
Summary:
### Summary: Round-Robin pri under leveled compaction allows subcompactions by default is not compatible with PlainTable
```c++
bool Compaction::ShouldFormSubcompactions() const {
if (cfd_ == nullptr) {
return false;
}
// Round-Robin pri under leveled compaction allows subcompactions by default
// and the number of subcompactions can be larger than max_subcompactions_
if (cfd_->ioptions()->compaction_pri == kRoundRobin &&
cfd_->ioptions()->compaction_style == kCompactionStyleLevel) {
return output_level_ > 0;
}
if (max_subcompactions_ <= 1) {
return false;
}
```
PlainTable does not support Subcompaction, including when AdaptiveTable is applied to PlainTable. subcompaction by default will result in the following error in some scenarios.
```c++
void PlainTableIterator::Seek(const Slice& target) {
if (use_prefix_seek_ != !table_->IsTotalOrderMode()) {
// This check is done here instead of NewIterator() to permit creating an
// iterator with total_order_seek = true even if we won't be able to Seek()
// it. This is needed for compaction: it creates iterator with
// total_order_seek = true but usually never does Seek() on it,
// only SeekToFirst().
status_ = Status::InvalidArgument(
"total_order_seek not implemented for PlainTable.");
offset_ = next_offset_ = table_->file_info_.data_end_offset;
return;
}
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12843
Reviewed By: ajkr
Differential Revision: D59433477
Pulled By: cbi42
fbshipit-source-id: fb780ba7f7e8efdfedb7480abf14dd38e0b63677
Summary:
We are seeing a number of crash test failures coming from checkpoint and backup code, likely from WalManager::GetSortedWalFiles -> ... -> WalManager::ReadFirstLine and this code path is not needed, because we don't need to know the sequence numbers of WAL files going into a checkpoint or backup. We can minimize the impact of whatever inconsistency is causing that problem by not relying on it where it's not needed.
Similarly, when we only need a roughly accurate set of current WAL files, we don't need to query all the archived WAL files (and redundantly the live ones again).
So this reduces filesystem queries and DB mutex acquires in creating backups and checkpoints.
Needed follow-up:
Figure out what is causing various failures with an apparent inconsistency where GetSortedWalFiles fails on reading a WAL file. If it's an injected failure, perhaps it's not propagating that injected failure appropriately. It might also be an inconsistency between what the DB knows is flushed and what WalManager reads from the filesystem (which we know is dubious and should be phased out, which this is arguably another step toward). Or completing that phase-out might solve the problem without a full diagnosis.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12831
Test Plan:
existing tests (easily caught when I went too far in initally developing this change)
Update to BackupUsingDirectIO test so that there's a WAL file in what is backed up. (Was relying on some oddity.)
Reviewed By: cbi42
Differential Revision: D59252649
Pulled By: pdillinger
fbshipit-source-id: 7ad4187a1c70caa59a6d6c1c643ef95232b929f5
Summary:
We didn't implement file system prefetch for OS Win. During table open, it uses `FilePrefetchBuffer` instead and only do 1 read instead of 4 in BufferedIO.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12816
Reviewed By: jaykorean
Differential Revision: D59181835
Pulled By: ajkr
fbshipit-source-id: 18b8f0247408cd1a80f289357ede5232ae5a3c66
Summary:
**Context/Summary:**
It seems unreasonable to take the archived log size into account when calculating log size **for flush** in method CreateCheckpoint. If the user sets WAL_ttl_seconds or WAL_size_limit_MB, the argument _log_size_for_flush_ can easily be reached due to the size of the archived dir. As a result, the flush may always be triggered.
**Test**
corverd by ./checkpoint_test
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12680
Reviewed By: jaykorean
Differential Revision: D59097904
Pulled By: ajkr
fbshipit-source-id: 0ed29c1b078d8f40b85288541b008e00dbc517d3
Summary:
the return value for `ErrorHandler::SetBGError(error)` seems to be not well-defined, it can be `bg_error_` (no matter if the `bg_error_` is set to the input error), ok status or [`recovery_error_`](3ee4d5a11a/db/error_handler.cc (L669)) from `StartRecoverFromRetryableBGIOError()`. The `recovery_error_` returned may be an OK status.
We have only a few places that use the return value of `SetBGError()` and they don't need to do so. Using the return value may even be wrong for example in 3ee4d5a11a/db/db_impl/db_impl_write.cc (L2365) where a non-ok `s` could be overwritten to OK. This PR changes SetBGError() to return void and clean up relevant code.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12792
Test Plan: existing unit tests and go over all places where return value of `SetBGError()` is used.
Reviewed By: hx235
Differential Revision: D58904898
Pulled By: cbi42
fbshipit-source-id: d58a20ba5a40e3f35367c6034a32c755088c3653
Summary:
I'd like to get this in so the Rust folks can integrate with their splendid logging/tracing frameworks; will be hugely appreciated. 🙏🏻
The infolog capabilities for C embeddings are quite spartan. LOG files were generated involuntarily until redirection to stderr was added by https://github.com/facebook/rocksdb/issues/12262; still insufficient for apps which cannot tolerate pollution of their stdio and tend to have existing logging frameworks to tie into for that.
Adds a very minimal derive of Logger around a C callback, written in the spirit, useful for FFI interfaces from other languages to integrate infolog.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12537
Reviewed By: ajkr
Differential Revision: D57597766
Pulled By: cbi42
fbshipit-source-id: ec684ce4ddf77a0a6ebbf013a1bacb4ff2e49eb0
Summary:
Fix a race for accessing `bg_error_` after mutex is released. We make some copies before releasing to avoid this.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12803
Reviewed By: cbi42
Differential Revision: D58957557
Pulled By: jowlyzhang
fbshipit-source-id: 3c7369a3b8c8707aebc0044ff98288c898c05cb8
Summary:
**Context/Summary:**
The relevant code logs info of newly created WAL and proceeds to "ConstructFragmentedRangeTombstones()" even when the previous step fails. This PR fixes it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12798
Test Plan: Existing tests
Reviewed By: cbi42
Differential Revision: D58917246
Pulled By: hx235
fbshipit-source-id: f395210d91e50617195cb9a8047cf5d82db0c40e
Summary:
the assertion `assert(!IsFragmentedRangeTombstonesConstructed(false));` assumes ConstructFragmentedRangeTombstones() is called only once for a memtable. This is not true since SwitchMemtable() can be called multiple times on the same live memtable, if a previous attempt fails. So remove the assertion in this PR and simplify relevant code.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12796
Test Plan: the exact condition to trigger manifest write in SwitchMemtable() is complicated. Will monitor crash test to see if there's no more failure.
