Summary:
Improve the iterators performance when the user explicitly sets the readahead size via `ReadOptions.readahead_size`.
1. Stop creating new table readers when the user explicitly sets readahead size.
2. Make use of an internal buffer based on `FilePrefetchBuffer` instead of using `ReadaheadRandomAccessFileReader`, to handle the user readahead requests (for both buffered and direct io cases).
3. Add `readahead_size` to db_bench.
**Benchmarks:**
https://gist.github.com/sagar0/53693edc320a18abeaeca94ca32f5737
For 1 MB readahead, Buffered IO performance improves by 28% and Direct IO performance improves by 50%.
For 512KB readahead, Buffered IO performance improves by 30% and Direct IO performance improves by 67%.
**Test Plan:**
Updated `DBIteratorTest.ReadAhead` test to make sure that:
- no new table readers are created for iterators on setting ReadOptions.readahead_size
- At least "readahead" number of bytes are actually getting read on each iterator read.
TODO later:
- Use similar logic for compactions as well.
- This ties in nicely with #4052 and paves the way for removing ReadaheadRandomAcessFile later.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5246
Differential Revision: D15107946
Pulled By: sagar0
fbshipit-source-id: 2c1149729ca7d779e4e8b7710ba6f4e8cbfd3bea
Summary:
In long scans, virtual function calls of Next(), Valid(), key() and value() are not trivial. By introducing NextAndGetResult(), Some of the Next(), Valid() and key() calls are consolidated into one virtual function call to reduce CPU.
Also did some inline tricks and add some "final" randomly in some functions. Even without the "final" annotation, most Next() calls are inlined with -O3, but sometimes with a final it is inlined by O2 too. It doesn't hurt to add those final annotations.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5197
Differential Revision: D14945977
Pulled By: siying
fbshipit-source-id: 7003969f9a5f1d5717f0bda503b91d19ba75ed88
Summary:
This is second attempt for #5101. Original commit message:
`BlockBasedTableIterator` avoid reading next block on `Next()` if it detects the iterator will be out of bound, by checking against index key. The optimization was added in #2239, and by the time it only check the bound per block. It seems later change make it a per-key check, which introduce unnecessary key comparisons.
This patch come with two fixes:
Fix 1: To optimize checking for bounds, we need comparing the bounds with index key as well. However BlockBasedTableIterator doesn't know whether its index iterator is internally using user keys or internal keys. The patch fixes that by extending InternalIterator with a user_key() function that is overridden by In IndexBlockIter.
Fix 2: In #5101 we return `IsOutOfBound()=true` when block index key is out of bound. But the index key can be larger than smallest key of the next file on the level. That file can be within upper bound and should not be filtered out.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5142
Differential Revision: D14907113
Pulled By: siying
fbshipit-source-id: ac95775c5b4e7b700f76ab43e39f45402c98fbfb
Summary:
This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching.
Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to -
1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch()
2. Bloom filter cachelines can be prefetched, hiding the cache miss latency
The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress.
Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32).
Batch Sizes
1 | 2 | 4 | 8 | 16 | 32
Random pattern (Stride length 0)
4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get
4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching)
4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching)
Good locality (Stride length 16)
4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753
4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781
4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135
Good locality (Stride length 256)
4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232
4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268
4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62
Medium locality (Stride length 4096)
4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555
4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465
4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891
dbbench command used (on a DB with 4 levels, 12 million keys)-
TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011
Differential Revision: D14348703
Pulled By: anand1976
fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
Summary:
This reverts commit f29dc1b906.
In BlockBasedTableIterator, index_iter_->key() is sometimes a user key, so it is wrong to call ExtractUserKey() against it. This is a bug introduced by #5101.
Temporarily revert the diff to keep the branch clean.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5132
Differential Revision: D14718584
Pulled By: siying
fbshipit-source-id: 0ac55dc9b5dbc18c7809092146bdf7eb9364b9ad
Summary:
`BlockBasedTableIterator` avoid reading next block on `Next()` if it detects the iterator will be out of bound, by checking against index key. The optimization was added in #2239, and by the time it only check the bound per block. It seems later change make it a per-key check, which introduce unnecessary key comparisons.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5101
Differential Revision: D14678707
Pulled By: siying
fbshipit-source-id: 2372446116753c7892ea4cec7b4b49ef87ba463e
Summary:
Following files were run through automatic formatter:
db/db_impl.cc
db/db_impl.h
db/db_impl_compaction_flush.cc
db/db_impl_debug.cc
db/db_impl_files.cc
db/db_impl_readonly.h
db/db_impl_write.cc
db/dbformat.cc
db/dbformat.h
table/block.cc
table/block.h
table/block_based_filter_block.cc
table/block_based_filter_block.h
table/block_based_filter_block_test.cc
table/block_based_table_builder.cc
table/block_based_table_reader.cc
table/block_based_table_reader.h
table/block_builder.cc
table/block_builder.h
table/block_fetcher.cc
table/block_prefix_index.cc
table/block_prefix_index.h
table/block_test.cc
table/format.cc
table/format.h
I could easily run all the files, but I don't want people to feel that
I'm doing it for lines of code changes :)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5114
Differential Revision: D14633040
Pulled By: siying
fbshipit-source-id: 3f346cb53bf21e8c10704400da548dfce1e89a52
Summary:
Currently `perf_context.user_key_comparison_count` is bump only in `InternalKeyComparator`. For places user comparator is used directly the counter is not bump. Fixing the majority of it.
