mirror of https://github.com/facebook/rocksdb.git
14 Commits
| Author | SHA1 | Message | Date |
|---|---|---|---|
Rulin Huang
|
20777b96cb |
Optimizations in notify-one (#12545)
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
|
|
|
akankshamahajan
|
1856734821 |
Branch cut 9.1.fb (#12476)
Summary: Pull Request resolved: https://github.com/facebook/rocksdb/pull/12476 Reviewed By: jowlyzhang Differential Revision: D55319508 Pulled By: akankshamahajan15 fbshipit-source-id: 2b6db671e027511282775c0fea155335d8e73cc2 |
|
|
Alan Paxton
|
d9a441113e |
JNI get_helper code sharing / multiGet() use efficient batch C++ support (#12344)
Summary: Implement RAII-based helpers for JNIGet() and multiGet() Replace JNI C++ helpers `rocksdb_get_helper, rocksdb_get_helper_direct`, `multi_get_helper`, `multi_get_helper_direct`, `multi_get_helper_release_keys`, `txn_get_helper`, and `txn_multi_get_helper`. The model is to entirely do away with a single helper, instead a number of utility methods allow each separate JNI `Get()` and `MultiGet()` method to organise their parameters efficiently, then call the underlying C++ `db->Get()`, `db->MultiGet()`, `txn->Get()`, or `txn->MultiGet()` method itself, and use further utilities to retrieve results. Roughly speaking: * get keys into C++ form * Call C++ Get() * get results and status into Java form We achieve a useful performance gain as part of this work; by using the updated C++ multiGet we immediately pick up its performance gains (batch improvements to multiGet C++ were previously implemented, but not until now used by Java/JNI). multiGetBB already uses the batched C++ multiGet(), and all other benchmarks show consistent improvement after the changes: ## Before: ``` Benchmark (columnFamilyTestType) (keyCount) (keySize) (multiGetSize) (valueSize) Mode Cnt Score Error Units MultiGetNewBenchmarks.multiGetBB200 no_column_family 10000 1024 100 256 thrpt 25 5315.459 ± 20.465 ops/s MultiGetNewBenchmarks.multiGetBB200 no_column_family 10000 1024 100 1024 thrpt 25 5673.115 ± 78.299 ops/s MultiGetNewBenchmarks.multiGetBB200 no_column_family 10000 1024 100 4096 thrpt 25 2616.860 ± 46.994 ops/s MultiGetNewBenchmarks.multiGetBB200 no_column_family 10000 1024 100 16384 thrpt 25 1700.058 ± 24.034 ops/s MultiGetNewBenchmarks.multiGetBB200 no_column_family 10000 1024 100 65536 thrpt 25 791.171 ± 13.955 ops/s MultiGetNewBenchmarks.multiGetList10 no_column_family 10000 1024 100 256 thrpt 25 6129.929 ± 94.200 ops/s MultiGetNewBenchmarks.multiGetList10 no_column_family 10000 1024 100 1024 thrpt 25 7012.405 ± 97.886 ops/s MultiGetNewBenchmarks.multiGetList10 no_column_family 10000 1024 100 4096 thrpt 25 2799.014 ± 39.352 ops/s MultiGetNewBenchmarks.multiGetList10 no_column_family 10000 1024 100 16384 thrpt 25 1417.205 ± 22.272 ops/s MultiGetNewBenchmarks.multiGetList10 no_column_family 10000 1024 100 65536 thrpt 25 655.594 ± 13.050 ops/s MultiGetNewBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 256 thrpt 25 6147.247 ± 82.711 ops/s MultiGetNewBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 1024 thrpt 25 7004.213 ± 79.251 ops/s MultiGetNewBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 4096 thrpt 25 2715.154 ± 110.017 ops/s MultiGetNewBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 16384 thrpt 25 1408.070 ± 31.714 ops/s MultiGetNewBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 65536 thrpt 25 623.829 ± 57.374 ops/s MultiGetNewBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 256 thrpt 25 6119.243 ± 116.313 ops/s MultiGetNewBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 1024 thrpt 25 6931.873 ± 128.094 ops/s MultiGetNewBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 4096 thrpt 25 2678.253 ± 39.113 ops/s MultiGetNewBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 16384 thrpt 25 1337.384 ± 19.500 ops/s MultiGetNewBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 65536 thrpt 25 625.596 ± 14.525 ops/s ``` ## After: ``` Benchmark (columnFamilyTestType) (keyCount) (keySize) (multiGetSize) (valueSize) Mode Cnt Score Error Units MultiGetBenchmarks.multiGetBB200 no_column_family 10000 1024 100 256 thrpt 25 5191.074 ± 78.250 ops/s MultiGetBenchmarks.multiGetBB200 no_column_family 10000 1024 100 1024 thrpt 25 5378.692 ± 260.682 ops/s MultiGetBenchmarks.multiGetBB200 no_column_family 10000 1024 100 4096 thrpt 25 2590.183 ± 34.844 ops/s MultiGetBenchmarks.multiGetBB200 no_column_family 10000 1024 100 16384 thrpt 25 1634.793 ± 34.022 ops/s MultiGetBenchmarks.multiGetBB200 no_column_family 10000 1024 100 65536 thrpt 25 786.455 ± 8.462 ops/s MultiGetBenchmarks.multiGetBB200 1_column_family 10000 1024 100 256 thrpt 25 5285.055 ± 11.676 ops/s MultiGetBenchmarks.multiGetBB200 1_column_family 10000 1024 100 1024 thrpt 25 5586.758 ± 213.008 ops/s MultiGetBenchmarks.multiGetBB200 1_column_family 10000 1024 100 4096 thrpt 25 2527.172 ± 17.106 ops/s MultiGetBenchmarks.multiGetBB200 1_column_family 10000 1024 100 16384 thrpt 25 1819.547 ± 12.958 ops/s MultiGetBenchmarks.multiGetBB200 1_column_family 10000 1024 100 65536 thrpt 25 803.861 ± 9.963 ops/s MultiGetBenchmarks.multiGetBB200 20_column_families 10000 1024 100 256 thrpt 25 5253.793 ± 28.020 ops/s MultiGetBenchmarks.multiGetBB200 20_column_families 10000 1024 100 1024 thrpt 25 5705.591 ± 20.556 ops/s MultiGetBenchmarks.multiGetBB200 20_column_families 10000 1024 100 4096 thrpt 25 2523.377 ± 15.415 ops/s MultiGetBenchmarks.multiGetBB200 20_column_families 10000 1024 100 16384 thrpt 25 1815.344 ± 11.309 ops/s MultiGetBenchmarks.multiGetBB200 20_column_families 10000 1024 100 65536 thrpt 25 820.792 ± 3.192 ops/s MultiGetBenchmarks.multiGetBB200 100_column_families 10000 1024 100 256 thrpt 25 5262.184 ± 20.477 ops/s MultiGetBenchmarks.multiGetBB200 100_column_families 10000 1024 100 1024 thrpt 25 5706.959 ± 23.123 ops/s MultiGetBenchmarks.multiGetBB200 100_column_families 10000 1024 100 4096 thrpt 25 2520.362 ± 9.170 ops/s MultiGetBenchmarks.multiGetBB200 100_column_families 10000 1024 100 16384 thrpt 25 1789.185 ± 14.239 ops/s MultiGetBenchmarks.multiGetBB200 100_column_families 10000 1024 100 65536 thrpt 25 818.401 ± 12.132 ops/s MultiGetBenchmarks.multiGetList10 no_column_family 10000 1024 100 256 thrpt 25 6978.310 ± 14.084 ops/s MultiGetBenchmarks.multiGetList10 no_column_family 10000 1024 100 1024 thrpt 25 7664.242 ± 22.304 ops/s MultiGetBenchmarks.multiGetList10 no_column_family 10000 1024 100 4096 thrpt 25 2881.778 ± 81.054 ops/s MultiGetBenchmarks.multiGetList10 no_column_family 10000 1024 100 16384 thrpt 25 1599.826 ± 7.190 ops/s MultiGetBenchmarks.multiGetList10 no_column_family 10000 1024 100 65536 thrpt 25 737.520 ± 6.809 ops/s MultiGetBenchmarks.multiGetList10 1_column_family 10000 1024 100 256 thrpt 25 6974.376 ± 10.716 ops/s MultiGetBenchmarks.multiGetList10 1_column_family 10000 1024 100 1024 thrpt 25 7637.440 ± 45.877 