Reviewed By: hx235
Differential Revision: D58913310
Pulled By: cbi42
fbshipit-source-id: 458bb9eebcf6743e9001186fcb757e4b50e8a5d2
Summary:
I believe this was possible with recyclable logs before recent work like https://github.com/facebook/rocksdb/issues/12734, but this cleans up a couple of possible crashes revealed by the crash test. A WAL with a nullptr file writer (already closed) can persist in `logs_` if a later WAL fails to sync. In case of any WAL sync failures, we don't record WAL syncs to the manifest. Thus, even if a WAL is fully synced and closed, we might need to keep it on the `logs_` list so that we know to record its sync to the manifest if there should be a successful sync next time. (However, I believe that's future-looking because currently any failure in WAL sync is considered non-recoverable.)
I don't believe this was likely enough before recent changes to warrant a release note (if it was possible).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12789
Test Plan: A unit test that would reveal the crashes, now fixed
Reviewed By: cbi42
Differential Revision: D58874154
Pulled By: pdillinger
fbshipit-source-id: bc69407cd9cbcd080af9585d502d4e33dafc3d29
Summary:
**Context:**
We currently have partial error injection:
- DB operation: all read, SST write
- DB open: all read, SST write, all metadata write.
This PR completes the error injection (with some limitations below):
- DB operation & open: all read, all write, all metadata write, all metadata read
**Summary:**
- Inject retryable metadata read, metadata write error concerning directory (e.g, dir sync, ) or file metadata (e.g, name, size, file creation/deletion...)
- Inject retryable errors to all major file types: random access file, sequential file, writable file
- Allow db stress test operations to handle above injected errors gracefully without crashing
- Change all error injection to thread-local implementation for easier disabling and enabling in the same thread. For example, we can control error handling thread to have no error injection. It's also cleaner in code.
- Limitation: compared to before, we now don't have write fault injection for backup/restore CopyOrCreateFiles work threads since they use anonymous background threads as well as read injection for db open bg thread
- Add a new flag to test error recovery without error injection so we can test the path where error recovery actually succeeds
- Some Refactory & fix to db stress test framework (see PR review comments)
- Fix some minor bugs surfaced (see PR review comments)
- Limitation: had to disable backup restore with metadata read/write injection since it surfaces too many testing issues. Will add it back later to focus on surfacing actual code/internal bugs first.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12713
Test Plan:
- Existing UT
- CI with no trivial error failure
Reviewed By: pdillinger
Differential Revision: D58326608
Pulled By: hx235
fbshipit-source-id: 011b5195aaeb6011641ae0a9194f7f2a0e325ad7
Summary:
This PR adds user property collector factory `CompactForTieringCollectorFactory` to support observe SST file and mark it as need compaction for fast tracking data to the proper tier.
A triggering ratio `compaction_trigger_ratio_` can be configured to achieve the following:
1) Setting the ratio to be equal to or smaller than 0 disables this collector
2) Setting the ratio to be within (0, 1] will write the number of observed eligible entries into a user property and marks a file as need-compaction when aforementioned condition is met.
3) Setting the ratio to be higher than 1 can be used to just writes the user table property, and not mark any file as need compaction.
For a column family that does not enable tiering feature, even if an effective configuration is provided, this collector is still disabled. For a file that is already on the last level, this collector is also disabled.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12760
Test Plan: Added unit tests
Reviewed By: pdillinger
Differential Revision: D58734976
Pulled By: jowlyzhang
fbshipit-source-id: 6daab2c4f62b5c6689c3c03e3b3907bbbe6b7a81
Summary:
Fix issue https://github.com/facebook/rocksdb/issues/12687.
A block cache may be shared by multiple column families. Therefore, when getting the aggregated property of the block cache, we need to deduplicate by instances of the block cache, meaning the same instance should only be counted once.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12755
Reviewed By: jowlyzhang
Differential Revision: D58508819
Pulled By: ajkr
fbshipit-source-id: 3b746841d7eac59f900387ec3b8c19dbcd20aae4
Summary:
Instead of completely disallowing `MultiCfIterator` when one or more child iterators will do manual prefix iteration (as suggested in https://github.com/facebook/rocksdb/issues/12770 ), just let `MultiCfIterator` operate as is even when there's a possibility of undefined result from child iterators. If one or more child iterators cause the heap to be empty, just return early and `Valid()` will return false.
It is still possible that heap is not empty when one or more child iterators are returning wrong keys. Basically, MultiCfIterator behaves the same as what we described in https://github.com/facebook/rocksdb/wiki/Prefix-Seek#manual-prefix-iterating - "RocksDB will not return error when it is misused and the iterating result will be undefined."
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12773
Test Plan:
MultiCfIterator added back to the stress test
```
python3 tools/db_crashtest.py blackbox --simple --max_key=25000000 --write_buffer_size=4194304 --use_attribute_group=0 --use_put_entity_one_in=1 --use_multi_get=1 --use_multi_cf_iterator=1 --verify_iterator_with_expected_state_one_in=2
```
Reviewed By: cbi42
Differential Revision: D58612055
Pulled By: jaykorean
fbshipit-source-id: e0dd942bed98382c59d463412dd8f163e6790b93
Summary:
This PR fix a possible manual flush hanging scenario because of its expectation that others will clear out excessive memtables was not met. The root cause is the FlushRequest rescheduling logic is using a stricter criteria for what a write stall is about to happen means than `WaitUntilFlushWouldNotStallWrites` does. Currently, the former thinks a write stall is about to happen when the last memtable is half full, and it will instead reschedule queued FlushRequest and not actually proceed with the flush. While the latter thinks if we already start to use the last memtable, we should wait until some other background flush jobs clear out some memtables before proceed this manual flush.