Index iterator and filter code also use user comparator directly and don't bump the counter. It is not fixed in this patch.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5098
Differential Revision: D14603753
Pulled By: siying
fbshipit-source-id: 1cd41035644ca9e49b97a51030a5d1e15f5f3cae
Summary:
The code convention we are following, Google C++ Style, discourage
alias in header files, especially public headers:
https://google.github.io/styleguide/cppguide.html#Aliases
Remove some of them. Might removed some from .cc files as well to be consistent.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5113
Differential Revision: D14633030
Pulled By: siying
fbshipit-source-id: b990edc919d5de60295992284f980195e501d424
Summary:
Since `SstFileReader` don't know largest seqno of a file, it will fail this check when it open a file with global seqno: ca89ac2ba9/table/block_based_table_reader.cc (L730)
Changes:
* Pass largest_seqno=kMaxSequenceNumber from `SstFileReader` and allow it to bypass the above check.
* `BlockBasedTable::VerifyChecksum` also double check if checksum will match when excluding global seqno (this is to make the new test in sst_table_reader_test pass).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5097
Differential Revision: D14607434
Pulled By: riversand963
fbshipit-source-id: 9008599227c5fccbf9b73fee46b3bf4a1523f023
Summary:
- If block cache disabled or not used for meta-blocks, `BlockBasedTableReader::Rep::uncompression_dict` owns the `UncompressionDict`. It is preloaded during `PrefetchIndexAndFilterBlocks`.
- If block cache is enabled and used for meta-blocks, block cache owns the `UncompressionDict`, which holds dictionary and digested dictionary when needed. It is never prefetched though there is a TODO for this in the code. The cache key is simply the compression dictionary block handle.
- New stats for compression dictionary accesses in block cache: "BLOCK_CACHE_COMPRESSION_DICT_*" and "compression_dict_block_read_count"
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4881
Differential Revision: D13663801
Pulled By: ajkr
fbshipit-source-id: bdcc54044e180855cdcc57639b493b0e016c9a3f
Summary:
Previously we were cleaning up range tombstone meta-block by calling `ReleaseCachedEntry`, which wouldn't work if `value != nullptr && cache_handle == nullptr`. This happened at least in the case with mmap reads and block cache both enabled. I noticed `NewDataBlockIterator` intends to handle all these cases, so migrated to that instead of `NewUnfragmentedRangeTombstoneIterator`.
Also changed the table-opening logic to fail on `ReadRangeDelBlock` failure, since that can cause data corruption. Added a test case to verify this behavior. Note the test case does not fail on `TryReopen` because failure to preload table handlers is not considered critical. However, it does fail on any read involving that file since it cannot return correct data.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4810
Differential Revision: D13534296
Pulled By: ajkr
fbshipit-source-id: 55dde1111717cea6ec4bf38418daab81ccef3599
Summary:
Refactored and simplified `BlockBasedTable::Open` to be similar to `BlockBasedTableBuilder::Finish` as both these functions complement each other. Also added `BlockBasedTableBuilder::WriteFooter` along the way.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4636
Differential Revision: D12933319
Pulled By: sagar0
fbshipit-source-id: 1ff1d02f6d80a63b5ba720a1fc75e71c7344137b
Summary:
Fix block based table reader not using memory_allocator when allocating index blocks and compression dictionary blocks.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4678
Differential Revision: D13054594
Pulled By: yiwu-arbug
fbshipit-source-id: 379f25bcc665395662511c4f873f4b7b55104ce2
Summary:
Removed `one_time_use` flag, which removed the need for some
tests, and changed all `NewRangeTombstoneIterator` methods to return
`FragmentedRangeTombstoneIterators`.
These changes also led to removing `RangeDelAggregatorV2::AddUnfragmentedTombstones`
and one of the `MemTableListVersion::AddRangeTombstoneIterators` methods.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4692
Differential Revision: D13106570
Pulled By: abhimadan
fbshipit-source-id: cbab5432d7fc2d9cdfd8d9d40361a1bffaa8f845
Summary:
We carry compression type and "cachable" variables for every block in the block cache, while they take well-known values. 8-byte is wasted for each block (2-byte for useful information but it takes 8 bytes because of padding). With this change, these two variables are removed.
The cachable information is only useful in the process of reading the block. We use other information to infer from it. For compressed blocks, the compression type is a part of the block content itself so we can get it from there.
Some code is slightly refactored so that the cachable information can flow better.
