rocksdb/table/block.h

381 lines
13 KiB
C
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same 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.
#pragma once
#include <stddef.h>
#include <stdint.h>
Optimize BlockIter::Prev() by caching decoded entries Summary: Right now the way we do BlockIter::Prev() is like this - Go to the beginning of the restart interval - Keep moving forward (and decoding keys using ParseNextKey()) until we reach the desired key This can be optimized by caching the decoded entries in the first pass and reusing them in consecutive BlockIter::Prev() calls Before caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2423972 ops/sec; 268.2 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2413867 ops/sec; 267.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.410 micros/op 2440881 ops/sec; 270.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2417298 ops/sec; 267.4 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2421682 ops/sec; 267.9 MB/s ``` After caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.324 micros/op 3088955 ops/sec; 341.7 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.335 micros/op 2980999 ops/sec; 329.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.341 micros/op 2929681 ops/sec; 324.1 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.344 micros/op 2908490 ops/sec; 321.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.338 micros/op 2958404 ops/sec; 327.3 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: andrewkr, yiwu, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, yoshinorim Differential Revision: https://reviews.facebook.net/D59463
2016-06-14 21:27:46 +02:00
#include <string>
#include <vector>
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
#include <malloc.h>
#endif
#include "db/dbformat.h"
#include "db/pinned_iterators_manager.h"
#include "rocksdb/iterator.h"
#include "rocksdb/options.h"
#include "rocksdb/statistics.h"
#include "table/block_prefix_index.h"
#include "table/internal_iterator.h"
#include "format.h"
namespace rocksdb {
struct BlockContents;
class Comparator;
class BlockIter;
class BlockPrefixIndex;
// BlockReadAmpBitmap is a bitmap that map the rocksdb::Block data bytes to
// a bitmap with ratio bytes_per_bit. Whenever we access a range of bytes in
// the Block we update the bitmap and increment READ_AMP_ESTIMATE_USEFUL_BYTES.
class BlockReadAmpBitmap {
public:
explicit BlockReadAmpBitmap(size_t block_size, size_t bytes_per_bit,
Statistics* statistics)
: bitmap_(nullptr), bytes_per_bit_pow_(0), statistics_(statistics) {
assert(block_size > 0 && bytes_per_bit > 0);
// convert bytes_per_bit to be a power of 2
while (bytes_per_bit >>= 1) {
bytes_per_bit_pow_++;
}
// num_bits_needed = ceil(block_size / bytes_per_bit)
size_t num_bits_needed = (block_size >> bytes_per_bit_pow_) +
(block_size % (1 << bytes_per_bit_pow_) != 0);
// bitmap_size = ceil(num_bits_needed / kBitsPerEntry)
size_t bitmap_size = (num_bits_needed / kBitsPerEntry) +
(num_bits_needed % kBitsPerEntry != 0);
// Create bitmap and set all the bits to 0
bitmap_ = new std::atomic<uint32_t>[bitmap_size];
memset(bitmap_, 0, bitmap_size * kBytesPersEntry);
RecordTick(GetStatistics(), READ_AMP_TOTAL_READ_BYTES,
num_bits_needed << bytes_per_bit_pow_);
}
~BlockReadAmpBitmap() { delete[] bitmap_; }
void Mark(uint32_t start_offset, uint32_t end_offset) {
assert(end_offset >= start_offset);
// Every new bit we set will bump this counter
uint32_t new_useful_bytes = 0;
// Index of first bit in mask (start_offset / bytes_per_bit)
uint32_t start_bit = start_offset >> bytes_per_bit_pow_;
// Index of last bit in mask (end_offset / bytes_per_bit)
uint32_t end_bit = end_offset >> bytes_per_bit_pow_;
// Index of middle bit (unique to this range)
uint32_t mid_bit = start_bit + 1;
// It's guaranteed that ranges sent to Mark() wont overlap, this mean that
// we dont need to set the middle bits, we can simply set only one bit of
// the middle bits, and check this bit if we want to know if the whole
// range is set or not.