If we make them use the same criteria, we can guarantee that at any time when`WaitUntilFlushWouldNotStallWrites` is waiting, it's not because the rescheduling logic is holding it back.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12771
Test Plan: Added unit test
Reviewed By: ajkr
Differential Revision: D58603746
Pulled By: jowlyzhang
fbshipit-source-id: 9fa1c87c0175d47a40f584dfb1b497baa576755b
Summary:
When user-defined timestamps in Memtable only feature is enabled, all scheduled flushes go through a check to see if it's eligible to be rescheduled to retain user-defined timestamps. However when the user makes a manual flush request, their intention is for all the in memory data to be persisted into SST files as soon as possible. These two sides have some conflict of interest, the user can implement some workaround like https://github.com/facebook/rocksdb/issues/12631 to explicitly mark which one takes precedence. The implementation for this can be nuanced since the user needs to be aware of all the scenarios that can trigger a manual flush and handle the concurrency well etc.
In this PR, we updated the default behavior to give manual flush precedence when it's requested. The user-defined timestamps rescheduling mechanism is turned off when a manual flush is requested. Likewise, all error recovery triggered flushes skips the rescheduling mechanism too.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12737
Test Plan: Add unit tests
Reviewed By: ajkr
Differential Revision: D58538246
Pulled By: jowlyzhang
fbshipit-source-id: 0b9b3d1af3e8d882f2d6a2406adda19324ba0694
Summary:
POSIX semantics for LinkFile (hard links) allow linking a file
that is still being written two, with both the source and destination
showing any subsequent writes to the source. This may not be practical
semantics for some FileSystem implementations such as remote storage.
They might only link the flushed or sync-ed file contents at time of
LinkFile, or might even have undefined behavior if LinkFile is called on
a file still open for write (not yet "sealed"). This change builds on https://github.com/facebook/rocksdb/issues/12731
to bring more hygiene to our handling of WAL files in Checkpoint.
Specifically, we now Close WAL files as soon as they are either
(a) inactive and fully synced, or (b) inactive and obsolete (so maybe
never fully synced), rather than letting Close() happen in handling
obsolete files (maybe a background thread). This should not be a
performance issue as Close() should be trivial cost relative to other
IO ops, but just in case:
* We don't Close() while holding a mutex, to avoid blocking, and
* The old behavior is available with a new kill switch option
`background_close_inactive_wals`.
Stacked on https://github.com/facebook/rocksdb/issues/12731
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12734
Test Plan:
Extended existing unit test, especially adding a hygiene
check to FaultInjectionTestFS to detect LinkFile() on a file still open
for writes. FaultInjectionTestFS already has relevant tracking data, and
tests can opt out of the new check, as in a smoke test I have left for
the old, deprecated functionality `background_close_inactive_wals=true`.
Also ran lengthy blackbox_crash_test to ensure the hygiene check is OK
with the crash test. (The only place I can find we use LinkFile in
production is Checkpoint.)
Reviewed By: cbi42
Differential Revision: D58295284
Pulled By: pdillinger
fbshipit-source-id: 64d90ed8477e2366c19eaf9c4c5ad60b82cac5c6
Summary:
Crash test showed a potential use-after-free where a file marked as obsolete and eligible for uncache on destruction is destroyed in the VersionSet destructor, which only happens as part of DB shutdown. At that point, the in-memory column families have already been destroyed, so attempting to uncache could use-after-free on stuff like getting the `user_comparator()` from the `internal_comparator()`.
I attempted to make it smarter, but wasn't able to untangle the destruction dependencies in a way that was safe, understandable, and maintainable.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12751
Test Plan:
Reproduced by adding uncache_aggressiveness to an existing (but otherwise unrelated) test. This makes it a fair regression test.
Also added testing to ensure that trivial moves and DB close & reopen are well behaved with uncache_aggressiveness. Specifically, this issue doesn't seem to be because things are uncached inappropriately in those cases.
Reviewed By: ltamasi
Differential Revision: D58390058
Pulled By: pdillinger
fbshipit-source-id: 66ac9cb13bf02638fa80ee5b7218153d8bc7cfd3
Summary:
This option is recommended to be set for production use:
We recommend to set track_and_verify_wals_in_manifest to true
for production
https://github.com/facebook/rocksdb/wiki/Track-WAL-in-MANIFEST
This adds this setting to the C API, so it can be used by other languages.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12749
Reviewed By: ltamasi
Differential Revision: D58382892
Pulled By: ajkr
fbshipit-source-id: 885de4539745a3119b6b2a162ab4fca9fa975283
Summary:
I haven't been able to reproduce the failure, seen in https://github.com/facebook/rocksdb/actions/runs/9420830905/job/25953696902?pr=12734
```
[ RUN ] DBBlockCacheTypeTestInstance/DBBlockCacheTypeTest.Uncache/2
db/db_block_cache_test.cc:1415: Failure
Expected equality of these values:
cache->GetOccupancyCount()
Which is: 37
kBaselineCount + kNumDataBlocks + meta_blocks_per_file
Which is: 15
Google Test trace:
db/db_block_cache_test.cc:1346: ua=10000
db/db_block_cache_test.cc:1344: partitioned=1
db/db_block_cache_test.cc:1418: Failure
...
```
But it's consistent with a SuperVersion reference sticking around beyond the CompactRange, as I can reproduce the result with a dangling Iterator. Like some other tests have had trouble with periodic stats popping up randomly, I suspect that could be the explanation in this case.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12748
Test Plan: Watch for similar future failures
Reviewed By: ltamasi
Differential Revision: D58366031
Pulled By: pdillinger
fbshipit-source-id: b812ca8837b8c8b9cbda1b201d76316d145fa3ec
Summary:
The write_dbid_to_manifest option is documented as "We recommend setting this flag to true". However, there is no way to set this flag from the C API.
Add the following functions to the C API:
* rocksdb_get_db_identity
* rocksdb_options_get_write_dbid_to_manifest
* rocksdb_options_set_write_dbid_to_manifest
Add a test that this option preserves the ID across checkpoints.
c.cc:
* Remove outdated comments about missing C API functions that exist.
* Document that CopyString is intended for binary data and is not NUL terminated.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12736
Reviewed By: ltamasi
Differential Revision: D58202117
Pulled By: ajkr
fbshipit-source-id: 707b110df5c4bd118d65548327428a53a9dc3019
Summary:
Currently, when files become obsolete, the block cache entries associated with them just age out naturally. With pure LRU, this is not too bad, as once you "use" enough cache entries to (re-)fill the cache, you are guranteed to have purged the obsolete entries. However, HyperClockCache is a counting clock cache with a somewhat longer memory, so could be more negatively impacted by previously-hot cache entries becoming obsolete, and taking longer to age out than newer single-hit entries.