Another change is to only use class BlockContents for compressed block, and narrow the class Block to only be used for uncompressed blocks, including blocks in compressed block cache. This can make the Block class less confusing. It also saves tens of bytes for each block in compressed block cache.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4650
Differential Revision: D12969070
Pulled By: siying
fbshipit-source-id: 548b62724e9eb66993026429fd9c7c3acd1f95ed
Summary:
Ran the following commands to recursively change all the files under RocksDB:
```
find . -type f -name "*.cc" -exec sed -i 's/ unique_ptr/ std::unique_ptr/g' {} +
find . -type f -name "*.cc" -exec sed -i 's/<unique_ptr/<std::unique_ptr/g' {} +
find . -type f -name "*.cc" -exec sed -i 's/ shared_ptr/ std::shared_ptr/g' {} +
find . -type f -name "*.cc" -exec sed -i 's/<shared_ptr/<std::shared_ptr/g' {} +
```
Running `make format` updated some formatting on the files touched.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4638
Differential Revision: D12934992
Pulled By: sagar0
fbshipit-source-id: 45a15d23c230cdd64c08f9c0243e5183934338a8
Summary:
Rename the interface, as it is mean to be a generic interface for memory allocation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4590
Differential Revision: D10866340
Pulled By: yiwu-arbug
fbshipit-source-id: 85cb753351a40cb856c046aeaa3f3b369eef3d16
Summary:
This allows tombstone fragmenting to only be performed when the table is opened, and cached for subsequent accesses.
On the same DB used in #4449, running `readrandom` results in the following:
```
readrandom : 0.983 micros/op 1017076 ops/sec; 78.3 MB/s (63103 of 100000 found)
```
Now that Get performance in the presence of range tombstones is reasonable, I also compared the performance between a DB with range tombstones, "expanded" range tombstones (several point tombstones that cover the same keys the equivalent range tombstone would cover, a common workaround for DeleteRange), and no range tombstones. The created DBs had 5 million keys each, and DeleteRange was called at regular intervals (depending on the total number of range tombstones being written) after 4.5 million Puts. The table below summarizes the results of a `readwhilewriting` benchmark (in order to provide somewhat more realistic results):
```
Tombstones? | avg micros/op | stddev micros/op | avg ops/s | stddev ops/s
----------------- | ------------- | ---------------- | ------------ | ------------
None | 0.6186 | 0.04637 | 1,625,252.90 | 124,679.41
500 Expanded | 0.6019 | 0.03628 | 1,666,670.40 | 101,142.65
500 Unexpanded | 0.6435 | 0.03994 | 1,559,979.40 | 104,090.52
1k Expanded | 0.6034 | 0.04349 | 1,665,128.10 | 125,144.57
1k Unexpanded | 0.6261 | 0.03093 | 1,600,457.50 | 79,024.94
5k Expanded | 0.6163 | 0.05926 | 1,636,668.80 | 154,888.85
5k Unexpanded | 0.6402 | 0.04002 | 1,567,804.70 | 100,965.55
10k Expanded | 0.6036 | 0.05105 | 1,667,237.70 | 142,830.36
10k Unexpanded | 0.6128 | 0.02598 | 1,634,633.40 | 72,161.82
25k Expanded | 0.6198 | 0.04542 | 1,620,980.50 | 116,662.93
25k Unexpanded | 0.5478 | 0.0362 | 1,833,059.10 | 121,233.81
50k Expanded | 0.5104 | 0.04347 | 1,973,107.90 | 184,073.49
50k Unexpanded | 0.4528 | 0.03387 | 2,219,034.50 | 170,984.32
```
After a large enough quantity of range tombstones are written, range tombstone Gets can become faster than reading from an equivalent DB with several point tombstones.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4493
Differential Revision: D10842844
Pulled By: abhimadan
fbshipit-source-id: a7d44534f8120e6aabb65779d26c6b9df954c509
Summary:
PR https://github.com/facebook/rocksdb/pull/4226 introduced per-level perf context which allows breaking down perf context by levels.
This PR takes advantage of the feature to populate a few counters related to bloom filters
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4581
Differential Revision: D10518010
Pulled By: miasantreble
fbshipit-source-id: 011244561783ec860d32d5b0fa6bce6e78d70ef8
Summary:
This is a conceptually simple change, but it touches many files to
pass the allocator through function calls.
We introduce CacheAllocator, which can be used by clients to configure
custom allocator for cache blocks. Our motivation is to hook this up
with folly's `JemallocNodumpAllocator`
(f43ce6d686/folly/experimental/JemallocNodumpAllocator.h),
but there are many other possible use cases.
Additionally, this commit cleans up memory allocation in
`util/compression.h`, making sure that all allocations are wrapped in a
unique_ptr as soon as possible.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4437
Differential Revision: D10132814
Pulled By: yiwu-arbug
fbshipit-source-id: be1343a4b69f6048df127939fea9bbc96969f564
Summary:
Given that index value is a BlockHandle, which is basically an <offset, size> pair we can apply delta encoding on the values. The first value at each index restart interval encoded the full BlockHandle but the rest encode only the size. Refer to IndexBlockIter::DecodeCurrentValue for the detail of the encoding. This reduces the index size which helps using the block cache more efficiently. The feature is enabled with using format_version 4.
The feature comes with a bit of cpu overhead which should be paid back by the higher cache hits due to smaller index block size.