if (mid_bit < end_bit) {
if (GetAndSet(mid_bit) == 0) {
new_useful_bytes += (end_bit - mid_bit) << bytes_per_bit_pow_;
} else {
// If the middle bit is set, it's guaranteed that start and end bits
// are also set
return;
}
} else {
// This range dont have a middle bit, the whole range fall in 1 or 2 bits
}
if (GetAndSet(start_bit) == 0) {
new_useful_bytes += (1 << bytes_per_bit_pow_);
}
if (GetAndSet(end_bit) == 0) {
new_useful_bytes += (1 << bytes_per_bit_pow_);
}
if (new_useful_bytes > 0) {
RecordTick(GetStatistics(), READ_AMP_ESTIMATE_USEFUL_BYTES,
new_useful_bytes);
}
}
Statistics* GetStatistics() {
return statistics_.load(std::memory_order_relaxed);
}
void SetStatistics(Statistics* stats) { statistics_.store(stats); }
uint32_t GetBytesPerBit() { return 1 << bytes_per_bit_pow_; }
private:
// Get the current value of bit at `bit_idx` and set it to 1
inline bool GetAndSet(uint32_t bit_idx) {
const uint32_t byte_idx = bit_idx / kBitsPerEntry;
const uint32_t bit_mask = 1 << (bit_idx % kBitsPerEntry);
return bitmap_[byte_idx].fetch_or(bit_mask, std::memory_order_relaxed) &
bit_mask;
}
const uint32_t kBytesPersEntry = sizeof(uint32_t); // 4 bytes
const uint32_t kBitsPerEntry = kBytesPersEntry * 8; // 32 bits
// Bitmap used to record the bytes that we read, use atomic to protect
// against multiple threads updating the same bit
std::atomic<uint32_t>* bitmap_;
// (1 << bytes_per_bit_pow_) is bytes_per_bit. Use power of 2 to optimize
// muliplication and division
uint8_t bytes_per_bit_pow_;
// Pointer to DB Statistics object, Since this bitmap may outlive the DB
// this pointer maybe invalid, but the DB will update it to a valid pointer
// by using SetStatistics() before calling Mark()
std::atomic<Statistics*> statistics_;
};
class Block {
public:
// Initialize the block with the specified contents.
explicit Block(BlockContents&& contents, size_t read_amp_bytes_per_bit = 0,
Statistics* statistics = nullptr);
~Block() = default;
size_t size() const { return size_; }
const char* data() const { return data_; }
bool cachable() const { return contents_.cachable; }
size_t usable_size() const {
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
if (contents_.allocation.get() != nullptr) {
return malloc_usable_size(contents_.allocation.get());
}
#endif // ROCKSDB_MALLOC_USABLE_SIZE
return size_;
}
uint32_t NumRestarts() const;
2014-09-18 00:08:19 +02:00
CompressionType compression_type() const {
return contents_.compression_type;
}
// If hash index lookup is enabled and `use_hash_index` is true. This block
// will do hash lookup for the key prefix.
//
// NOTE: for the hash based lookup, if a key prefix doesn't match any key,
// the iterator will simply be set as "invalid", rather than returning
// the key that is just pass the target key.
//
// If iter is null, return new Iterator
// If iter is not null, update this one and return it as Iterator*
//
// If total_order_seek is true, hash_index_ and prefix_index_ are ignored.
// This option only applies for index block. For data block, hash_index_
// and prefix_index_ are null, so this option does not matter.
InternalIterator* NewIterator(const Comparator* comparator,
BlockIter* iter = nullptr,
bool total_order_seek = true,
Statistics* stats = nullptr);
void SetBlockPrefixIndex(BlockPrefixIndex* prefix_index);
// Report an approximation of how much memory has been used.