Part of the reason we still have this natural aging-out is that there's almost no connection between block cache entries and the file they are associated with. Everything is hashed into the same pool(s) of entries with nothing like a secondary index based on file. Keeping track of such an index could be expensive.
This change adds a new, mutable CF option `uncache_aggressiveness` for erasing obsolete block cache entries. The process can be speculative, lossy, or unproductive because not all potential block cache entries associated with files will be resident in memory, and attempting to remove them all could be wasted CPU time. Rather than a simple on/off switch, `uncache_aggressiveness` basically tells RocksDB how much CPU you're willing to burn trying to purge obsolete block cache entries. When such efforts are not sufficiently productive for a file, we stop and move on.
The option is in ColumnFamilyOptions so that it is dynamically changeable for already-open files, and customizeable by CF.
Note that this block cache removal happens as part of the process of purging obsolete files, which is often in a background thread (depending on `background_purge_on_iterator_cleanup` and `avoid_unnecessary_blocking_io` options) rather than along CPU critical paths.
Notable auxiliary code details:
* Possibly fixing some issues with trivial moves with `only_delete_metadata`: unnecessary TableCache::Evict in that case and missing from the ObsoleteFileInfo move operator. (Not able to reproduce an current failure.)
* Remove suspicious TableCache::Erase() from VersionSet::AddObsoleteBlobFile() (TODO follow-up item)
Marked EXPERIMENTAL until more thorough validation is complete.
Direct stats of this functionality are omitted because they could be misleading. Block cache hit rate is a better indicator of benefit, and CPU profiling a better indicator of cost.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12694
Test Plan:
* Unit tests added, including refactoring an existing test to make better use of parameterized tests.
* Added to crash test.
* Performance, sample command:
```
for I in `seq 1 10`; do for UA in 300; do for CT in lru_cache fixed_hyper_clock_cache auto_hyper_clock_cache; do rm -rf /dev/shm/test3; TEST_TMPDIR=/dev/shm/test3 /usr/bin/time ./db_bench -benchmarks=readwhilewriting -num=13000000 -read_random_exp_range=6 -write_buffer_size=10000000 -bloom_bits=10 -cache_type=$CT -cache_size=390000000 -cache_index_and_filter_blocks=1 -disable_wal=1 -duration=60 -statistics -uncache_aggressiveness=$UA 2>&1 | grep -E 'micros/op|rocksdb.block.cache.data.(hit|miss)|rocksdb.number.keys.(read|written)|maxresident' | awk '/rocksdb.block.cache.data.miss/ { miss = $4 } /rocksdb.block.cache.data.hit/ { hit = $4 } { print } END { print "hit rate = " ((hit * 1.0) / (miss + hit)) }' | tee -a results-$CT-$UA; done; done; done
```
Averaging 10 runs each case, block cache data block hit rates
```
lru_cache
UA=0 -> hit rate = 0.327, ops/s = 87668, user CPU sec = 139.0
UA=300 -> hit rate = 0.336, ops/s = 87960, user CPU sec = 139.0
fixed_hyper_clock_cache
UA=0 -> hit rate = 0.336, ops/s = 100069, user CPU sec = 139.9
UA=300 -> hit rate = 0.343, ops/s = 100104, user CPU sec = 140.2
auto_hyper_clock_cache
UA=0 -> hit rate = 0.336, ops/s = 97580, user CPU sec = 140.5
UA=300 -> hit rate = 0.345, ops/s = 97972, user CPU sec = 139.8
```
Conclusion: up to roughly 1 percentage point of improved block cache hit rate, likely leading to overall improved efficiency (because the foreground CPU cost of cache misses likely outweighs the background CPU cost of erasure, let alone I/O savings).
Reviewed By: ajkr
Differential Revision: D57932442
Pulled By: pdillinger
fbshipit-source-id: 84a243ca5f965f731f346a4853009780a904af6c
Summary:
As titled. Also added the newest user-defined timestamp into the `MemTableInfo`. This can be a useful info in the callback.
Added some unit tests as examples for how users can use two separate approaches to allow manual flush / manual compactions to go through when the user-defined timestamps in memtable only feature is enabled. One approach relies on selectively increase cutoff timestamp in `OnMemtableSeal` callback when it's initiated by a manual flush. Another approach is to increase cutoff timestamp in `OnManualFlushScheduled` callback. The caveats of the approaches are also documented in the unit test.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12631
Reviewed By: ajkr
Differential Revision: D58260528
Pulled By: jowlyzhang
fbshipit-source-id: bf446d7140affdf124744095e0a179fa6e427532
Summary:
**Context/Summary:** a better API design is decided lately so we decided to revert these two changes.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12738
Test Plan: - CI
Reviewed By: ajkr
Differential Revision: D58162165
Pulled By: hx235
fbshipit-source-id: 9bbe4d2fe9fbe39213f4cf137a2d419e6ffb8e16
Summary:
Background: there is one active WAL file but there can be
several more WAL files in various states. Those other WALs are always
in a "flushed" state but could be on the `logs_` list not yet fully
synced. We currently allow any WAL that is not the active WAL to be
hard-linked when creating a Checkpoint, as although it might still be
open for write, we are not appending any more data to it.
The problem is that a created Checkpoint is supposed to be fully synced
on return of that function, and a hard-linked WAL in the state described
above might not be fully synced. (Through some prudence in https://github.com/facebook/rocksdb/issues/10083,
it would synced if using track_and_verify_wals_in_manifest=true.)
The fix is a step toward a long term goal of removing the need to query
the filesystem to determine WAL files and their state. (I consider it
dubious any time we independently read from or query metadata from a
file we have open for writing, as this makes us more susceptible to
FileSystem deficiencies or races.) More specifically:
* Detect which WALs might not be fully synced, according to our DBImpl
metadata, and prevent hard linking those (with `trim_to_size=true`
from `GetLiveFilesStorageInfo()`. And while we're at it, use our known
flushed sizes for those WALs.
* To avoid a race between that and GetSortedWalFiles(), track a maximum
needed WAL number for the Checkpoint/GetLiveFilesStorageInfo.
