rocksdb/table/block.cc

356 lines
11 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.
//
// Decodes the blocks generated by block_builder.cc.
#include "table/block.h"
#include <algorithm>
#include <string>
#include <unordered_map>
#include <vector>
#include "rocksdb/comparator.h"
#include "table/format.h"
#include "table/block_prefix_index.h"
#include "util/coding.h"
#include "util/logging.h"
#include "util/perf_context_imp.h"
namespace rocksdb {
// Helper routine: decode the next block entry starting at "p",
// storing the number of shared key bytes, non_shared key bytes,
// and the length of the value in "*shared", "*non_shared", and
// "*value_length", respectively. Will not derefence past "limit".
//
// If any errors are detected, returns nullptr. Otherwise, returns a
// pointer to the key delta (just past the three decoded values).
static inline const char* DecodeEntry(const char* p, const char* limit,
uint32_t* shared,
uint32_t* non_shared,
uint32_t* value_length) {
if (limit - p < 3) return nullptr;
*shared = reinterpret_cast<const unsigned char*>(p)[0];
*non_shared = reinterpret_cast<const unsigned char*>(p)[1];
*value_length = reinterpret_cast<const unsigned char*>(p)[2];
if ((*shared | *non_shared | *value_length) < 128) {
// Fast path: all three values are encoded in one byte each
p += 3;
} else {
if ((p = GetVarint32Ptr(p, limit, shared)) == nullptr) return nullptr;
if ((p = GetVarint32Ptr(p, limit, non_shared)) == nullptr) return nullptr;
if ((p = GetVarint32Ptr(p, limit, value_length)) == nullptr) return nullptr;
}
if (static_cast<uint32_t>(limit - p) < (*non_shared + *value_length)) {
return nullptr;
}
return p;
}
void BlockIter::Next() {
assert(Valid());
ParseNextKey();
}
void BlockIter::Prev() {
assert(Valid());
// Scan backwards to a restart point before current_
const uint32_t original = current_;
while (GetRestartPoint(restart_index_) >= original) {
if (restart_index_ == 0) {
// No more entries
current_ = restarts_;
restart_index_ = num_restarts_;
return;
}
restart_index_--;
}
SeekToRestartPoint(restart_index_);
do {
// Loop until end of current entry hits the start of original entry
} while (ParseNextKey() && NextEntryOffset() < original);
}
void BlockIter::Seek(const Slice& target) {
PERF_TIMER_GUARD(block_seek_nanos);
if (data_ == nullptr) { // Not init yet
return;
}
uint32_t index = 0;
bool ok = false;
if (prefix_index_) {
ok = PrefixSeek(target, &index);
} else {
ok = BinarySeek(target, 0, num_restarts_ - 1, &index);
}
if (!ok) {
return;
}
SeekToRestartPoint(index);
// Linear search (within restart block) for first key >= target
while (true) {
if (!ParseNextKey() || Compare(key_.GetKey(), target) >= 0) {
return;
}
}
}
void BlockIter::SeekToFirst() {
if (data_ == nullptr) { // Not init yet
return;
}
SeekToRestartPoint(0);
ParseNextKey();
}
void BlockIter::SeekToLast() {
if (data_ == nullptr) { // Not init yet
return;
}
SeekToRestartPoint(num_restarts_ - 1);
while (ParseNextKey() && NextEntryOffset() < restarts_) {
// Keep skipping
}
}
void BlockIter::CorruptionError() {
current_ = restarts_;
restart_index_ = num_restarts_;
status_ = Status::Corruption("bad entry in block");
key_.Clear();
value_.clear();
}
bool BlockIter::ParseNextKey() {
current_ = NextEntryOffset();
const char* p = data_ + current_;
const char* limit = data_ + restarts_; // Restarts come right after data
if (p >= limit) {
// No more entries to return. Mark as invalid.
current_ = restarts_;
restart_index_ = num_restarts_;
return false;
}
// Decode next entry
uint32_t shared, non_shared, value_length;
p = DecodeEntry(p, limit, &shared, &non_shared, &value_length);
if (p == nullptr || key_.Size() < shared) {
CorruptionError();
return false;
} else {
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
if (shared == 0) {
// If this key dont share any bytes with prev key then we dont need
// to decode it and can use it's address in the block directly.
key_.SetKey(Slice(p, non_shared), false /* copy */);
} else {
// This key share `shared` bytes with prev key, we need to decode it
key_.TrimAppend(shared, p, non_shared);
}
value_ = Slice(p + non_shared, value_length);
while (restart_index_ + 1 < num_restarts_ &&
GetRestartPoint(restart_index_ + 1) < current_) {
++restart_index_;
}
return true;
}
}
// Binary search in restart array to find the first restart point
// with a key >= target (TODO: this comment is inaccurate)
bool BlockIter::BinarySeek(const Slice& target, uint32_t left, uint32_t right,
uint32_t* index) {
assert(left <= right);
while (left < right) {
uint32_t mid = (left + right + 1) / 2;
uint32_t region_offset = GetRestartPoint(mid);
uint32_t shared, non_shared, value_length;
const char* key_ptr =
DecodeEntry(data_ + region_offset, data_ + restarts_, &shared,
&non_shared, &value_length);
if (key_ptr == nullptr || (shared != 0)) {
CorruptionError();
return false;
}
Slice mid_key(key_ptr, non_shared);
int cmp = Compare(mid_key, target);
if (cmp < 0) {
// Key at "mid" is smaller than "target". Therefore all
// blocks before "mid" are uninteresting.
left = mid;
} else if (cmp > 0) {
// Key at "mid" is >= "target". Therefore all blocks at or
// after "mid" are uninteresting.
right = mid - 1;
} else {
left = right = mid;
}
}
*index = left;
return true;
}
// Compare target key and the block key of the block of `block_index`.