size_t ApproximateMemoryUsage() const;
private:
BlockContents contents_;
const char* data_; // contents_.data.data()
size_t size_; // contents_.data.size()
uint32_t restart_offset_; // Offset in data_ of restart array
std::unique_ptr<BlockPrefixIndex> prefix_index_;
std::unique_ptr<BlockReadAmpBitmap> read_amp_bitmap_;
// No copying allowed
Block(const Block&);
void operator=(const Block&);
};
class BlockIter : public InternalIterator {
public:
BlockIter()
: comparator_(nullptr),
data_(nullptr),
restarts_(0),
num_restarts_(0),
current_(0),
restart_index_(0),
status_(Status::OK()),
Optimize BlockIter::Prev() by caching decoded entries Summary: Right now the way we do BlockIter::Prev() is like this - Go to the beginning of the restart interval - Keep moving forward (and decoding keys using ParseNextKey()) until we reach the desired key This can be optimized by caching the decoded entries in the first pass and reusing them in consecutive BlockIter::Prev() calls Before caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2423972 ops/sec; 268.2 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2413867 ops/sec; 267.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.410 micros/op 2440881 ops/sec; 270.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2417298 ops/sec; 267.4 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2421682 ops/sec; 267.9 MB/s ``` After caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.324 micros/op 3088955 ops/sec; 341.7 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.335 micros/op 2980999 ops/sec; 329.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.341 micros/op 2929681 ops/sec; 324.1 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.344 micros/op 2908490 ops/sec; 321.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.338 micros/op 2958404 ops/sec; 327.3 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: andrewkr, yiwu, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, yoshinorim Differential Revision: https://reviews.facebook.net/D59463
2016-06-14 21:27:46 +02:00
prefix_index_(nullptr),
key_pinned_(false),
read_amp_bitmap_(nullptr),
last_bitmap_offset_(0) {}
BlockIter(const Comparator* comparator, const char* data, uint32_t restarts,
uint32_t num_restarts, BlockPrefixIndex* prefix_index,
BlockReadAmpBitmap* read_amp_bitmap)
: BlockIter() {
Initialize(comparator, data, restarts, num_restarts, prefix_index,
read_amp_bitmap);
}
void Initialize(const Comparator* comparator, const char* data,
uint32_t restarts, uint32_t num_restarts,
BlockPrefixIndex* prefix_index,
BlockReadAmpBitmap* read_amp_bitmap) {
assert(data_ == nullptr); // Ensure it is called only once
assert(num_restarts > 0); // Ensure the param is valid
comparator_ = comparator;
data_ = data;
restarts_ = restarts;
num_restarts_ = num_restarts;
current_ = restarts_;
restart_index_ = num_restarts_;
prefix_index_ = prefix_index;
read_amp_bitmap_ = read_amp_bitmap;
last_bitmap_offset_ = current_ + 1;
}
void SetStatus(Status s) {
status_ = s;
}
virtual bool Valid() const override { return current_ < restarts_; }
virtual Status status() const override { return status_; }
virtual Slice key() const override {
assert(Valid());
return key_.GetKey();
}
virtual Slice value() const override {
assert(Valid());
if (read_amp_bitmap_ && current_ < restarts_ &&
current_ != last_bitmap_offset_) {
read_amp_bitmap_->Mark(current_ /* current entry offset */,
NextEntryOffset() - 1);
last_bitmap_offset_ = current_;
}
return value_;
}
virtual void Next() override;
virtual void Prev() override;
virtual void Seek(const Slice& target) override;
virtual void SeekToFirst() override;
virtual void SeekToLast() override;
#ifndef NDEBUG
~BlockIter() {
// Assert that the BlockIter is never deleted while Pinning is Enabled.