ops/s MultiGetBenchmarks.multiGetList10 1_column_family 10000 1024 100 4096 thrpt 25 2820.472 ± 42.231 ops/s MultiGetBenchmarks.multiGetList10 1_column_family 10000 1024 100 16384 thrpt 25 1716.663 ± 8.527 ops/s MultiGetBenchmarks.multiGetList10 1_column_family 10000 1024 100 65536 thrpt 25 755.848 ± 7.514 ops/s MultiGetBenchmarks.multiGetList10 20_column_families 10000 1024 100 256 thrpt 25 6943.651 ± 20.040 ops/s MultiGetBenchmarks.multiGetList10 20_column_families 10000 1024 100 1024 thrpt 25 7679.415 ± 9.114 ops/s MultiGetBenchmarks.multiGetList10 20_column_families 10000 1024 100 4096 thrpt 25 2844.564 ± 13.388 ops/s MultiGetBenchmarks.multiGetList10 20_column_families 10000 1024 100 16384 thrpt 25 1729.545 ± 5.983 ops/s MultiGetBenchmarks.multiGetList10 20_column_families 10000 1024 100 65536 thrpt 25 783.218 ± 1.530 ops/s MultiGetBenchmarks.multiGetList10 100_column_families 10000 1024 100 256 thrpt 25 6944.276 ± 29.995 ops/s MultiGetBenchmarks.multiGetList10 100_column_families 10000 1024 100 1024 thrpt 25 7670.301 ± 8.986 ops/s MultiGetBenchmarks.multiGetList10 100_column_families 10000 1024 100 4096 thrpt 25 2839.828 ± 12.421 ops/s MultiGetBenchmarks.multiGetList10 100_column_families 10000 1024 100 16384 thrpt 25 1730.005 ± 9.209 ops/s MultiGetBenchmarks.multiGetList10 100_column_families 10000 1024 100 65536 thrpt 25 787.096 ± 1.977 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 256 thrpt 25 6896.944 ± 21.530 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 1024 thrpt 25 7622.407 ± 12.824 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 4096 thrpt 25 2927.538 ± 19.792 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 16384 thrpt 25 1598.041 ± 4.312 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 no_column_family 10000 1024 100 65536 thrpt 25 744.564 ± 9.236 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 1_column_family 10000 1024 100 256 thrpt 25 6853.760 ± 78.041 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 1_column_family 10000 1024 100 1024 thrpt 25 7360.917 ± 355.365 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 1_column_family 10000 1024 100 4096 thrpt 25 2848.774 ± 13.409 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 1_column_family 10000 1024 100 16384 thrpt 25 1727.688 ± 3.329 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 1_column_family 10000 1024 100 65536 thrpt 25 776.088 ± 7.517 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 20_column_families 10000 1024 100 256 thrpt 25 6910.339 ± 14.366 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 20_column_families 10000 1024 100 1024 thrpt 25 7633.660 ± 10.830 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 20_column_families 10000 1024 100 4096 thrpt 25 2787.799 ± 81.775 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 20_column_families 10000 1024 100 16384 thrpt 25 1726.517 ± 6.830 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 20_column_families 10000 1024 100 65536 thrpt 25 787.597 ± 3.362 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 100_column_families 10000 1024 100 256 thrpt 25 6922.445 ± 10.493 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 100_column_families 10000 1024 100 1024 thrpt 25 7604.710 ± 48.043 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 100_column_families 10000 1024 100 4096 thrpt 25 2848.788 ± 15.783 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 