* Because of the level of consistency provided by those two, we no
longer need to consider syncing as part of the FlushWAL in
GetLiveFilesStorageInfo. (We determine the max WAL number consistent
with the manifest file size, while holding DB mutex. Should make
track_and_verify_wals_in_manifest happy.) This makes the premise of
test PutRaceWithCheckpointTrackedWalSync obsolete (sync point callback
no longer hit) so the test is removed, with crash test as backstop for
related issues. See https://github.com/facebook/rocksdb/issues/10185
Stacked on https://github.com/facebook/rocksdb/issues/12729
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12731
Test Plan:
Expanded an existing test, which now fails before fix.
Also long runs of blackbox_crash_test with amplified checkpoint frequency.
Reviewed By: cbi42
Differential Revision: D58199629
Pulled By: pdillinger
fbshipit-source-id: 376e55f4a2b082cd2adb6408a41209de14422382
Summary:
In places (e.g. GetSortedWals()) RocksDB relies on querying the file size or even reading the contents of files currently open for writing, and as in POSIX semantics, expects to see the flushed size and contents regardless of what has been synced. FaultInjectionTestFS historically did not emulate this behavior, only showing synced data from such read operations. (Different from FaultInjectionTestEnv--sigh.)
This change makes the "proper" behavior the default behavior, at least for GetFileSize and FSSequentialFile. However, this new functionality is disabled in db_stress because of undiagnosed, unresolved issues.
Also removes unused and confusing field `pos_at_last_flush_`
This change is needed to support testing a relevant bug fix (in a follow-up diff). Other suggested follow-up:
* Fix db_stress not to rely on the old behavior, and fix a related FIXME in db_stress_test_base.cc in LockWAL testing.
* Fill in some corner cases in the FileSystem API for reading unsynced data (see new TODO items).
* Consider deprecating and removing Flush() API functions from FileSystem APIs. It is not clear to me that there is a supported scenario in which they do anything but confuse API users and developers. If there is a use for them, it doesn't appear to be tested.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12729
Test Plan: applies to all unit tests successfully, just updating the unit test from https://github.com/facebook/rocksdb/issues/12556 due to relying on the errant behavior. Also added a specific unit test
Reviewed By: hx235
Differential Revision: D58091835
Pulled By: pdillinger
fbshipit-source-id: f47a63b2b000f5875b6293a98577bff663d7fd33
Summary:
When https://github.com/facebook/rocksdb/issues/12343 added support to bulk load external files while column family enables user-defined timestamps, it's a requirement that the external file doesn't overlap with the DB in key ranges. More specifically, the external file should not contain a user key (without timestamp) that already have some entries in the DB.
All the `*Overlap*` functions like `RangeOverlapWithMemtable`, `RangeOverlapWithCompaction` are using `CompareWithoutTimestamp` to check for overlap already. One thing that is missing here is we need to extend the external file's user key boundary for this check to avoid missing the checks for the boundary user keys. For example, with the current way of checking things where `external_file_info.smallest.user_key()` is used as the left boundary, and `external_file_info.largest.user_key()` is used as the right boundary, a file with this entry: (b, 40) can fit into a DB with these two entries: (b, 30), (c, 20).
To avoid this, we extend the user key boundaries used for overlap check, by updating the left boundary with the maximum timestamp and the right boundary with the minimum timestamp.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12735
Test Plan: Added unit test
Reviewed By: ltamasi
Differential Revision: D58152117
Pulled By: jowlyzhang
fbshipit-source-id: 9cba61e7357f6d76ad44c258381c35073ebbf347
Summary:
rocksdb_batched_multi_get_cf has performance improvement than normal multi_get, however it needs a cf_handle arg, so add a C-API to get and destroy the default cf_handle, as many user only use the default cf.
Fixes https://github.com/facebook/rocksdb/issues/12316
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12514
Reviewed By: hx235
Differential Revision: D55922517
Pulled By: ajkr
fbshipit-source-id: c4cc4289f2cfd9efbb8f390a44a9d8d1ed08d9f0
Summary:
This PR adds a `DB::WriteWithCallback` API that does the same things as `DB::Write` while takes an argument `UserWriteCallback` to execute custom callback functions during the write.
We currently support two types of callback functions: `OnWriteEnqueued` and `OnWalWriteFinish`. The former is invoked after the write is enqueued, and the later is invoked after WAL write finishes when applicable.
These callback functions are intended for users to use to improve synchronization between concurrent writes, their execution is on the write's critical path so it will impact the write's latency if not used properly. The documentation for the callback interface mentioned this and suggest user to keep these callback functions' implementation minimum.
Although transaction interfaces' writes doesn't yet allow user to specify such a user write callback argument, the `DBImpl::Write*` type of APIs do not differentiate between regular DB writes or writes coming from the transaction layer when it comes to supporting this `UserWriteCallback`. These callbacks works for all the write modes including: default write mode, Options.two_write_queues, Options.unordered_write, Options.enable_pipelined_write
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12603
Test Plan: Added unit test in ./write_callback_test
Reviewed By: anand1976
Differential Revision: D58044638
Pulled By: jowlyzhang
fbshipit-source-id: 87a84a0221df8f589ec8fc4d74597e72ce97e4cd
Summary:
These functions were very similar and did not make sense for maintaining separately. This is not a pure refactor but I think bringing the behaviors closer together should reduce long term risk of unintentionally divergent behavior. This change is motivated by some forthcoming WAL handling fixes for Checkpoint and Backups.
* Sync() is always used on closed WALs, like the old SyncClosedWals. SyncWithoutFlush() is only used on the active (maybe) WAL. Perhaps SyncWithoutFlush() should be used whenever available, but I don't know which is preferred, as the previous state of the code was inconsistent.
* Syncing the WAL dir is selective based on need, like old SyncWAL, rather than done always like old SyncClosedLogs. This could be a performance improvement that was never applied to SyncClosedLogs but now is. We might still sync the dir more times than necessary in the case of parallel SyncWAL variants, but on a good FileSystem that's probably not too different performance-wise from us implementing something to have threads wait on each other.
Cosmetic changes:
* Rename internal function SyncClosedLogs to SyncClosedWals
* Merging the sync points into the common implementation between the two entry points isn't pretty, but should be fine.