assert(!pinned_iters_mgr_ ||
(pinned_iters_mgr_ && !pinned_iters_mgr_->PinningEnabled()));
Introduce ReadOptions::pin_data (support zero copy for keys) Summary: This patch update the Iterator API to introduce new functions that allow users to keep the Slices returned by key() valid as long as the Iterator is not deleted ReadOptions::pin_data : If true keep loaded blocks in memory as long as the iterator is not deleted Iterator::IsKeyPinned() : If true, this mean that the Slice returned by key() is valid as long as the iterator is not deleted Also add a new option BlockBasedTableOptions::use_delta_encoding to allow users to disable delta_encoding if needed. Benchmark results (using https://phabricator.fb.com/P20083553) ``` // $ du -h /home/tec/local/normal.4K.Snappy/db10077 // 6.1G /home/tec/local/normal.4K.Snappy/db10077 // $ du -h /home/tec/local/zero.8K.LZ4/db10077 // 6.4G /home/tec/local/zero.8K.LZ4/db10077 // Benchmarks for shard db10077 // _build/opt/rocks/benchmark/rocks_copy_benchmark \ // --normal_db_path="/home/tec/local/normal.4K.Snappy/db10077" \ // --zero_db_path="/home/tec/local/zero.8K.LZ4/db10077" // First run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 1.73s 576.97m // BM_StringPiece 103.74% 1.67s 598.55m // ============================================================================ // Match rate : 1000000 / 1000000 // Second run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 611.99ms 1.63 // BM_StringPiece 203.76% 300.35ms 3.33 // ============================================================================ // Match rate : 1000000 / 1000000 ``` Test Plan: Unit tests Reviewers: sdong, igor, anthony, yhchiang, rven Reviewed By: rven Subscribers: dhruba, lovro, adsharma Differential Revision: https://reviews.facebook.net/D48999
2015-12-16 21:08:30 +01:00
}
virtual void SetPinnedItersMgr(
PinnedIteratorsManager* pinned_iters_mgr) override {
pinned_iters_mgr_ = pinned_iters_mgr;
Introduce ReadOptions::pin_data (support zero copy for keys) Summary: This patch update the Iterator API to introduce new functions that allow users to keep the Slices returned by key() valid as long as the Iterator is not deleted ReadOptions::pin_data : If true keep loaded blocks in memory as long as the iterator is not deleted Iterator::IsKeyPinned() : If true, this mean that the Slice returned by key() is valid as long as the iterator is not deleted Also add a new option BlockBasedTableOptions::use_delta_encoding to allow users to disable delta_encoding if needed. Benchmark results (using https://phabricator.fb.com/P20083553) ``` // $ du -h /home/tec/local/normal.4K.Snappy/db10077 // 6.1G /home/tec/local/normal.4K.Snappy/db10077 // $ du -h /home/tec/local/zero.8K.LZ4/db10077 // 6.4G /home/tec/local/zero.8K.LZ4/db10077 // Benchmarks for shard db10077 // _build/opt/rocks/benchmark/rocks_copy_benchmark \ // --normal_db_path="/home/tec/local/normal.4K.Snappy/db10077" \ // --zero_db_path="/home/tec/local/zero.8K.LZ4/db10077" // First run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 1.73s 576.97m // BM_StringPiece 103.74% 1.67s 598.55m // ============================================================================ // Match rate : 1000000 / 1000000 // Second run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 611.99ms 1.63 // BM_StringPiece 203.76% 300.35ms 3.33 // ============================================================================ // Match rate : 1000000 / 1000000 ``` Test Plan: Unit tests Reviewers: sdong, igor, anthony, yhchiang, rven Reviewed By: rven Subscribers: dhruba, lovro, adsharma Differential Revision: https://reviews.facebook.net/D48999
2015-12-16 21:08:30 +01:00
}
PinnedIteratorsManager* pinned_iters_mgr_ = nullptr;
#endif
Introduce ReadOptions::pin_data (support zero copy for keys) Summary: This patch update the Iterator API to introduce new functions that allow users to keep the Slices returned by key() valid as long as the Iterator is not deleted ReadOptions::pin_data : If true keep loaded blocks in memory as long as the iterator is not deleted Iterator::IsKeyPinned() : If true, this mean that the Slice returned by key() is valid as long as the iterator is not deleted Also add a new option BlockBasedTableOptions::use_delta_encoding to allow users to disable delta_encoding if needed. Benchmark results (using https://phabricator.fb.com/P20083553) ``` // $ du -h /home/tec/local/normal.4K.Snappy/db10077 // 6.1G /home/tec/local/normal.4K.Snappy/db10077 // $ du -h /home/tec/local/zero.8K.LZ4/db10077 // 6.4G /home/tec/local/zero.8K.LZ4/db10077 // Benchmarks for shard db10077 // _build/opt/rocks/benchmark/rocks_copy_benchmark \ // --normal_db_path="/home/tec/local/normal.4K.Snappy/db10077" \ // --zero_db_path="/home/tec/local/zero.8K.LZ4/db10077" // First run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 1.73s 576.97m // BM_StringPiece 103.74% 1.67s 598.55m // ============================================================================ // Match rate : 1000000 / 1000000 // Second run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 611.99ms 1.63 // BM_StringPiece 203.76% 300.35ms 3.33 // ============================================================================ // Match rate : 1000000 / 1000000 ``` Test Plan: Unit tests Reviewers: sdong, igor, anthony, yhchiang, rven Reviewed By: rven Subscribers: dhruba, lovro, adsharma Differential Revision: https://reviews.facebook.net/D48999
