-
Notifications
You must be signed in to change notification settings - Fork 6.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix some errors in async prefetching in FilePrefetchBuffer #9734
Conversation
90fade1
to
ae1a6e2
Compare
@akankshamahajan15 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
ae1a6e2
to
11b4d41
Compare
@akankshamahajan15 has updated the pull request. You must reimport the pull request before landing. |
Summary: db_bench was failing due to some corner cases which has been fixed in this PR. Test Plan: ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 -async_io=1
11b4d41
to
10880b8
Compare
@akankshamahajan15 has updated the pull request. You must reimport the pull request before landing. |
@akankshamahajan15 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
It's helpful to include the original error message in the PR description. |
@akankshamahajan15 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@@ -66,6 +66,17 @@ void FilePrefetchBuffer::CalculateOffsetAndLen(size_t alignment, | |||
// chunk_len is greater than 0. | |||
bufs_[index].buffer_.RefitTail(static_cast<size_t>(chunk_offset_in_buffer), | |||
static_cast<size_t>(chunk_len)); | |||
} else if (chunk_len > 0) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For async prefetching we don’t refit the data so it allocate new memory. Planning to handle this in next iteration
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I didn't understand this case. By the time we get here, PrefetchAsync would have already consumed as much data as it can from the buffers. IIUC, one or both buffers should be empty, i.e chunk_len will be 0. When exactly will this case be true?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This happens when data is in cache and it updates the read pattern in FilePrefetchBuffer
because of which when it reads the next pattern, it needs to prefetch the data and some chunk is left in the buffer.
One example:
1. Read offset: 5649638, size: 4774
2. curr: 1, bufs_[curr_].offset_: 5623808, bufs_[curr_].buffer_.CurrentSize(): 16384, bufs_[curr_ ^1].offset: 5640192, bufs_[curr_^1].buffer_.CurrentSize(): 12288
3. For offset: 5649638, prev_offset: 5644838, prev_len: 4800 (For previous offset: In cache offset: 5644838).
4. So it clears the first buffer, swap it with second and second buffer has chunk_len : 4096.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I will try to add the unit test to catch this error. But this was faced in db_stress
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Only verified that this PR affects only an experimental feature readOptions::async_io
. Will defer to @anand1976 to fully evaluate.
@akankshamahajan15 has updated the pull request. You must reimport the pull request before landing. |
@akankshamahajan15 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Summary: In ReadOption `async_io` which prefetches the data asynchronously, db_bench and db_stress runs were failing because wrong data was prefetched which resulted in Error: Checksum mismatched. Wrong data was copied because capacity was less than actual size needed. It has been fixed in this PR. Since there are two separate methods for async and sync prefetching, these changes are in async prefetching methods and any changes would not effect normal prefetching. I ran the regressions to make sure normal prefetching is fine. Pull Request resolved: #9734 Test Plan: 1. CircleCI jobs 2. Ran db_bench ``` . /db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 -async_io=1 -adaptive_readahead=1 ``` 3. Ran db_stress test ``` export CRASH_TEST_EXT_ARGS=" --async_io=1 --adaptive_readahead=1" make crash_test -j ``` 4. Run regressions for async_io disabled. Old flow without any async changes: ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.0 Date: Thu Mar 17 13:11:34 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 483618.390 micros/op 2 ops/sec; 338.9 MB/s (249 of 249 found) ``` With async prefetching changes and async_io disabled to make sure in normal prefetching there is no regression. ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 --async_io=0 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.1 Date: Wed Mar 23 15:56:37 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 481819.816 micros/op 2 ops/sec; 340.2 MB/s (250 of 250 found) ``` Reviewed By: riversand963 Differential Revision: D35058471 Pulled By: akankshamahajan15 fbshipit-source-id: 9233a1e6d97cea0c7a8111bfb9e8ac3251c341ce