Results with sysbench read-only using 4k blocks and using 16 index restart interval:
Format 2:
19585 rocksdb read-only range=100
Format 3:
19569 rocksdb read-only range=100
Format 4:
19352 rocksdb read-only range=100
Pull Request resolved: https://github.com/facebook/rocksdb/pull/3983
Differential Revision: D8361343
Pulled By: maysamyabandeh
fbshipit-source-id: f882ee082322acac32b0072e2bdbb0b5f854e651
Summary:
RocksDB used to store global_seqno in external SST files written by
SstFileWriter. During file ingestion, RocksDB uses `pwrite` to update the
`global_seqno`. Since random write is not supported in some non-POSIX compliant
file systems, external SST file ingestion is not supported on these file
systems. To address this limitation, we no longer update `global_seqno` during
file ingestion. Later RocksDB uses the MANIFEST and other information in table
properties to deduce global seqno for externally-ingested SST files.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4172
Differential Revision: D8961465
Pulled By: riversand963
fbshipit-source-id: 4382ec85270a96be5bc0cf33758ca2b167b05071
Summary:
Right now we use one hard-coded prefetch size to prefetch data from the tail of the SST files. However, this may introduce a waste for some use cases, while not efficient for others.
Introduce a way to adjust this prefetch size by tracking 32 recent times, and pick a value with which the wasted read is less than 10%
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4156
Differential Revision: D8916847
Pulled By: siying
fbshipit-source-id: 8413f9eb3987e0033ed0bd910f83fc2eeaaf5758
Summary:
BlockIter is getting crowded including details that specific only to either index or data blocks. The patch moves down such details to DataBlockIter and IndexBlockIter, both inheriting from BlockIter.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4121
Differential Revision: D8816832
Pulled By: maysamyabandeh
fbshipit-source-id: d492e74155c11d8a0c1c85cd7ee33d24c7456197
Summary:
https://github.com/facebook/rocksdb/pull/3881 fixed a bug where PinnableSlice pin mmap files which could be deleted with background compaction. This is however a non-issue for ReadOnlyDB when there is no compaction running and max_open_files is -1. This patch reenables the pinning feature for that case.
Closes https://github.com/facebook/rocksdb/pull/4053
Differential Revision: D8662546
Pulled By: maysamyabandeh
fbshipit-source-id: 402962602eb0f644e17822748332999c3af029fd
Summary:
Previously in https://github.com/facebook/rocksdb/pull/3601 bloom filter will only be checked if `prefix_extractor` in the mutable_cf_options matches the one found in the SST file.
This PR relaxes the requirement by checking if all keys in the range [user_key, iterate_upper_bound) all share the same prefix after transforming using the BF in the SST file. If so, the bloom filter is considered compatible and will continue to be looked at.
Closes https://github.com/facebook/rocksdb/pull/3899
Differential Revision: D8157459
Pulled By: miasantreble
fbshipit-source-id: 18d17cba56a1005162f8d5db7a27aba277089c41
Summary:
Pass in `for_compaction` to `BlockBasedTableIterator` via `BlockBasedTableReader::NewIterator`.
In 7103559f49, `for_compaction` was set in `BlockBasedTable::Rep` via `BlockBasedTable::SetupForCompaction`. In hindsight it was not the right decision; it also caused TSAN to complain.
Closes https://github.com/facebook/rocksdb/pull/4048
Differential Revision: D8601056
Pulled By: sagar0
fbshipit-source-id: 30127e898c15c38c1080d57710b8c5a6d64a0ab3
Summary:
Top-level index in partitioned index/filter blocks are small and could be pinned in memory. So far we use that by cache_index_and_filter_blocks to false. This however make it difficult to keep account of the total memory usage. This patch introduces pin_top_level_index_and_filter which in combination with cache_index_and_filter_blocks=true keeps the top-level index in cache and yet pinned them to avoid cache misses and also cache lookup overhead.
Closes https://github.com/facebook/rocksdb/pull/4037
Differential Revision: D8596218
Pulled By: maysamyabandeh
fbshipit-source-id: 3a5f7f9ca6b4b525b03ff6bd82354881ae974ad2
Summary:
This PR extends the improvements in #3282 to also work when using Direct IO.
We see **4.5X performance improvement** in seekrandom benchmark doing long range scans, when using direct reads, on flash.
**Description:**
This change improves the performance of iterators doing long range scans (e.g. big/full index or table scans in MyRocks) by using readahead and prefetching additional data on each disk IO, and storing in a local buffer. This prefetching is automatically enabled on noticing more than 2 IOs for the same table file during iteration. The readahead size starts with 8KB and is exponentially increased on each additional sequential IO, up to a max of 256 KB. This helps in cutting down the number of IOs needed to complete the range scan.
**Implementation Details:**
- Used `FilePrefetchBuffer` as the underlying buffer to store the readahead data. `FilePrefetchBuffer` can now take file_reader, readahead_size and max_readahead_size as input to the constructor, and automatically do readahead.
- `FilePrefetchBuffer::TryReadFromCache` can now call `FilePrefetchBuffer::Prefetch` if readahead is enabled.
- `AlignedBuffer` (which is the underlying store for `FilePrefetchBuffer`) now takes a few additional args in `AlignedBuffer::AllocateNewBuffer` to allow copying data from the old buffer.
- Made sure not to re-read partial chunks of data that were already available in the buffer, from device again.
- Fixed a couple of cases where `AlignedBuffer::cursize_` was not being properly kept up-to-date.
**Constraints:**
- Similar to #3282, this gets currently enabled only when ReadOptions.readahead_size = 0 (which is the default value).
- Since the prefetched data is stored in a temporary buffer allocated on heap, this could increase the memory usage if you have many iterators doing long range scans simultaneously.