100_column_families 10000 1024 100 16384 thrpt 25 1730.837 ± 6.497 ops/s MultiGetBenchmarks.multiGetListExplicitCF20 100_column_families 10000 1024 100 65536 thrpt 25 794.557 ± 1.869 ops/s MultiGetBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 256 thrpt 25 6918.716 ± 15.766 ops/s MultiGetBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 1024 thrpt 25 7626.692 ± 9.394 ops/s MultiGetBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 4096 thrpt 25 2871.382 ± 72.155 ops/s MultiGetBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 16384 thrpt 25 1598.786 ± 4.819 ops/s MultiGetBenchmarks.multiGetListRandomCF30 no_column_family 10000 1024 100 65536 thrpt 25 748.469 ± 7.234 ops/s MultiGetBenchmarks.multiGetListRandomCF30 1_column_family 10000 1024 100 256 thrpt 25 6922.666 ± 17.131 ops/s MultiGetBenchmarks.multiGetListRandomCF30 1_column_family 10000 1024 100 1024 thrpt 25 7623.890 ± 8.805 ops/s MultiGetBenchmarks.multiGetListRandomCF30 1_column_family 10000 1024 100 4096 thrpt 25 2850.698 ± 18.004 ops/s MultiGetBenchmarks.multiGetListRandomCF30 1_column_family 10000 1024 100 16384 thrpt 25 1727.623 ± 4.868 ops/s MultiGetBenchmarks.multiGetListRandomCF30 1_column_family 10000 1024 100 65536 thrpt 25 774.534 ± 10.025 ops/s MultiGetBenchmarks.multiGetListRandomCF30 20_column_families 10000 1024 100 256 thrpt 25 5486.251 ± 13.582 ops/s MultiGetBenchmarks.multiGetListRandomCF30 20_column_families 10000 1024 100 1024 thrpt 25 4920.656 ± 44.557 ops/s MultiGetBenchmarks.multiGetListRandomCF30 20_column_families 10000 1024 100 4096 thrpt 25 3922.913 ± 25.686 ops/s MultiGetBenchmarks.multiGetListRandomCF30 20_column_families 10000 1024 100 16384 thrpt 25 2873.106 ± 4.336 ops/s MultiGetBenchmarks.multiGetListRandomCF30 20_column_families 10000 1024 100 65536 thrpt 25 802.404 ± 8.967 ops/s MultiGetBenchmarks.multiGetListRandomCF30 100_column_families 10000 1024 100 256 thrpt 25 4817.996 ± 18.042 ops/s MultiGetBenchmarks.multiGetListRandomCF30 100_column_families 10000 1024 100 1024 thrpt 25 4243.922 ± 13.929 ops/s MultiGetBenchmarks.multiGetListRandomCF30 100_column_families 10000 1024 100 4096 thrpt 25 3175.998 ± 7.773 ops/s MultiGetBenchmarks.multiGetListRandomCF30 100_column_families 10000 1024 100 16384 thrpt 25 2321.990 ± 12.501 ops/s MultiGetBenchmarks.multiGetListRandomCF30 100_column_families 10000 1024 100 65536 thrpt 25 1753.028 ± 7.130 ops/s ``` Closes https://github.com/facebook/rocksdb/issues/11518 Pull Request resolved: https://github.com/facebook/rocksdb/pull/12344 Reviewed By: cbi42 Differential Revision: D54809714 Pulled By: pdillinger fbshipit-source-id: bee3b949720abac073bce043b59ce976a11e99eb |
|
|
Levi Tamasi
|
81765866c4 |
Update HISTORY/version/format compatibility script for the 8.10 release (#12154)
Summary: Pull Request resolved: https://github.com/facebook/rocksdb/pull/12154 Reviewed By: jaykorean, akankshamahajan15 Differential Revision: D52216271 Pulled By: ltamasi fbshipit-source-id: 13bab72802eeec8f6e3544be9ebcd7f725a64d2e |
|
|
Alan Paxton
|
5a063ecd34 |
Java API consistency between RocksDB.put() , .merge() and Transaction.put() , .merge() (#11019)
Summary:
### Implement new Java API get()/put()/merge() methods, and transactional variants.
The Java API methods are very inconsistent in terms of how they pass parameters (byte[], ByteBuffer), and what variants and defaulted parameters they support. We try to bring some consistency to this.
