rocksdb/table/block_based/index_builder.cc

220 lines
9.1 KiB
C++
Raw Normal View History

// 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/index_builder.h"
#include <assert.h>
#include <cinttypes>
#include <list>
#include <string>
#include "rocksdb/comparator.h"
#include "rocksdb/flush_block_policy.h"
#include "table/block_based/partitioned_filter_block.h"
#include "table/format.h"
// Without anonymous namespace here, we fail the warning -Wmissing-prototypes
namespace rocksdb {
// using namespace rocksdb;
// Create a index builder based on its type.
IndexBuilder* IndexBuilder::CreateIndexBuilder(
BlockBasedTableOptions::IndexType index_type,
const InternalKeyComparator* comparator,
const InternalKeySliceTransform* int_key_slice_transform,
const bool use_value_delta_encoding,
const BlockBasedTableOptions& table_opt) {
IndexBuilder* result = nullptr;
switch (index_type) {
case BlockBasedTableOptions::kBinarySearch: {
result = new ShortenedIndexBuilder(
comparator, table_opt.index_block_restart_interval,
table_opt.format_version, use_value_delta_encoding,
Add an option to put first key of each sst block in the index (#5289) Summary: The first key is used to defer reading the data block until this file gets to the top of merging iterator's heap. For short range scans, most files never make it to the top of the heap, so this change can reduce read amplification by a lot sometimes. Consider the following workload. There are a few data streams (we'll be calling them "logs"), each stream consisting of a sequence of blobs (we'll be calling them "records"). Each record is identified by log ID and a sequence number within the log. RocksDB key is concatenation of log ID and sequence number (big endian). Reads are mostly relatively short range scans, each within a single log. Writes are mostly sequential for each log, but writes to different logs are randomly interleaved. Compactions are disabled; instead, when we accumulate a few tens of sst files, we create a new column family and start writing to it. So, a typical sst file consists of a few ranges of blocks, each range corresponding to one log ID (we use FlushBlockPolicy to cut blocks at log boundaries). A typical read would go like this. First, iterator Seek() reads one block from each sst file. Then a series of Next()s move through one sst file (since writes to each log are mostly sequential) until the subiterator reaches the end of this log in this sst file; then Next() switches to the next sst file and reads sequentially from that, and so on. Often a range scan will only return records from a small number of blocks in small number of sst files; in this case, the cost of initial Seek() reading one block from each file may be bigger than the cost of reading the actually useful blocks. Neither iterate_upper_bound nor bloom filters can prevent reading one block from each file in Seek(). But this PR can: if the index contains first key from each block, we don't have to read the block until this block actually makes it to the top of merging iterator's heap, so for short range scans we won't read any blocks from most of the sst files. This PR does the deferred block loading inside value() call. This is not ideal: there's no good way to report an IO error from inside value(). As discussed with siying offline, it would probably be better to change InternalIterator's interface to explicitly fetch deferred value and get status. I'll do it in a separate PR. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5289 Differential Revision: D15256423 Pulled By: al13n321 fbshipit-source-id: 750e4c39ce88e8d41662f701cf6275d9388ba46a
2019-06-25 03:50:35 +00:00
table_opt.index_shortening, /* include_first_key */ false);
} break;
case BlockBasedTableOptions::kHashSearch: {
result = new HashIndexBuilder(
comparator, int_key_slice_transform,
table_opt.index_block_restart_interval, table_opt.format_version,
use_value_delta_encoding, table_opt.index_shortening);
} break;
case BlockBasedTableOptions::kTwoLevelIndexSearch: {
result = PartitionedIndexBuilder::CreateIndexBuilder(
comparator, use_value_delta_encoding, table_opt);
} break;
Add an option to put first key of each sst block in the index (#5289) Summary: The first key is used to defer reading the data block until this file gets to the top of merging iterator's heap. For short range scans, most files never make it to the top of the heap, so this change can reduce read amplification by a lot sometimes. Consider the following workload. There are a few data streams (we'll be calling them "logs"), each stream consisting of a sequence of blobs (we'll be calling them "records"). Each record is identified by log ID and a sequence number within the log. RocksDB key is concatenation of log ID and sequence number (big endian). Reads are mostly relatively short range scans, each within a single log. Writes are mostly sequential for each log, but writes to different logs are randomly interleaved. Compactions are disabled; instead, when we accumulate a few tens of sst files, we create a new column family and start writing to it. So, a typical sst file consists of a few ranges of blocks, each range corresponding to one log ID (we use FlushBlockPolicy to cut blocks at log boundaries). A typical read would go like this. First, iterator Seek() reads one block from each sst file. Then a series of Next()s move through one sst file (since writes to each log are mostly sequential) until the subiterator reaches the end of this log in this sst file; then Next() switches to the next sst file and reads sequentially from that, and so on. Often a range scan will only return records from a small number of blocks in small number of sst files; in this case, the cost of initial Seek() reading one block from each file may be bigger than the cost of reading the actually useful blocks. Neither iterate_upper_bound nor bloom filters can prevent reading one block from each file in Seek(). But this PR can: if the index contains first key from each block, we don't have to read the block until this block actually makes it to the top of merging iterator's heap, so for short range scans we won't read any blocks from most of the sst files. This PR does the deferred block loading inside value() call. This is not ideal: there's no good way to report an IO error from inside value(). As discussed with siying offline, it would probably be better to change InternalIterator's interface to explicitly fetch deferred value and get status. I'll do it in a separate PR. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5289 Differential Revision: D15256423 Pulled By: al13n321 fbshipit-source-id: 750e4c39ce88e8d41662f701cf6275d9388ba46a
2019-06-25 03:50:35 +00:00
case BlockBasedTableOptions::kBinarySearchWithFirstKey: {
result = new ShortenedIndexBuilder(
comparator, table_opt.index_block_restart_interval,
table_opt.format_version, use_value_delta_encoding,
table_opt.index_shortening, /* include_first_key */ true);
} break;
default: {
assert(!"Do not recognize the index type ");
} break;
}
return result;
}
PartitionedIndexBuilder* PartitionedIndexBuilder::CreateIndexBuilder(
const InternalKeyComparator* comparator,
const bool use_value_delta_encoding,
const BlockBasedTableOptions& table_opt) {
return new PartitionedIndexBuilder(comparator, table_opt,
use_value_delta_encoding);
}
PartitionedIndexBuilder::PartitionedIndexBuilder(
const InternalKeyComparator* comparator,
const BlockBasedTableOptions& table_opt,
const bool use_value_delta_encoding)
: IndexBuilder(comparator),
index_block_builder_(table_opt.index_block_restart_interval,
true /*use_delta_encoding*/,
use_value_delta_encoding),
index_block_builder_without_seq_(table_opt.index_block_restart_interval,
true /*use_delta_encoding*/,
use_value_delta_encoding),
sub_index_builder_(nullptr),
table_opt_(table_opt),
// We start by false. After each partition we revise the value based on
// what the sub_index_builder has decided. If the feature is disabled
// entirely, this will be set to true after switching the first
// sub_index_builder. Otherwise, it could be set to true even one of the
// sub_index_builders could not safely exclude seq from the keys, then it
// wil be enforced on all sub_index_builders on ::Finish.
seperator_is_key_plus_seq_(false),
use_value_delta_encoding_(use_value_delta_encoding) {}
PartitionedIndexBuilder::~PartitionedIndexBuilder() {
delete sub_index_builder_;
}
void PartitionedIndexBuilder::MakeNewSubIndexBuilder() {
assert(sub_index_builder_ == nullptr);
sub_index_builder_ = new ShortenedIndexBuilder(
comparator_, table_opt_.index_block_restart_interval,
table_opt_.format_version, use_value_delta_encoding_,
Add an option to put first key of each sst block in the index (#5289) Summary: The first key is used to defer reading the data block until this file gets to the top of merging iterator's heap. For short range scans, most files never make it to the top of the heap, so this change can reduce read amplification by a lot sometimes. Consider the following workload. There are a few data streams (we'll be calling them "logs"), each stream consisting of a sequence of blobs (we'll be calling them "records"). Each record is identified by log ID and a sequence number within the log. RocksDB key is concatenation of log ID and sequence number (big endian). Reads are mostly relatively short range scans, each within a single log. Writes are mostly sequential for each log, but writes to different logs are randomly interleaved. Compactions are disabled; instead, when we accumulate a few tens of sst files, we create a new column family and start writing to it. So, a typical sst file consists of a few ranges of blocks, each range corresponding to one log ID (we use FlushBlockPolicy to cut blocks at log boundaries). A typical read would go like this. First, iterator Seek() reads one block from each sst file. Then a series of Next()s move through one sst file (since writes to each log are mostly sequential) until the subiterator reaches the end of this log in this sst file; then Next() switches to the next sst file and reads sequentially from that, and so on. Often a range scan will only return records from a small number of blocks in small number of sst files; in this case, the cost of initial Seek() reading one block from each file may be bigger than the cost of reading the actually useful blocks. Neither iterate_upper_bound nor bloom filters can prevent reading one block from each file in Seek(). But this PR can: if the index contains first key from each block, we don't have to read the block until this block actually makes it to the top of merging iterator's heap, so for short range scans we won't read any blocks from most of the sst files. This PR does the deferred block loading inside value() call. This is not ideal: there's no good way to report an IO error from inside value(). As discussed with siying offline, it would probably be better to change InternalIterator's interface to explicitly fetch deferred value and get status. I'll do it in a separate PR. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5289 Differential Revision: D15256423 Pulled By: al13n321 fbshipit-source-id: 750e4c39ce88e8d41662f701cf6275d9388ba46a
2019-06-25 03:50:35 +00:00
table_opt_.index_shortening, /* include_first_key */ false);
flush_policy_.reset(FlushBlockBySizePolicyFactory::NewFlushBlockPolicy(
table_opt_.metadata_block_size, table_opt_.block_size_deviation,
// Note: this is sub-optimal since sub_index_builder_ could later reset
// seperator_is_key_plus_seq_ but the probability of that is low.