- Enabled only for user reads, and disabled for compactions. Compaction reads are controlled by the options `use_direct_io_for_flush_and_compaction` and `compaction_readahead_size`, and the current feature takes precautions not to mess with them.
**Benchmarks:**
I used the same benchmark as used in #3282.
Data fill:
```
TEST_TMPDIR=/data/users/$USER/benchmarks/iter ./db_bench -benchmarks=fillrandom -num=1000000000 -compression_type="none" -level_compaction_dynamic_level_bytes
```
Do a long range scan: Seekrandom with large number of nexts
```
TEST_TMPDIR=/data/users/$USER/benchmarks/iter ./db_bench -benchmarks=seekrandom -use_direct_reads -duration=60 -num=1000000000 -use_existing_db -seek_nexts=10000 -statistics -histogram
```
```
Before:
seekrandom : 37939.906 micros/op 26 ops/sec; 29.2 MB/s (1636 of 1999 found)
With this change:
seekrandom : 8527.720 micros/op 117 ops/sec; 129.7 MB/s (6530 of 7999 found)
```
~4.5X perf improvement. Taken on an average of 3 runs.
Closes https://github.com/facebook/rocksdb/pull/3884
Differential Revision: D8082143
Pulled By: sagar0
fbshipit-source-id: 4d7a8561cbac03478663713df4d31ad2620253bb
Summary:
https://github.com/facebook/rocksdb/pull/3764 introduced an optimization feature to skip duplicate prefix entires in full bloom filters. Unfortunately it also introduces a bug in partitioned full filters, where the duplicate prefix should still be inserted if it is in a new partition. The patch fixes the bug by resetting the duplicate detection logic each time a partition is cut.
This bug could result into false negatives, which means that DB could skip an existing key.
Closes https://github.com/facebook/rocksdb/pull/4024
Differential Revision: D8518866
Pulled By: maysamyabandeh
fbshipit-source-id: 044f4d988e606a330ecafd8c79daceb68b8796bf
Summary:
b555ed30a4 makes the BlockBasedTableIterator to be invalidated if the current position if over the upper bound. However, this can bring performance regression to the case of multiple Seek()s hitting the same data block but all out of upper bound.
For example, if an SST file has a data block containing following keys : {a, z}
The user sets the upper bound to be "x", and it executed following queries:
Seek("b")
Seek("c")
Seek("d")
Before the upper bound optimization, these queries always come to this same current data block of the iterator, but now inside each Seek() the data block is read from the block cache but is returned again.
To prevent this regression case, we keep the current data block iterator if it is upper bound.
Closes https://github.com/facebook/rocksdb/pull/4004
Differential Revision: D8463192
Pulled By: siying
fbshipit-source-id: 8710628b30acde7063a097c3184d6c4333a8ef81
Summary:
Index blocks have the same format as data blocks. The keys therefore similarly to the keys in the data blocks are internal keys, which means that in addition to the user key it also has 8 bytes that encodes sequence number and value type. This extra 8 bytes however is not necessary in index blocks since the index keys act as an separator between two data blocks. The only exception is when the last key of a block and the first key of the next block share the same user key, in which the sequence number is required to act as a separator.
The patch excludes the sequence from index keys only if the above special case does not happen for any of the index keys. It then records that in the property block. The reader looks at the property block to see if it should expect sequence numbers in the keys of the index block.s
Closes https://github.com/facebook/rocksdb/pull/3894
Differential Revision: D8118775
Pulled By: maysamyabandeh
fbshipit-source-id: 915479f028b5799ca91671d67455ecdefbd873bd
Summary:
Please refer to earlier discussion in [issue 3609](https://github.com/facebook/rocksdb/issues/3609).
There was also an alternative fix in [PR 3888](https://github.com/facebook/rocksdb/pull/3888), but the proposed solution requires complex change.
To summarize the cause of the problem. Upon creation of a column family, a `BlockBasedTableFactory` object is `new`ed and encapsulated by a `std::shared_ptr`. Since there is no other `std::shared_ptr` pointing to this `BlockBasedTableFactory`, when the column family is dropped, the `ColumnFamilyData` is `delete`d, causing the destructor of `std::shared_ptr`. Since there is no other `std::shared_ptr`, the underlying memory is also freed.
Later when the db exits, it releases all the table readers, including the table readers that have been operating on the dropped column family. This needs to access the `table_options` owned by `BlockBasedTableFactory` that has already been deleted. Therefore, a segfault is raised.
Previous workaround is to purge all obsolete files upon `ColumnFamilyData` destruction, which leads to a force release of table readers of the dropped column family. However this does not work when the user disables file deletion.
Our solution in this PR is making a copy of `table_options` in `BlockBasedTable::Rep`. This solution increases memory copy and usage, but is much simpler.
Test plan
```
$ make -j16
$ ./column_family_test --gtest_filter=ColumnFamilyTest.CreateDropAndDestroy:ColumnFamilyTest.CreateDropAndDestroyWithoutFileDeletion
```
Expected behavior:
All tests should pass.
Closes https://github.com/facebook/rocksdb/pull/3898
Differential Revision: D8149421
Pulled By: riversand963
fbshipit-source-id: eaecc2e064057ef607fbdd4cc275874f866c3438
Summary:
Currently it is not possible to change bloom filter config without restart the db, which is causing a lot of operational complexity for users.