* All APIs should support calls with ByteBuffer parameters.
* Similar methods (RocksDB.get() vs Transaction.get()) should support as similar as possible sets of parameters for predictability.
* get()-like methods should provide variants where the caller supplies the target buffer, for the sake of efficiency. Allocation costs in Java can be significant when large buffers are repeatedly allocated and freed.
### API Additions
1. RockDB.get implement indirect ByteBuffers. Added indirect ByteBuffers and supporting native methods for get().
2. RocksDB.Iterator implement missing (byte[], offset, length) variants for key() and value() parameters.
3. Transaction.get() implement missing methods, based on RocksDB.get. Added ByteBuffer.get with and without column family. Added byte[]-as-target get.
4. Transaction.iterator() implement a getIterator() which defaults ReadOptions; as per RocksDB.iterator(). Rationalize support API for this and RocksDB.iterator()
5. RocksDB.merge implement ByteBuffer methods; both direct and indirect buffers. Shadow the methods of RocksDB.put; RocksDB.put only offers ByteBuffer API with explicit WriteOptions. Duplicated this with RocksDB.merge
6. Transaction.merge implement methods as per RocksDB.merge methods. Transaction is already constructed with WriteOptions, so no explicit WriteOptions methods required.
7. Transaction.mergeUntracked implement the same API methods as Transaction.merge except the ones that use assumeTracked, because that’s not a feature of merge untracked.
### Support Changes (C++)
The current JNI code in C++ supports multiple variants of methods through a number of helper functions. There are numerous TODO suggestions in the code proposing that the helpers be re-factored/shared.
We have taken a different approach for the new methods; we have created wrapper classes `JDirectBufferSlice`, `JDirectBufferPinnableSlice`, `JByteArraySlice` and `JByteArrayPinnableSlice` RAII classes which construct slices from JNI parameters and can then be passed directly to RocksDB methods. For instance, the `Java_org_rocksdb_Transaction_getDirect` method is implemented like this:
```
try {
ROCKSDB_NAMESPACE::JDirectBufferSlice key(env, jkey_bb, jkey_off,
jkey_part_len);
ROCKSDB_NAMESPACE::JDirectBufferPinnableSlice value(env, jval_bb, jval_off,
jval_part_len);
ROCKSDB_NAMESPACE::KVException::ThrowOnError(
env, txn->Get(*read_options, column_family_handle, key.slice(),
&value.pinnable_slice()));
return value.Fetch();
} catch (const ROCKSDB_NAMESPACE::KVException& e) {
return e.Code();
}
```
Notice the try/catch mechanism with the `KVException` class, which combined with RAII and the wrapper classes means that there is no ad-hoc cleanup necessary in the JNI methods.
We propose to extend this mechanism to existing JNI methods as further work.
### Support Changes (Java)
Where there are multiple parameter-variant versions of the same method, we use fewer or just one supporting native method for all of them. This makes maintenance a bit easier and reduces the opportunity for coding errors mixing up (untyped) object handles.
In order to support this efficiently, some classes need to have default values for column families and read options added and cached so that they are not re-constructed on every method call.