2015-12-16 21:08:30 +01:00
Optimize BlockIter::Prev() by caching decoded entries Summary: Right now the way we do BlockIter::Prev() is like this - Go to the beginning of the restart interval - Keep moving forward (and decoding keys using ParseNextKey()) until we reach the desired key This can be optimized by caching the decoded entries in the first pass and reusing them in consecutive BlockIter::Prev() calls Before caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2423972 ops/sec; 268.2 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2413867 ops/sec; 267.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.410 micros/op 2440881 ops/sec; 270.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2417298 ops/sec; 267.4 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2421682 ops/sec; 267.9 MB/s ``` After caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.324 micros/op 3088955 ops/sec; 341.7 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.335 micros/op 2980999 ops/sec; 329.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.341 micros/op 2929681 ops/sec; 324.1 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.344 micros/op 2908490 ops/sec; 321.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.338 micros/op 2958404 ops/sec; 327.3 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: andrewkr, yiwu, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, yoshinorim Differential Revision: https://reviews.facebook.net/D59463
2016-06-14 21:27:46 +02:00
virtual bool IsKeyPinned() const override { return key_pinned_; }
Introduce ReadOptions::pin_data (support zero copy for keys) Summary: This patch update the Iterator API to introduce new functions that allow users to keep the Slices returned by key() valid as long as the Iterator is not deleted ReadOptions::pin_data : If true keep loaded blocks in memory as long as the iterator is not deleted Iterator::IsKeyPinned() : If true, this mean that the Slice returned by key() is valid as long as the iterator is not deleted Also add a new option BlockBasedTableOptions::use_delta_encoding to allow users to disable delta_encoding if needed. Benchmark results (using https://phabricator.fb.com/P20083553) ``` // $ du -h /home/tec/local/normal.4K.Snappy/db10077 // 6.1G /home/tec/local/normal.4K.Snappy/db10077 // $ du -h /home/tec/local/zero.8K.LZ4/db10077 // 6.4G /home/tec/local/zero.8K.LZ4/db10077 // Benchmarks for shard db10077 // _build/opt/rocks/benchmark/rocks_copy_benchmark \ // --normal_db_path="/home/tec/local/normal.4K.Snappy/db10077" \ // --zero_db_path="/home/tec/local/zero.8K.LZ4/db10077" // First run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 1.73s 576.97m // BM_StringPiece 103.74% 1.67s 598.55m // ============================================================================ // Match rate : 1000000 / 1000000 // Second run // ============================================================================ // rocks/benchmark/RocksCopyBenchmark.cpp relative time/iter iters/s // ============================================================================ // BM_StringCopy 611.99ms 1.63 // BM_StringPiece 203.76% 300.35ms 3.33 // ============================================================================ // Match rate : 1000000 / 1000000 ``` Test Plan: Unit tests Reviewers: sdong, igor, anthony, yhchiang, rven Reviewed By: rven Subscribers: dhruba, lovro, adsharma Differential Revision: https://reviews.facebook.net/D48999
2015-12-16 21:08:30 +01:00
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
2016-07-20 18:49:03 +02:00
virtual bool IsValuePinned() const override { return true; }
size_t TEST_CurrentEntrySize() { return NextEntryOffset() - current_; }
private:
const Comparator* comparator_;
const char* data_; // underlying block contents
uint32_t restarts_; // Offset of restart array (list of fixed32)
uint32_t num_restarts_; // Number of uint32_t entries in restart array
// current_ is offset in data_ of current entry. >= restarts_ if !Valid
uint32_t current_;
uint32_t restart_index_; // Index of restart block in which current_ falls
IterKey key_;
Slice value_;
Status status_;
BlockPrefixIndex* prefix_index_;