This PR aims to make it possible to dynamically change bloom filter config.
Closes https://github.com/facebook/rocksdb/pull/3601
Differential Revision: D7253114
Pulled By: miasantreble
fbshipit-source-id: f22595437d3e0b86c95918c484502de2ceca120c
Summary:
Before this PR, Iterator/InternalIterator may simultaneously have non-ok status() and Valid() = true. That state means that the last operation failed, but the iterator is nevertheless positioned on some unspecified record. Likely intended uses of that are:
* If some sst files are corrupted, a normal iterator can be used to read the data from files that are not corrupted.
* When using read_tier = kBlockCacheTier, read the data that's in block cache, skipping over the data that is not.
However, this behavior wasn't documented well (and until recently the wiki on github had misleading incorrect information). In the code there's a lot of confusion about the relationship between status() and Valid(), and about whether Seek()/SeekToLast()/etc reset the status or not. There were a number of bugs caused by this confusion, both inside rocksdb and in the code that uses rocksdb (including ours).
This PR changes the convention to:
* If status() is not ok, Valid() always returns false.
* Any seek operation resets status. (Before the PR, it depended on iterator type and on particular error.)
This does sacrifice the two use cases listed above, but siying said it's ok.
Overview of the changes:
* A commit that adds missing status checks in MergingIterator. This fixes a bug that actually affects us, and we need it fixed. `DBIteratorTest.NonBlockingIterationBugRepro` explains the scenario.
* Changes to lots of iterator types to make all of them conform to the new convention. Some bug fixes along the way. By far the biggest changes are in DBIter, which is a big messy piece of code; I tried to make it less big and messy but mostly failed.
* A stress-test for DBIter, to gain some confidence that I didn't break it. It does a few million random operations on the iterator, while occasionally modifying the underlying data (like ForwardIterator does) and occasionally returning non-ok status from internal iterator.
To find the iterator types that needed changes I searched for "public .*Iterator" in the code. Here's an overview of all 27 iterator types:
Iterators that didn't need changes:
* status() is always ok(), or Valid() is always false: MemTableIterator, ModelIter, TestIterator, KVIter (2 classes with this name anonymous namespaces), LoggingForwardVectorIterator, VectorIterator, MockTableIterator, EmptyIterator, EmptyInternalIterator.
* Thin wrappers that always pass through Valid() and status(): ArenaWrappedDBIter, TtlIterator, InternalIteratorFromIterator.
Iterators with changes (see inline comments for details):
* DBIter - an overhaul:
- It used to silently skip corrupted keys (`FindParseableKey()`), which seems dangerous. This PR makes it just stop immediately after encountering a corrupted key, just like it would for other kinds of corruption. Let me know if there was actually some deeper meaning in this behavior and I should put it back.
- It had a few code paths silently discarding subiterator's status. The stress test caught a few.
- The backwards iteration code path was expecting the internal iterator's set of keys to be immutable. It's probably always true in practice at the moment, since ForwardIterator doesn't support backwards iteration, but this PR fixes it anyway. See added DBIteratorTest.ReverseToForwardBug for an example.
- Some parts of backwards iteration code path even did things like `assert(iter_->Valid())` after a seek, which is never a safe assumption.
- It used to not reset status on seek for some types of errors.
- Some simplifications and better comments.
- Some things got more complicated from the added error handling. I'm open to ideas for how to make it nicer.
* MergingIterator - check status after every operation on every subiterator, and in some places assert that valid subiterators have ok status.
* ForwardIterator - changed to the new convention, also slightly simplified.
* ForwardLevelIterator - fixed some bugs and simplified.
* LevelIterator - simplified.
* TwoLevelIterator - changed to the new convention. Also fixed a bug that would make SeekForPrev() sometimes silently ignore errors from first_level_iter_.
* BlockBasedTableIterator - minor changes.
* BlockIter - replaced `SetStatus()` with `Invalidate()` to make sure non-ok BlockIter is always invalid.
* PlainTableIterator - some seeks used to not reset status.
* CuckooTableIterator - tiny code cleanup.
* ManagedIterator - fixed some bugs.
* BaseDeltaIterator - changed to the new convention and fixed a bug.
* BlobDBIterator - seeks used to not reset status.
* KeyConvertingIterator - some small change.
Closes https://github.com/facebook/rocksdb/pull/3810
Differential Revision: D7888019
Pulled By: al13n321
fbshipit-source-id: 4aaf6d3421c545d16722a815b2fa2e7912bc851d
Summary:
Currently the partitioned index iterator creates a new ReadOptions which ignores the fill_cache config set to ReadOptions passed by the user. The patch propagates fill_cache from the user's ReadOptions to that of partition index iterator.
Also it clarifies the contract of fill_cache that i) it does not apply to filters, ii) it still charges block cache for the size of the data block, it still pin the block if it is already in the block cache.
Closes https://github.com/facebook/rocksdb/pull/3739
Differential Revision: D7678308
Pulled By: maysamyabandeh
fbshipit-source-id: 53ed96424ae922e499e2d4e3580ddc3f0db893da
Summary:
b555ed30a4 introduces a regression, which causes blocks always to be pinned in block based iterators. Fix it.
