From d1aca4a5ae6b3f94dddf3f69cbef4463f3858ede Mon Sep 17 00:00:00 2001 From: akankshamahajan Date: Fri, 11 Nov 2022 13:34:49 -0800 Subject: [PATCH] Fix async_io regression in scans (#10939) Summary: Fix async_io regression in scans due to incorrect check which was causing the valid data in buffer to be cleared during seek. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10939 Test Plan: - stress tests export CRASH_TEST_EXT_ARGS="--async_io=1" make crash_test -j32 - Ran db_bench command which was caught the regression: ./db_bench --db=/rocksdb_async_io_testing/prefix_scan --disable_wal=1 --use_existing_db=true --benchmarks="seekrandom" -key_size=32 -value_size=512 -num=50000000 -use_direct_reads=false -seek_nexts=963 -duration=30 -ops_between_duration_checks=1 --async_io=true --compaction_readahead_size=4194304 --log_readahead_size=0 --blob_compaction_readahead_size=0 --initial_auto_readahead_size=65536 --num_file_reads_for_auto_readahead=0 --max_auto_readahead_size=524288 seekrandom : 3777.415 micros/op 264 ops/sec 30.000 seconds 7942 operations; 132.3 MB/s (7942 of 7942 found) Reviewed By: anand1976 Differential Revision: D41173899 Pulled By: akankshamahajan15 fbshipit-source-id: 2d75b06457d65b1851c92382565d9c3fac329dfe --- HISTORY.md | 1 + file/file_prefetch_buffer.cc | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/HISTORY.md b/HISTORY.md index 207d22f26..439223c2a 100644 --- a/HISTORY.md +++ b/HISTORY.md @@ -8,6 +8,7 @@ * Fix memory corruption error in scans if async_io is enabled. Memory corruption happened if there is IOError while reading the data leading to empty buffer and other buffer already in progress of async read goes again for reading. * Fix failed memtable flush retry bug that could cause wrongly ordered updates, which would surface to writers as `Status::Corruption` in case of `force_consistency_checks=true` (default). It affects use cases that enable both parallel flush (`max_background_flushes > 1` or `max_background_jobs >= 8`) and non-default memtable count (`max_write_buffer_number > 2`). * Fixed an issue where the `READ_NUM_MERGE_OPERANDS` ticker was not updated when the base key-value or tombstone was read from an SST file. +* Fixed a regression in scan for async_io. During seek, valid buffers were getting cleared causing a regression. ### New Features * Add basic support for user-defined timestamp to Merge (#10819). diff --git a/file/file_prefetch_buffer.cc b/file/file_prefetch_buffer.cc index 9222acdfa..9ea9129e2 100644 --- a/file/file_prefetch_buffer.cc +++ b/file/file_prefetch_buffer.cc @@ -664,7 +664,7 @@ bool FilePrefetchBuffer::TryReadFromCacheAsync( // submitted in PrefetchAsync should match with this request. Otherwise // buffers will be outdated. // Random offset called. So abort the IOs. - if (bufs_[curr_].offset_ != offset) { + if (prev_offset_ != offset) { AbortAllIOs(); bufs_[curr_].buffer_.Clear(); bufs_[curr_ ^ 1].buffer_.Clear();