Optimize BlockIter::Prev() by caching decoded entries Summary: Right now the way we do BlockIter::Prev() is like this - Go to the beginning of the restart interval - Keep moving forward (and decoding keys using ParseNextKey()) until we reach the desired key This can be optimized by caching the decoded entries in the first pass and reusing them in consecutive BlockIter::Prev() calls Before caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2423972 ops/sec; 268.2 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2413867 ops/sec; 267.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.410 micros/op 2440881 ops/sec; 270.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2417298 ops/sec; 267.4 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2421682 ops/sec; 267.9 MB/s ``` After caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.324 micros/op 3088955 ops/sec; 341.7 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.335 micros/op 2980999 ops/sec; 329.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.341 micros/op 2929681 ops/sec; 324.1 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.344 micros/op 2908490 ops/sec; 321.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.338 micros/op 2958404 ops/sec; 327.3 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: andrewkr, yiwu, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, yoshinorim Differential Revision: https://reviews.facebook.net/D59463
2016-06-14 21:27:46 +02:00
bool key_pinned_;
// read-amp bitmap
BlockReadAmpBitmap* read_amp_bitmap_;
// last `current_` value we report to read-amp bitmp
mutable uint32_t last_bitmap_offset_;
Optimize BlockIter::Prev() by caching decoded entries Summary: Right now the way we do BlockIter::Prev() is like this - Go to the beginning of the restart interval - Keep moving forward (and decoding keys using ParseNextKey()) until we reach the desired key This can be optimized by caching the decoded entries in the first pass and reusing them in consecutive BlockIter::Prev() calls Before caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2423972 ops/sec; 268.2 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2413867 ops/sec; 267.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.410 micros/op 2440881 ops/sec; 270.0 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.414 micros/op 2417298 ops/sec; 267.4 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.413 micros/op 2421682 ops/sec; 267.9 MB/s ``` After caching ``` DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readreverse" --db="/dev/shm/bench_prev_opt/" --use_existing_db --disable_auto_compactions DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.324 micros/op 3088955 ops/sec; 341.7 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.335 micros/op 2980999 ops/sec; 329.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.341 micros/op 2929681 ops/sec; 324.1 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.344 micros/op 2908490 ops/sec; 321.8 MB/s DB path: [/dev/shm/bench_prev_opt/] readreverse : 0.338 micros/op 2958404 ops/sec; 327.3 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: andrewkr, yiwu, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba, yoshinorim Differential Revision: https://reviews.facebook.net/D59463
2016-06-14 21:27:46 +02:00
struct CachedPrevEntry {
explicit CachedPrevEntry(uint32_t _offset, const char* _key_ptr,
size_t _key_offset, size_t _key_size, Slice _value)
: offset(_offset),
key_ptr(_key_ptr),
key_offset(_key_offset),
key_size(_key_size),
value(_value) {}
// offset of entry in block
uint32_t offset;
// Pointer to key data in block (nullptr if key is delta-encoded)
const char* key_ptr;
// offset of key in prev_entries_keys_buff_ (0 if key_ptr is not nullptr)
size_t key_offset;
// size of key
size_t key_size;
// value slice pointing to data in block
Slice value;
};
std::string prev_entries_keys_buff_;
std::vector<CachedPrevEntry> prev_entries_;
int32_t prev_entries_idx_ = -1;
inline int Compare(const Slice& a, const Slice& b) const {
return comparator_->Compare(a, b);
}
// Return the offset in data_ just past the end of the current entry.
inline uint32_t NextEntryOffset() const {
// NOTE: We don't support files bigger than 2GB
return static_cast<uint32_t>((value_.data() + value_.size()) - data_);
}
uint32_t GetRestartPoint(uint32_t index) {
assert(index < num_restarts_);
return DecodeFixed32(data_ + restarts_ + index * sizeof(uint32_t));
}
void SeekToRestartPoint(uint32_t index) {
key_.Clear();
restart_index_ = index;
// current_ will be fixed by ParseNextKey();
// ParseNextKey() starts at the end of value_, so set value_ accordingly
uint32_t offset = GetRestartPoint(index);
value_ = Slice(data_ + offset, 0);
}
void CorruptionError();
bool ParseNextKey();
bool BinarySeek(const Slice& target, uint32_t left, uint32_t right,
uint32_t* index);
int CompareBlockKey(uint32_t block_index, const Slice& target);
bool BinaryBlockIndexSeek(const Slice& target, uint32_t* block_ids,
uint32_t left, uint32_t right,
uint32_t* index);
bool PrefixSeek(const Slice& target, uint32_t* index);
};
} // namespace rocksdb