Recommended follow-up:
* Clean up more confusing naming like log_dir_synced_
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12707
Test Plan: existing tests
Reviewed By: anand1976
Differential Revision: D57870856
Pulled By: pdillinger
fbshipit-source-id: 5455fba016d25dd5664fa41b253f18db2ca8919a
Summary:
We tested on icelake server (vcpu=160). The default configuration is allow_concurrent_memtable_write=1, thread number =activate core number. With our optimizations, the improvement can reach up to 184% in fillseq case. op/s is as the performance indicator in db_bench, and the following are performance improvements in some cases in db_bench.
| case name | optimized/original |
|-------------------:|--------------------:|
| fillrandom | 182% |
| fillseq | 184% |
| fillsync | 136% |
| overwrite | 179% |
| randomreplacekeys | 180% |
| randomtransaction | 161% |
| updaterandom | 163% |
| xorupdaterandom | 165% |
With analysis, we find that although the process of writing memtable is processed in parallel, the process of waking up the writers is not processed in parallel, which means that only one writers is responsible for the sequential waking up other writers. The following is our method to optimize this process.
Assume that there are currently n threads in total, we parallelize SetState in LaunchParallelMemTableWriters. To wake up each writer to write its own memtable, the leader writer first wakes up the (n^0.5-1) caller writers, and then those callers and the leader will wake up n/x separately to write to the memtable. This reduces the number for the leader's to SetState n-1 writers to 2*(n^0.5) writers in turn.
A reproduction script:
./db_bench --benchmarks="fillrandom" --threads ${number of all activate vcpu} --seed 1708494134896523 --duration 60
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12545
Reviewed By: ajkr
Differential Revision: D57422827
Pulled By: cbi42
fbshipit-source-id: 94127937c0c61e4241720bd902c82c607b7b2431
Summary:
https://github.com/facebook/rocksdb/issues/12512 added the sanity check for this incompatible combination. However, it does the check during memtable insertion which can turn the DB into read-only mode. This PR moves the check earlier so that this write failure will not turn the DB into read-only mode and affect other DB operations.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12710
Test Plan: * updated unit test `DBRangeDelTest.RowCache` to write to DB after a failed DeleteRange(). The test fails before this PR.
Reviewed By: ajkr
Differential Revision: D57925188
Pulled By: cbi42
fbshipit-source-id: 8bf001bd3fcf05635411ba28bc4a037321942879
Summary:
Add the `--leader_path` option to specify the directory path of the leader for a follower RocksDB instance. This PR also adds a `count` command to the repl shell. While not specific to followers, it is useful for testing purposes.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12682
Reviewed By: jowlyzhang
Differential Revision: D57642296
Pulled By: anand1976
fbshipit-source-id: 53767d496ecadc363ff92cd958b8e15a7bf3b151
Summary:
**Context/Summary:**
the flag --json of manifest_dump in ldb tool has no effect
The bug may be introduced by pr https://github.com/facebook/rocksdb/pull/8378
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12703
Reviewed By: cbi42
Differential Revision: D57848094
Pulled By: ajkr
fbshipit-source-id: 3d1ce65528bf4ce9c53593a7208406ab90e8994b
Summary:
This PR adds UpdateTimestamp API of WriteBatch and WBWI, create WB, WBWI with all options and Iterator Refresh in C API
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10529
Reviewed By: cbi42
Differential Revision: D57826913
Pulled By: ajkr
fbshipit-source-id: d2ec840129f61a1d3a5a12e859728be98ebbad2f
Summary:
This change replaces the use of `std::unique_ptr` with `std::optional` for conditionally constructing a `ReadLock` object. The read lock object was recently introduced in PR https://github.com/facebook/rocksdb/issues/12624. This change makes the code more concise and clarifies that the lock is not meant to be transferred (as `std::unique_ptr` is movable). It also avoids a heap allocation.
There are no functional changes.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12704
Reviewed By: cbi42
Differential Revision: D57848192
Pulled By: ajkr
fbshipit-source-id: da48c77aac33b51ba5dcc238f98fc48ccf234a21
Summary:
These names are confusing with `Logger` etc. so moving to `WalFile` etc.
Other small, related name refactorings.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12695
Test Plan: Left most unit tests using old names as an API compatibility test. Non-test code compiles with deprecated names removed. No functional changes.
Reviewed By: ajkr
Differential Revision: D57747458
Pulled By: pdillinger
fbshipit-source-id: 7b77596b9c20d865d43b9dc66c30c8bd2b3b424f
Summary:
It should be no less than `level0_file_num_compaction_trigger`(which defaults to 4) when set to a positive value. Otherwise DB open will fail.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12701
Test Plan: crash test not failing DB open due to this option value.
Reviewed By: ajkr
Differential Revision: D57825062
Pulled By: cbi42
fbshipit-source-id: 22d8e12aeceb5cef815157845995a8448552e2d2
Summary:
I think the point of the `if (end_of_buffer_offset_ - buffer_.size() == 0)` was to only set `recycled_` when the first record was read. However, the condition was false when reading the first record when the WAL began with a `kSetCompressionType` record because we had already dropped the `kSetCompressionType` record from `buffer_`. To fix this, I used `first_record_read_` instead.
Also, it was pretty confusing to treat the WAL as non-recycled when a recyclable record first appeared in a non-first record. I changed it to return an error if that happens.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12643
Reviewed By: hx235
Differential Revision: D57238099
Pulled By: ajkr
fbshipit-source-id: e20a2a0c9cf0c9510a7b6af463650a05d559239e
Summary:
**Context/Summary:** https://github.com/facebook/rocksdb/pull/12556 `avoid_sync_during_shutdown=false` missed an edge case where `manual_wal_flush == true` so WAL sync will still miss unflushed WAL. This PR fixes it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12684
Test Plan: modified UT to include this case `manual_wal_flush==true`
Reviewed By: cbi42
Differential Revision: D57655861
Pulled By: hx235
fbshipit-source-id: c9f49fe260e8b38b3ea387558432dcd9a3dbec19
Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12689
These should be in `snake_case` (not `camelCase`) per our style guide.