sub_index_builder_->seperator_is_key_plus_seq_
? sub_index_builder_->index_block_builder_
: sub_index_builder_->index_block_builder_without_seq_));
partition_cut_requested_ = false;
}
void PartitionedIndexBuilder::RequestPartitionCut() {
partition_cut_requested_ = true;
}
void PartitionedIndexBuilder::AddIndexEntry(
std::string* last_key_in_current_block,
const Slice* first_key_in_next_block, const BlockHandle& block_handle) {
// Note: to avoid two consecuitive flush in the same method call, we do not
// check flush policy when adding the last key
if (UNLIKELY(first_key_in_next_block == nullptr)) { // no more keys
if (sub_index_builder_ == nullptr) {
MakeNewSubIndexBuilder();
}
sub_index_builder_->AddIndexEntry(last_key_in_current_block,
first_key_in_next_block, block_handle);
if (sub_index_builder_->seperator_is_key_plus_seq_) {
// then we need to apply it to all sub-index builders
seperator_is_key_plus_seq_ = true;
}
sub_index_last_key_ = std::string(*last_key_in_current_block);
entries_.push_back(
{sub_index_last_key_,
std::unique_ptr<ShortenedIndexBuilder>(sub_index_builder_)});
sub_index_builder_ = nullptr;
cut_filter_block = true;
} else {
// apply flush policy only to non-empty sub_index_builder_
if (sub_index_builder_ != nullptr) {
std::string handle_encoding;
block_handle.EncodeTo(&handle_encoding);
bool do_flush =
partition_cut_requested_ ||
flush_policy_->Update(*last_key_in_current_block, handle_encoding);
if (do_flush) {
entries_.push_back(
{sub_index_last_key_,
std::unique_ptr<ShortenedIndexBuilder>(sub_index_builder_)});
cut_filter_block = true;
sub_index_builder_ = nullptr;
}
}
if (sub_index_builder_ == nullptr) {
MakeNewSubIndexBuilder();
}
sub_index_builder_->AddIndexEntry(last_key_in_current_block,
first_key_in_next_block, block_handle);
sub_index_last_key_ = std::string(*last_key_in_current_block);
if (sub_index_builder_->seperator_is_key_plus_seq_) {
// then we need to apply it to all sub-index builders
seperator_is_key_plus_seq_ = true;
}
}
}
Status PartitionedIndexBuilder::Finish(
IndexBlocks* index_blocks, const BlockHandle& last_partition_block_handle) {
if (partition_cnt_ == 0) {
partition_cnt_ = entries_.size();
}
// It must be set to null after last key is added
assert(sub_index_builder_ == nullptr);
if (finishing_indexes == true) {
Entry& last_entry = entries_.front();
std::string handle_encoding;
last_partition_block_handle.EncodeTo(&handle_encoding);
std::string handle_delta_encoding;
PutVarsignedint64(
&handle_delta_encoding,
last_partition_block_handle.size() - last_encoded_handle_.size());
last_encoded_handle_ = last_partition_block_handle;
const Slice handle_delta_encoding_slice(handle_delta_encoding);
index_block_builder_.Add(last_entry.key, handle_encoding,
&handle_delta_encoding_slice);
if (!seperator_is_key_plus_seq_) {
index_block_builder_without_seq_.Add(ExtractUserKey(last_entry.key),
handle_encoding,
&handle_delta_encoding_slice);
}
entries_.pop_front();
}
// If there is no sub_index left, then return the 2nd level index.
if (UNLIKELY(entries_.empty())) {
if (seperator_is_key_plus_seq_) {
index_blocks->index_block_contents = index_block_builder_.Finish();
} else {
index_blocks->index_block_contents =
index_block_builder_without_seq_.Finish();
}
top_level_index_size_ = index_blocks->index_block_contents.size();
index_size_ += top_level_index_size_;
return Status::OK();
} else {
// Finish the next partition index in line and Incomplete() to indicate we
// expect more calls to Finish
Entry& entry = entries_.front();
// Apply the policy to all sub-indexes
entry.value->seperator_is_key_plus_seq_ = seperator_is_key_plus_seq_;
auto s = entry.value->Finish(index_blocks);
index_size_ += index_blocks->index_block_contents.size();
finishing_indexes = true;
return s.ok() ? Status::Incomplete() : s;
}
}
size_t PartitionedIndexBuilder::NumPartitions() const { return partition_cnt_; }
} // namespace rocksdb