// Return -1 if error.
int BlockIter::CompareBlockKey(uint32_t block_index, const Slice& target) {
uint32_t region_offset = GetRestartPoint(block_index);
uint32_t shared, non_shared, value_length;
const char* key_ptr = DecodeEntry(data_ + region_offset, data_ + restarts_,
&shared, &non_shared, &value_length);
if (key_ptr == nullptr || (shared != 0)) {
CorruptionError();
return 1; // Return target is smaller
}
Slice block_key(key_ptr, non_shared);
return Compare(block_key, target);
}
// Binary search in block_ids to find the first block
// with a key >= target
bool BlockIter::BinaryBlockIndexSeek(const Slice& target, uint32_t* block_ids,
uint32_t left, uint32_t right,
uint32_t* index) {
assert(left <= right);
uint32_t left_bound = left;
while (left <= right) {
uint32_t mid = (left + right) / 2;
int cmp = CompareBlockKey(block_ids[mid], target);
if (!status_.ok()) {
return false;
}
if (cmp < 0) {
// Key at "target" is larger than "mid". Therefore all
// blocks before or at "mid" are uninteresting.
left = mid + 1;
} else {
// Key at "target" is <= "mid". Therefore all blocks
// after "mid" are uninteresting.
// If there is only one block left, we found it.
if (left == right) break;
right = mid;
}
}
if (left == right) {
// In one of the two following cases:
// (1) left is the first one of block_ids
// (2) there is a gap of blocks between block of `left` and `left-1`.
// we can further distinguish the case of key in the block or key not
// existing, by comparing the target key and the key of the previous
// block to the left of the block found.
if (block_ids[left] > 0 &&
(left == left_bound || block_ids[left - 1] != block_ids[left] - 1) &&
CompareBlockKey(block_ids[left] - 1, target) > 0) {
current_ = restarts_;
return false;
}
*index = block_ids[left];
return true;
} else {
assert(left > right);
// Mark iterator invalid
current_ = restarts_;
return false;
}
}
bool BlockIter::PrefixSeek(const Slice& target, uint32_t* index) {
assert(prefix_index_);
uint32_t* block_ids = nullptr;
uint32_t num_blocks = prefix_index_->GetBlocks(target, &block_ids);
if (num_blocks == 0) {
current_ = restarts_;
return false;
} else {
return BinaryBlockIndexSeek(target, block_ids, 0, num_blocks - 1, index);
}
}
uint32_t Block::NumRestarts() const {
assert(size_ >= 2*sizeof(uint32_t));
return DecodeFixed32(data_ + size_ - sizeof(uint32_t));
}
Block::Block(BlockContents&& contents)
: contents_(std::move(contents)),
data_(contents_.data.data()),
size_(contents_.data.size()) {
if (size_ < sizeof(uint32_t)) {
size_ = 0; // Error marker
} else {
restart_offset_ =
static_cast<uint32_t>(size_) - (1 + NumRestarts()) * sizeof(uint32_t);
if (restart_offset_ > size_ - sizeof(uint32_t)) {
// The size is too small for NumRestarts() and therefore
// restart_offset_ wrapped around.
size_ = 0;
}
}
}
InternalIterator* Block::NewIterator(const Comparator* cmp, BlockIter* iter,
bool total_order_seek) {
if (size_ < 2*sizeof(uint32_t)) {
if (iter != nullptr) {
iter->SetStatus(Status::Corruption("bad block contents"));
return iter;
} else {
return NewErrorInternalIterator(Status::Corruption("bad block contents"));
}
}
const uint32_t num_restarts = NumRestarts();
if (num_restarts == 0) {
if (iter != nullptr) {
iter->SetStatus(Status::OK());
return iter;
} else {
return NewEmptyInternalIterator();
}
} else {
BlockPrefixIndex* prefix_index_ptr =
total_order_seek ? nullptr : prefix_index_.get();
if (iter != nullptr) {
iter->Initialize(cmp, data_, restart_offset_, num_restarts,
prefix_index_ptr);
} else {
iter = new BlockIter(cmp, data_, restart_offset_, num_restarts,
prefix_index_ptr);
}
}
return iter;
}
void Block::SetBlockPrefixIndex(BlockPrefixIndex* prefix_index) {
prefix_index_.reset(prefix_index);
}
size_t Block::ApproximateMemoryUsage() const {
size_t usage = usable_size();
if (prefix_index_) {
usage += prefix_index_->ApproximateMemoryUsage();
}
return usage;
}
} // namespace rocksdb