Summary: In ReadOption `async_io` which prefetches the data asynchronously, db_bench and db_stress runs were failing because wrong data was prefetched which resulted in Error: Checksum mismatched. Wrong data was copied because capacity was less than actual size needed. It has been fixed in this PR. Since there are two separate methods for async and sync prefetching, these changes are in async prefetching methods and any changes would not effect normal prefetching. I ran the regressions to make sure normal prefetching is fine. Pull Request resolved: #9734 Test Plan: 1. CircleCI jobs 2. Ran db_bench ``` . /db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 -async_io=1 -adaptive_readahead=1 ``` 3. Ran db_stress test ``` export CRASH_TEST_EXT_ARGS=" --async_io=1 --adaptive_readahead=1" make crash_test -j ``` 4. Run regressions for async_io disabled. Old flow without any async changes: ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.0 Date: Thu Mar 17 13:11:34 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 483618.390 micros/op 2 ops/sec; 338.9 MB/s (249 of 249 found) ``` With async prefetching changes and async_io disabled to make sure in normal prefetching there is no regression. ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 --async_io=0 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.1 Date: Wed Mar 23 15:56:37 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 481819.816 micros/op 2 ops/sec; 340.2 MB/s (250 of 250 found) ``` Reviewed By: riversand963 Differential Revision: D35058471 Pulled By: akankshamahajan15 fbshipit-source-id: 9233a1e6d97cea0c7a8111bfb9e8ac3251c341ce
…9734) Summary: In ReadOption `async_io` which prefetches the data asynchronously, db_bench and db_stress runs were failing because wrong data was prefetched which resulted in Error: Checksum mismatched. Wrong data was copied because capacity was less than actual size needed. It has been fixed in this PR. Since there are two separate methods for async and sync prefetching, these changes are in async prefetching methods and any changes would not effect normal prefetching. I ran the regressions to make sure normal prefetching is fine. Pull Request resolved: facebook#9734 Test Plan: 1. CircleCI jobs 2. Ran db_bench ``` . /db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 -async_io=1 -adaptive_readahead=1 ``` 3. Ran db_stress test ``` export CRASH_TEST_EXT_ARGS=" --async_io=1 --adaptive_readahead=1" make crash_test -j ``` 4. Run regressions for async_io disabled. Old flow without any async changes: ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.0 Date: Thu Mar 17 13:11:34 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 483618.390 micros/op 2 ops/sec; 338.9 MB/s (249 of 249 found) ``` With async prefetching changes and async_io disabled to make sure in normal prefetching there is no regression. ``` ./db_bench -use_existing_db=true -db=/tmp/prefix_scan_prefetch_main -benchmarks="seekrandom" -key_size=32 -value_size=512 -num=5000000 -use_direct_reads=true -seek_nexts=327680 -duration=120 -ops_between_duration_checks=1 --async_io=0 Initializing RocksDB Options from the specified file Initializing RocksDB Options from command-line flags RocksDB: version 7.1 Date: Wed Mar 23 15:56:37 2022 CPU: 24 * Intel Core Processor (Broadwell) CPUCache: 16384 KB Keys: 32 bytes each (+ 0 bytes user-defined timestamp) Values: 512 bytes each (256 bytes after compression) Entries: 5000000 Prefix: 0 bytes Keys per prefix: 0 RawSize: 2594.0 MB (estimated) FileSize: 1373.3 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: Snappy Compression sampling rate: 0 Memtablerep: SkipListFactory Perf Level: 1 ------------------------------------------------ DB path: [/tmp/prefix_scan_prefetch_main] seekrandom : 481819.816 micros/op 2 ops/sec; 340.2 MB/s (250 of 250 found) ``` Reviewed By: riversand963 Differential Revision: D35058471 Pulled By: akankshamahajan15 fbshipit-source-id: 9233a1e6d97cea0c7a8111bfb9e8ac3251c341ce
Summary: In ReadOption
async_io
which prefetches the data asynchronously, db_bench and db_stress runs were failing because wrong data was prefetched which resulted in Error: Checksum mismatched. Wrong data was copied because capacity was less than actual size needed. It has been fixed in this PR.Since there are two separate methods for async and sync prefetching, these changes are in async prefetching methods and any changes would not effect normal prefetching. I ran the regressions to make sure normal prefetching is fine.
Test Plan:
CircleCI jobs
Ran db_bench
Old flow without any async changes:
With async prefetching changes and async_io disabled to make sure in normal prefetching there is no regression.