This PR closes https://github.com/facebook/rocksdb/issues/9776
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11019
Reviewed By: ajkr
Differential Revision: D52039446
Pulled By: jowlyzhang
fbshipit-source-id: 45d0140a4887e42134d2e56520e9b8efbd349660
|
|
|
Hui Xiao
|
ab15d33566 |
Update history, version and format testing for 8.8 (#12004)
Summary: Pull Request resolved: https://github.com/facebook/rocksdb/pull/12004 Reviewed By: cbi42 Differential Revision: D50586984 Pulled By: hx235 fbshipit-source-id: 1480a8c2757340ebf83510557104aaa0e437b3ae |
|
|
Peter Dillinger
|
141b872bd4 |
Improve efficiency of create_missing_column_families, light refactor (#11920)
Summary: In preparing some seqno_to_time_mapping improvements, I found that some of the wrap-up work for creating column families was unnecessarily repeated in the case of DB::Open with create_missing_column_families. This change fixes that (`CreateColumnFamily()` -> `CreateColumnFamilyImpl()` in `DBImpl::Open()`), motivated by avoiding repeated calls to `RegisterRecordSeqnoTimeWorker()` but with the side benefit of avoiding repeated calls to `WriteOptionsFile()` for each CF. Also in this change: * Add a `Status::UpdateIfOk()` function for combining statuses in a common pattern * Rename `max_time_duration` -> `min_preserve_seconds` (include units as much as possible) * Improved comments in several places Pull Request resolved: https://github.com/facebook/rocksdb/pull/11920 Test Plan: tests added / updated Reviewed By: jaykorean Differential Revision: D49919147 Pulled By: pdillinger fbshipit-source-id: 3d0318c1d070c842c5331da0a5b415caedc104f1 |
|
|
Changyu Bi
|
49da91ec09 |
Update files for version 8.8 (#11878)
Summary: Pull Request resolved: https://github.com/facebook/rocksdb/pull/11878 Reviewed By: ajkr Differential Revision: D49568389 Pulled By: cbi42 fbshipit-source-id: b2022735799be9b5e81e03dfb418f8b104632ecf |
|
|
akankshamahajan
|
3d67b5e8e5 |
Lookup ahead in block cache ahead to tune Readaheadsize (#11860)
Summary: Implement block cache lookup to determine readahead_size during scans. It's enabled if auto_readahead_size, block_cache and iterate_upper_bound - all three are set. Design - 1. Whenever there is a cache miss and FilePrefetchBuffer is called, a callback is made to determine readahead_size for that prefetching. 2. The callback iterates over index and do block cache lookup for each data block handle until existing readahead_size is reached. Then It removes the cache hit data blocks from end to calculate optimized readahead_size. 3. Since index_iter_ is moved, it stores block handles in a queue, and use that queue to get block handle instead of doing index_iter_->Next(). 4. This is for Sync scans. Async scans support is in progress. NOTE: The issue right now is after Seek and Next, if Prev is called, there is no way to do Prev operation. index_iter_ is already pointing to a different block. So it returns "Not supported" in that case with error message - "auto tuning of readahead size is not supported with Prev op" Pull Request resolved: https://github.com/facebook/rocksdb/pull/11860 Test Plan: - Added new unit test - crash_tests - Running scans locally to check for any regression Reviewed By: anand1976 Differential Revision: D49548118 Pulled By: akankshamahajan15 fbshipit-source-id: f1aee409a71b4ad9e5bf3610f43edf30c6630c78 |
|
|
akankshamahajan
|
5b5b011cdd |
Avoid double block cache lookup during Seek with async_io option (#11616)
Summary: With the async_io option, the Seek happens in 2 phases. Phase 1 starts an asynchronous read on a block cache miss, and phase 2 waits for it to complete and finishes the seek. In both phases, BlockBasedTable::NewDataBlockIterator is called, which tries to lookup the block cache for the data block first before looking in the prefetch buffer. It's optimized by doing the block cache lookup only in the first phase and save some CPU. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11616 Test Plan: Added unit test Reviewed By: jaykorean Differential Revision: D47477887 Pulled By: akankshamahajan15 fbshipit-source-id: 0355e0a68fc0ea2eb92340ae42735afcdbcbfd79 |
|
|
Andrew Kryczka
|
cf95821fb6 |