Closes https://github.com/facebook/rocksdb/pull/3582
Differential Revision: D7189534
Pulled By: siying
fbshipit-source-id: 117dc7a03d0a0e360424db02efb366e12da2be03
Summary:
Use a customzied BlockBasedTableIterator and LevelIterator to replace current implementations leveraging two-level-iterator. Hope the customized logic will make code easier to understand. As a side effect, BlockBasedTableIterator reduces the allocation for the data block iterator object, and avoid the virtual function call to it, because we can directly reference BlockIter, a final class. Similarly, LevelIterator reduces virtual function call to the dummy iterator iterating the file metadata. It also enabled further optimization.
The upper bound check is also moved from index block to data block. This implementation fits this iterator better. After the change, forwared iterator is slightly optimized to ensure we trim those iterators.
The two-level-iterator now is only used by partitioned index, so it is simplified.
Closes https://github.com/facebook/rocksdb/pull/3406
Differential Revision: D6809041
Pulled By: siying
fbshipit-source-id: 7da3b9b1d3c8e9d9405302c15920af1fcaf50ffa
Summary:
`ReadBlockFromFile` uses a stack buffer to hold small data blocks before passing them to the compression library, which outputs uncompressed data in a heap buffer. In the case of `kNoCompression` there is a `memcpy` to copy from stack buffer to heap buffer.
This PR optimizes `ReadBlockFromFile` to skip the stack buffer for files whose blocks are known to be uncompressed. We determine this using the SST file property, "compression_name", if it's available.
Closes https://github.com/facebook/rocksdb/pull/3472
Differential Revision: D6920848
Pulled By: ajkr
fbshipit-source-id: 5c753e804efc178b9229ae5dbe6a4adc32031f07
Summary:
ReadOptions.fill_cache is set in compaction inputs and can be set by users in their queries too. It tells RocksDB not to put a data block used to block cache.
The memory used by the data block is, however, not trackable by users.
To make the system more manageable, we can cost the block to block cache while using it, and then release it after using.
Closes https://github.com/facebook/rocksdb/pull/3333
Differential Revision: D6670230
Pulled By: miasantreble
fbshipit-source-id: ab848d3ed286bd081a13ee1903de357b56cbc308
Summary:
This change improves the performance of iterators doing long range scans (e.g. big/full table scans in MyRocks) by using readahead and prefetching additional data on each disk IO. This prefetching is automatically enabled on noticing more than 2 IOs for the same table file during iteration. The readahead size starts with 8KB and is exponentially increased on each additional sequential IO, up to a max of 256 KB. This helps in cutting down the number of IOs needed to complete the range scan.
Constraints:
- The prefetched data is stored by the OS in page cache. So this currently works only for non direct-reads use-cases i.e applications which use page cache. (Direct-I/O support will be enabled in a later PR).
- This gets currently enabled only when ReadOptions.readahead_size = 0 (which is the default value).
Thanks to siying for the original idea and implementation.
**Benchmarks:**
Data fill:
```
TEST_TMPDIR=/data/users/$USER/benchmarks/iter ./db_bench -benchmarks=fillrandom -num=1000000000 -compression_type="none" -level_compaction_dynamic_level_bytes
```
Do a long range scan: Seekrandom with large number of nexts
```
TEST_TMPDIR=/data/users/$USER/benchmarks/iter ./db_bench -benchmarks=seekrandom -duration=60 -num=1000000000 -use_existing_db -seek_nexts=10000 -statistics -histogram
```
Page cache was cleared before each experiment with the command:
```
sudo sh -c "echo 3 > /proc/sys/vm/drop_caches"
```
```
Before:
seekrandom : 34020.945 micros/op 29 ops/sec; 32.5 MB/s (1636 of 1999 found)
With this change:
seekrandom : 8726.912 micros/op 114 ops/sec; 126.8 MB/s (5702 of 6999 found)
```
~3.9X performance improvement.
Also verified with strace and gdb that the readahead size is increasing as expected.
```
strace -e readahead -f -T -t -p <db_bench process pid>
```
Closes https://github.com/facebook/rocksdb/pull/3282
Differential Revision: D6586477
Pulled By: sagar0
fbshipit-source-id: 8a118a0ed4594fbb7f5b1cafb242d7a4033cb58c
Summary:
This is a pre-cleaning up before a major block based table iterator refactoring. BlockBasedTable::NewDataBlockIterator() will always return BlockIter. This simplifies the logic and code and enable further refactoring and optimization.
Closes https://github.com/facebook/rocksdb/pull/3398
Differential Revision: D6780165
Pulled By: siying
fbshipit-source-id: 273f7dc896724f682c0118fb69a359d9cc4418b4
Summary:
table/block.cc:
420 }
CID 1396127 (#1 of 1): Uninitialized scalar field (UNINIT_CTOR)
7. uninit_member: Non-static class member restart_offset_ is not initialized in this constructor nor in any functions that it calls.
421}
table/block_based_table_builder.cc:
CID 1418259 (#1 of 1): Uninitialized scalar field (UNINIT_CTOR)
7. uninit_member: Non-static class member compressed_cache_key_prefix_size is not initialized in this constructor nor in any functions that it calls.
table/block_based_table_reader.h:
3. uninit_member: Non-static class member index_type is not initialized in this constructor nor in any functions that it calls.
