mirror of
https://github.com/facebook/rocksdb.git
synced 2024-11-30 22:41:48 +00:00
b45738622a
Summary: This is mostly taken from https://github.com/facebook/rocksdb/issues/10427 with my own comments addressed. This PR plumbs the user’s `ReadOptions` down to `GetOrReadIndexBlock()`, `GetOrReadFilterBlock()`, and `GetFilterPartitionBlock()`. Now those functions no longer have to make up a `ReadOptions` with incomplete information. I also let `PartitionIndexReader::NewIterator()` pass through its caller's `ReadOptions::verify_checksums`, which was inexplicably dropped previously. Fixes https://github.com/facebook/rocksdb/issues/10463 Pull Request resolved: https://github.com/facebook/rocksdb/pull/11208 Test Plan: Functional: - Measured `-verify_checksum=false` applies to metadata blocks read outside of table open - setup command: `TEST_TMPDIR=/tmp/100M-DB/ ./db_bench -benchmarks=filluniquerandom,waitforcompaction -write_buffer_size=1048576 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -compression_type=none -num=1638400 -key_size=8 -value_size=56` - run command: `TEST_TMPDIR=/tmp/100M-DB/ ./db_bench -benchmarks=readrandom -use_existing_db=true -write_buffer_size=1048576 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -compression_type=none -num=1638400 -key_size=8 -value_size=56 -duration=10 -threads=32 -cache_size=131072 -statistics=true -verify_checksum=false -open_files=20 -cache_index_and_filter_blocks=true` - before: `rocksdb.block.checksum.compute.count COUNT : 384353` - after: `rocksdb.block.checksum.compute.count COUNT : 22` Performance: - Setup command (tmpfs, 128MB logical data size, cache indexes/filters without pinning so index/filter lookups go through table reader): `TEST_TMPDIR=/dev/shm/128M-DB/ ./db_bench -benchmarks=filluniquerandom,waitforcompaction -write_buffer_size=131072 -target_file_size_base=131072 -max_bytes_for_level_base=524288 -compression_type=none -num=4194304 -key_size=8 -value_size=24 -bloom_bits=8 -whole_key_filtering=1` - Measured point lookup performance. Database is fully cached to emphasize any new callstack overheads - Command: `TEST_TMPDIR=/dev/shm/128M-DB/ ./db_bench -benchmarks=readrandom[-W1][-X20] -use_existing_db=true -cache_index_and_filter_blocks=true -disable_auto_compactions=true -num=4194304 -key_size=8 -value_size=24 -bloom_bits=8 -whole_key_filtering=1 -duration=10 -cache_size=1048576000` - Before: `readrandom [AVG 20 runs] : 274848 (± 3717) ops/sec; 8.4 (± 0.1) MB/sec` - After: `readrandom [AVG 20 runs] : 277904 (± 4474) ops/sec; 8.5 (± 0.1) MB/sec` Reviewed By: hx235 Differential Revision: D43145366 Pulled By: ajkr fbshipit-source-id: 75ec062ece86a82cd788783de9de2c72df57f994
74 lines
2.5 KiB
C++
74 lines
2.5 KiB
C++
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
|
// This source code is licensed under both the GPLv2 (found in the
|
|
// COPYING file in the root directory) and Apache 2.0 License
|
|
// (found in the LICENSE.Apache file in the root directory).
|
|
//
|
|
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file. See the AUTHORS file for names of contributors.
|
|
#include "table/block_based/binary_search_index_reader.h"
|
|
|
|
namespace ROCKSDB_NAMESPACE {
|
|
Status BinarySearchIndexReader::Create(
|
|
const BlockBasedTable* table, const ReadOptions& ro,
|
|
FilePrefetchBuffer* prefetch_buffer, bool use_cache, bool prefetch,
|
|
bool pin, BlockCacheLookupContext* lookup_context,
|
|
std::unique_ptr<IndexReader>* index_reader) {
|
|
assert(table != nullptr);
|
|
assert(table->get_rep());
|
|
assert(!pin || prefetch);
|
|
assert(index_reader != nullptr);
|
|
|
|
CachableEntry<Block> index_block;
|
|
if (prefetch || !use_cache) {
|
|
const Status s =
|
|
ReadIndexBlock(table, prefetch_buffer, ro, use_cache,
|
|
/*get_context=*/nullptr, lookup_context, &index_block);
|
|
if (!s.ok()) {
|
|
return s;
|
|
}
|
|
|
|
if (use_cache && !pin) {
|
|
index_block.Reset();
|
|
}
|
|
}
|
|
|
|
index_reader->reset(
|
|
new BinarySearchIndexReader(table, std::move(index_block)));
|
|
|
|
return Status::OK();
|
|
}
|
|
|
|
InternalIteratorBase<IndexValue>* BinarySearchIndexReader::NewIterator(
|
|
const ReadOptions& read_options, bool /* disable_prefix_seek */,
|
|
IndexBlockIter* iter, GetContext* get_context,
|
|
BlockCacheLookupContext* lookup_context) {
|
|
const BlockBasedTable::Rep* rep = table()->get_rep();
|
|
const bool no_io = (read_options.read_tier == kBlockCacheTier);
|
|
CachableEntry<Block> index_block;
|
|
const Status s = GetOrReadIndexBlock(no_io, get_context, lookup_context,
|
|
&index_block, read_options);
|
|
if (!s.ok()) {
|
|
if (iter != nullptr) {
|
|
iter->Invalidate(s);
|
|
return iter;
|
|
}
|
|
|
|
return NewErrorInternalIterator<IndexValue>(s);
|
|
}
|
|
|
|
Statistics* kNullStats = nullptr;
|
|
// We don't return pinned data from index blocks, so no need
|
|
// to set `block_contents_pinned`.
|
|
auto it = index_block.GetValue()->NewIndexIterator(
|
|
internal_comparator()->user_comparator(),
|
|
rep->get_global_seqno(BlockType::kIndex), iter, kNullStats, true,
|
|
index_has_first_key(), index_key_includes_seq(), index_value_is_full());
|
|
|
|
assert(it != nullptr);
|
|
index_block.TransferTo(it);
|
|
|
|
return it;
|
|
}
|
|
} // namespace ROCKSDB_NAMESPACE
|