Update for 8.5.fb branch cut (#11642)
Summary: Updated the main branch for the 8.5.fb branch cut. Also made unreleased_history/release.sh backdate to the last commit instead of the current date in case the release manager is a laggard like myself. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11642 Reviewed By: cbi42 Differential Revision: D47783574 Pulled By: ajkr fbshipit-source-id: 4e2a80f5ccd542dc7dd0d22dfd7e59cb136325a1 |
|
|
akankshamahajan
|
749b179c04 |
Remove reallocation of AlignedBuffer in direct_io sync reads if already aligned (#11600)
Summary: Remove reallocation of AlignedBuffer in direct_io sync reads in RandomAccessFileReader::Read if buffer passed is already aligned. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11600 Test Plan: Setup: `TEST_TMPDIR=./tmp-db/ ./db_bench -benchmarks=filluniquerandom -disable_auto_compactions=true -target_file_size_base=1048576 -write_buffer_size=1048576 -compression_type=none` Benchmark: `TEST_TMPDIR=./tmp-db/ perf record ./db_bench --cache_size=8388608 --use_existing_db=true --disable_auto_compactions=true --benchmarks=seekrandom --use_direct_reads=true -use_direct_io_for_flush_and_compaction=true -reads=1000 -seek_nexts=1 -max_auto_readahead_size=131072 -initial_auto_readahead_size=16384 -adaptive_readahead=true -num_file_reads_for_auto_readahead=0` Perf profile- Before: ``` 8.73% db_bench libc.so.6 [.] __memmove_evex_unaligned_erms 3.34% db_bench [kernel.vmlinux] [k] filemap_get_read_batch ``` After: ``` 2.50% db_bench [kernel.vmlinux] [k] filemap_get_read_batch 2.29% db_bench libc.so.6 [.] __memmove_evex_unaligned_erms ``` `make crash_test -j `with direct_io enabled completed succesfully locally. Ran few benchmarks with direct_io from seek_nexts varying between 912 to 327680 and different readahead_size parameters and it showed no regression so far. Reviewed By: ajkr Differential Revision: D47478598 Pulled By: akankshamahajan15 fbshipit-source-id: 6a48e21cb34696f5d09c22a6311a3a1cb5f9cf33 |
|
|
Peter Dillinger
|
b1b6f87fbe |
Some small improvements to HyperClockCache (#11601)
Summary: Stacked on https://github.com/facebook/rocksdb/issues/11572 * Minimize use of std::function and lambdas to minimize chances of compiler heap-allocating closures (unnecessary stress on allocator). It appears that converting FindSlot to a template enables inlining the lambda parameters, avoiding heap allocations. * Clean up some logic with FindSlot (FIXMEs from https://github.com/facebook/rocksdb/issues/11572) * Fix handling of rare case of probing all slots, with new unit test. (Previously Insert would not roll back displacements in that case, which would kill performance if it were to happen.) * Add an -early_exit option to cache_bench for gathering memory stats before deallocation. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11601 Test Plan: unit test added for probing all slots ## Seeing heap allocations Run `MALLOC_CONF="stats_print:true" ./cache_bench -cache_type=hyper_clock_cache` before https://github.com/facebook/rocksdb/issues/11572 vs. after this change. Before, we see this in the interesting bin statistics: ``` size nrequests ---- --------- 32 578460 64 24340 8192 578460 ``` And after: ``` size nrequests ---- --------- 32 (insignificant) 64 24370 8192 579130 ``` ## Performance test Build with `make USE_CLANG=1 PORTABLE=0 DEBUG_LEVEL=0 -j32 cache_bench` Run `./cache_bench -cache_type=hyper_clock_cache -ops_per_thread=5000000` in before and after configurations, simultaneously: ``` Before: Complete in 33.244 s; Rough parallel ops/sec = 2406442 After: Complete in 32.773 s; Rough parallel ops/sec = 2441019 ``` Reviewed By: jowlyzhang Differential Revision: D47375092 Pulled By: pdillinger fbshipit-source-id: 46f0f57257ddb374290a0a38c651764ea60ba410 |
|
|
Peter Dillinger
|
7a9b264f36 |
Some fixes to unreleased_history/ (#11504)
Summary: * Add a "Performance Improvements" section * Add required copyright headers Pull Request resolved: https://github.com/facebook/rocksdb/pull/11504 Test Plan: manual Reviewed By: hx235 Differential Revision: D46405128 Pulled By: pdillinger fbshipit-source-id: 4f878dfd0170d381d3051a44c13479c860e812c0 |