Reviewed By: jowlyzhang
Differential Revision: D57676418
fbshipit-source-id: 82ad6a87d1540f0b29c2f864ca0128287fe95a9e
Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12681
When rebuilding transactions during recovery, `MemtableInserter::PutCFImpl` currently calls `WriteBatchInternal::Put` regardless of value type, which is incorrect for `PutEntity` entries, as well as `TimedPut`s and the blob indexes used by the old BlobDB implementation. The patch fixes the handling of `PutEntity` and returns `NotSupported` for `TimedPut`s and blob indices.
Reviewed By: jaykorean, jowlyzhang
Differential Revision: D57636355
fbshipit-source-id: 833de4e4aa0b42ff6638b72c4181f981d12d0f15
Summary:
We recently noticed that some memtable flushed and file
ingestions could proceed during LockWAL, in violation of its stated
contract. (Note: we aren't 100% sure its actually needed by MySQL, but
we want it to be in a clean state nonetheless.)
Despite earlier skepticism that this could be done safely (https://github.com/facebook/rocksdb/issues/12666), I
found a place to wait to wait for LockWAL to be cleared before allowing
these operations to proceed: WaitForPendingWrites()
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12652
Test Plan:
Added to unit tests. Extended how db_stress validates LockWAL
and re-enabled combination of ingestion and LockWAL in crash test, in
follow-up to https://github.com/facebook/rocksdb/issues/12642
Ran blackbox_crash_test for a long while with relevant features
amplified.
Suggested follow-up: fix FaultInjectionTestFS to report file sizes
consistent with what the user has requested to be flushed.
Reviewed By: jowlyzhang
Differential Revision: D57622142
Pulled By: pdillinger
fbshipit-source-id: aef265fce69465618974b4ec47f4636257c676ce
Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12677
The patch contains two fixes related to printing `PutEntity` records with `ldb dump_wal`:
1) It adds the key to the printout (it was missing earlier).
2) It restores the formatting flags of the output stream after dumping the wide-column structure so that any `hex` flag that might have been set does not affect subsequent printing of e.g. sequence numbers.
Reviewed By: jaykorean, jowlyzhang
Differential Revision: D57591295
fbshipit-source-id: af4e3e219f0082ad39bbdfd26f8c5a57ebb898be
Summary:
This PR implements deletion of obsolete files in a follower RocksDB instance. The follower tails the leader's MANIFEST and creates links to newly added SST files. These links need to be deleted once those files become obsolete in order to reclaim space. There are three cases to be considered -
1. New files added and links created, but the Version could not be installed due to some missing files. Those links need to be preserved so a subsequent catch up attempt can succeed. We insert the next file number in the `VersionSet` to `pending_outputs_` to prevent their deletion.
2. Files deleted from the previous successfully installed `Version`. These are deleted as usual in `PurgeObsoleteFiles`.
3. New files added by a `VersionEdit` and deleted by a subsequent `VersionEdit`, both processed in the same catchup attempt. Links will be created for the new files when verifying a candidate `Version`. Those need to be deleted explicitly as they're never added to `VersionStorageInfo`, and thus not deleted by `PurgeObsoleteFiles`.
Test plan -
New unit tests in `db_follower_test`.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12657
Reviewed By: jowlyzhang
Differential Revision: D57462697
Pulled By: anand1976
fbshipit-source-id: 898f15570638dd4930f839ffd31c560f9cb73916
Summary:
This test is flaky and a recent failure prints the following:
```
[ RUN ] DBTestWithParam/DBTestWithParam.ThreadStatusSingleCompaction/0
thread id: 1842811, thread status:
thread id: 1842803, thread status:
db/db_test.cc:4697: Failure
Expected equality of these values:
op_count
Which is: 0
expected_count
Which is: 1
[ FAILED ] DBTestWithParam/DBTestWithParam.ThreadStatusSingleCompaction/0, where GetParam() = (1, false) (307 ms)
```
Empty thread status implies that operation_type of the threads are all OP_UNKNOWN. From 3ed46e0668/monitoring/thread_status_updater.cc (L197), this can be due to thread_data->operation_type being OP_UNKNOWN or that thread_data->cf_key it not in `cf_info_map_`, potentially due to how cf_key_ is accessed with relaxed memory order. This PR adds some debug print to print the cf_name to check this.
This PR also prints num_running_compaction and lsm state to check if a compaction is indeed running, and removes some not needed options and ensures that exactly 4 L0 files are created.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12661
Test Plan:
- Cannot repro the failure locally: `gtest-parallel --repeat=10000 --workers=200 ./db_test --gtest_filter="*ThreadStatusSingleCompaction*"`
- New failure message will look like:
```
[ RUN ] DBTestWithParam/DBTestWithParam.ThreadStatusSingleCompaction/0
op_count: 1, expected_count 2
thread id: 6104100864, thread status: , cf_name
thread id: 6103527424, thread status: Compaction, cf_name default
running compaction: 1 lsm state: 4
db/db_test.cc:4885: Failure
Value of: match
Actual: false
Expected: true
[ FAILED ] DBTestWithParam/DBTestWithParam.ThreadStatusSingleCompaction/0, where GetParam() = (1, false) (115 ms)
```
Reviewed By: hx235
Differential Revision: D57422755
Pulled By: cbi42
fbshipit-source-id: 635663f26052b20e485dfa06a7c0f1f318ac1099
Summary:
Represent internal kTypeValuePreferredSeqno in the public API as kEntryTimedPut (because it is created by TimedPut, until the entry can be safely converted to a regular value entry in compaction)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12669
Test Plan: for follow-up work actually using it. But putting this in place in the public API gives us more flexibility in rolling out that follow-up work (e.g. as a user extension or patch if needed).
Reviewed By: jowlyzhang
Differential Revision: D57459637
Pulled By: pdillinger
fbshipit-source-id: 160ccf7c4e524ee479558846b2a207d51b8b3d9c
Summary:
`ReadOptions::pin_data` already has the effect of pinning the `Slice` returned by `Iterator::value()` when the value is stored inline (e.g., `kTypeValue`). This PR adds a bit of visibility into that via a new `Iterator` property, "rocksdb.iterator.is-value-pinned", as well as some documentation and tests.
See also: https://github.com/facebook/rocksdb/issues/12658
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12659
Reviewed By: cbi42
Differential Revision: D57391200
Pulled By: ajkr
fbshipit-source-id: 0caa8db27ca1aba86ee2addc3dfd6f0e003d32e2
Summary:
To avoid use-after-free on custom env on ASSERT_WHATEVER failure.