CID 1396147 (#1 of 1): Uninitialized scalar field (UNINIT_CTOR)
5. uninit_member: Non-static class member hash_index_allow_collision is not initialized in this constructor nor in any functions that it calls.
413 global_seqno(kDisableGlobalSequenceNumber) {}
414
table/cuckoo_table_reader.cc:
55 if (hash_funs == user_props.end()) {
56 status_ = Status::Corruption("Number of hash functions not found");
5. uninit_member: Non-static class member is_last_level_ is not initialized in this constructor nor in any functions that it calls.
7. uninit_member: Non-static class member identity_as_first_hash_ is not initialized in this constructor nor in any functions that it calls.
9. uninit_member: Non-static class member use_module_hash_ is not initialized in this constructor nor in any functions that it calls.
11. uninit_member: Non-static class member num_hash_func_ is not initialized in this constructor nor in any functions that it calls.
13. uninit_member: Non-static class member key_length_ is not initialized in this constructor nor in any functions that it calls.
15. uninit_member: Non-static class member user_key_length_ is not initialized in this constructor nor in any functions that it calls.
17. uninit_member: Non-static class member value_length_ is not initialized in this constructor nor in any functions that it calls.
19. uninit_member: Non-static class member bucket_length_ is not initialized in this constructor nor in any functions that it calls.
21. uninit_member: Non-static class member cuckoo_block_size_ is not initialized in this constructor nor in any functions that it calls.
23. uninit_member: Non-static class member cuckoo_block_bytes_minus_one_ is not initialized in this constructor nor in any functions that it calls.
CID 1322785 (#2 of 2): Uninitialized scalar field (UNINIT_CTOR)
25. uninit_member: Non-static class member table_size_ is not initialized in this constructor nor in any functions that it calls.
57 return;
table/plain_table_index.h:
2. uninit_member: Non-static class member index_size_ is not initialized in this constructor nor in any functions that it calls.
CID 1322801 (#1 of 1): Uninitialized scalar field (UNINIT_CTOR)
4. uninit_member: Non-static class member sub_index_size_ is not initialized in this constructor nor in any functions that it calls.
128 huge_page_tlb_size_(huge_page_tlb_size) {}
129
Closes https://github.com/facebook/rocksdb/pull/3113
Differential Revision: D6505719
Pulled By: yiwu-arbug
fbshipit-source-id: 38f44d8f9dfefb4c2e25d83b8df25a5201c75618
Summary:
This is the continuation of https://github.com/facebook/rocksdb/pull/2661 for filter partitions. When pin_l0 is set (along with cache_xxx), then open table open the filter partitions are loaded into the cache and pinned there.
Closes https://github.com/facebook/rocksdb/pull/2766
Differential Revision: D5671098
Pulled By: maysamyabandeh
fbshipit-source-id: 174f24018f1d7f1129621e7380287b65b67d2115
Summary:
This fixes the existing logic for pinning l0 index partitions. The patch preloads the partitions into block cache and pin them if they belong to level 0 and pin_l0 is set.
The drawback is that it does many small IOs when preloading all the partitions into the cache is direct io is enabled. Working for a solution for that.
Closes https://github.com/facebook/rocksdb/pull/2661
Differential Revision: D5554010
Pulled By: maysamyabandeh
fbshipit-source-id: 1e6f32a3524d71355c77d4138516dcfb601ca7b2
Summary:
Right now, if direct I/O is enabled, prefetching the last 512KB cannot be applied, except compaction inputs or readahead is enabled for iterators. This can create a lot of I/O for HDD cases. To solve the problem, the 512KB is prefetched in block based table if direct I/O is enabled. The prefetched buffer is passed in totegher with random access file reader, so that we try to read from the buffer before reading from the file. This can be extended in the future to support flexible user iterator readahead too.
Closes https://github.com/facebook/rocksdb/pull/2708
Differential Revision: D5593091
Pulled By: siying
fbshipit-source-id: ee36ff6d8af11c312a2622272b21957a7b5c81e7
Summary:
We need a tool to check any sst file corruption in the db.
It will check all the sst files in current version and read all the blocks (data, meta, index) with checksum verification. If any verification fails, the function will return non-OK status.
Closes https://github.com/facebook/rocksdb/pull/2498
Differential Revision: D5324269
Pulled By: lightmark
fbshipit-source-id: 6f8a272008b722402a772acfc804524c9d1a483b
Summary:
Breaking commit: d12691b86f
In the above commit, I moved the `TableCache` cleanup logic from `Version` destructor into `PurgeObsoleteFiles`. I missed cleaning up `TableCache` entries for the current `Version` during DB destruction.
This PR adds that logic to `VersionSet` destructor. One unfortunate side effect is now we're potentially deleting `TableReader`s after `column_family_set_.reset()`, which means we can't call `BlockBasedTableReader::Close` a second time as the block cache might already be destroyed.
Closes https://github.com/facebook/rocksdb/pull/2662
Differential Revision: D5515108
Pulled By: ajkr
fbshipit-source-id: 2cb820e19aa813e0d258d17f76b2d7b6b7ee0b18