This is motivated by a rare crash seen in DBErrorHandlingFSTest.WALWriteError (VersionSet::GetObsoleteFiles in a SstFileManagerImpl::ClearError thread) and wanting to rule out this being related to that.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12655
Test Plan: manually seeing ASSERT_WHATEVER failures, especially under ASAN
Reviewed By: cbi42
Differential Revision: D57358202
Pulled By: pdillinger
fbshipit-source-id: 4da2a0d73a54380b257e5cc1ab6c666e26b83973
Summary:
If timestamp size record doesn't fit into a block, without padding `Writer::EmitPhysicalRecord` fails on assert (either `assert(block_offset_ + kHeaderSize + n <= kBlockSize);` or `assert(block_offset_ + kRecyclableHeaderSize + n <= kBlockSize)`, depending on whether recycling log files is enabled) in debug build. In release, current block grows beyond 32K, `block_offset_` gets reset on next `AddRecord` and all the subsequent blocks are no longer aligned by block size.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12614
Reviewed By: ltamasi
Differential Revision: D57302140
Pulled By: jowlyzhang
fbshipit-source-id: cacb5cefb7586885e52a8137ae23a95e1aefca2d
Summary:
As titled. A proper fix should probably be failing file ingestion if the DB is in a lock wal state as it promises to "Freezes the logical state of the DB".
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12642
Reviewed By: pdillinger
Differential Revision: D57235869
Pulled By: jowlyzhang
fbshipit-source-id: c70031463842220f865621eb6f53424df27d29e9
Summary:
**Context/Summary:**
Previously `CompactFiles()` used `RangeOverlapWithCompaction()` to check for conflict when sanitizing input files while later used `FilesRangeOverlapWithCompaction()` to assert for no conflict. The latter function checks for more conflict scenarios than the former does, particularly the ones arising from `preclude_last_level_data_seconds > 0` (i.e, compaction can output to second-to-the-last level). So we ran into assertion violation in `CompactFiles()` like below
```
Assertion `output_level == 0 || !FilesRangeOverlapWithCompaction( input_files, output_level, Compaction::EvaluatePenultimateLevel(vstorage, ioptions_, start_level, output_level))' failed.
```
This PR make `CompactFiles()` used `FilesRangeOverlapWithCompaction()` and return Aborted status upon range conflict instead of crashing (during debug build) or proceed incorrectly (during non-debug build). To do so cleanly, I included a refactoring to make `FilesRangeOverlapWithCompaction()` part of `SanitizeAndConvertCompactionInputFiles()`, replacing `RangeOverlapWithCompaction()`.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12628
Test Plan: New UT crashed before the fix and return correct status after the fix.
Reviewed By: cbi42
Differential Revision: D57123536
Pulled By: hx235
fbshipit-source-id: f963a2c9e7ba1a9927a67fcc87f0dce126d3a430
Summary:
Follow-up from https://github.com/facebook/rocksdb/issues/12403
The crash test was periodically failing with the
"disableWAL option is not supported if recycle_log_file_num > 0" failure, despite not setting the disableWAL from the user side.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12639
Test Plan: db_stress reproducer now passes. Added WAL recycling to txn DB unit tests, which is generally more difficult for correctness. Many tests now cover this change and pass.
Reviewed By: anand1976
Differential Revision: D57227617
Pulled By: pdillinger
fbshipit-source-id: db9abefeb505bce624b45bc64009694d2a5baed9
Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12630
The patch cleans up, improves, and brings into sync (to the extent possible without API signature changes) the sanity checks around the `GetEntity` / `MultiGetEntity` family of APIs, including the read-your-own-writes (`WriteBatchWithIndex`) and transaction layers. The checks are centralized in two main sets of entry points, namely in `DB(Impl)` and the "main" `GetEntityFromBatchAndDB` / `MultiGetEntityFromBatchAndDB` overloads in `WriteBatchWithIndex`. This eliminates the need to duplicate the checks in the transaction classes.
Reviewed By: jaykorean
Differential Revision: D57125741
fbshipit-source-id: 4dd059ef644a9b173fbba767538943397e4cc6cd
Summary:
The state of `saved_seq_for_penul_check_` is not correctly maintained with the current flow. It's supposed to store the original sequence number for a `kTypeValuePreferredSeqno` entry for use in the `DecideOutputLevel` function. However, it's not always properly cleared.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12626
Test Plan:
Added unit test that would fail before the fix
./tiered_compaction_test --gtest_filter="*InterleavedTimedPutAndPut*"
Reviewed By: pdillinger
Differential Revision: D57123469
Pulled By: jowlyzhang
fbshipit-source-id: 8d73214b3b6dc152daf19b6bd6ee9063581dc277
Summary:
In `SaveValue()`, the read lock needs to be obtained before `VerifyEntryChecksum()` because the KV checksum verification reads the entire value metadata+data, which is all mutable when `ColumnFamilyOptions::inplace_update_support == true`.
In `MemTable::Update()`, the write lock needs to be obtained before mutating the value metadata (changing the value size) because it can be read concurrently.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12624
Test Plan:
```
$ make COMPILE_WITH_TSAN=1 -j56 db_stress
...
$ python3 tools/db_crashtest.py blackbox --simple --max_key=10 --inplace_update_support=1 --interval=10 --allow_concurrent_memtable_write=0
```
Reviewed By: cbi42
Differential Revision: D57034571
Pulled By: ajkr
fbshipit-source-id: 3dddf881ad87923143acdf6bfec12ce47bb13a48
Summary:
For manual compaction, FIFO compaction will always skip key range overlapping checking with SST files. If CompactRange() is called with CompactionRangeOptions::change_level=true, a CF with FIFO compaction will now return Status::NotSupported.
For file ingestion, we will always ingest into L0. Previously, it's possible to ingest files into non-L0 levels with FIFO compaction.
These changes also help to fix [this](a178d15baf/db/db_impl/db_impl_compaction_flush.cc (L1269)) assertion failure in crash tests.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/12618
Test Plan: added unit tests to verify the new behavior.
Reviewed By: hx235
Differential Revision: D56962401
Pulled By: cbi42
fbshipit-source-id: 19812a1509650b4162b379ca5bee02f2e9d9569d
