rocksdb/table/table_builder.cc

442 lines
14 KiB
C++
Raw Normal View History

// Copyright (c) 2013, 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.
#include "rocksdb/table_builder.h"
#include <assert.h>
#include <map>
#include "rocksdb/comparator.h"
#include "rocksdb/env.h"
#include "rocksdb/filter_policy.h"
#include "rocksdb/options.h"
#include "table/block_builder.h"
#include "table/filter_block.h"
#include "table/format.h"
#include "table/table.h"
#include "util/coding.h"
#include "util/crc32c.h"
#include "util/stop_watch.h"
namespace rocksdb {
namespace {
struct BytewiseLessThan {
bool operator()(const std::string& key1, const std::string& key2) {
// smaller entries will be placed in front.
return comparator->Compare(key1, key2) <= 0;
}
const Comparator* comparator = BytewiseComparator();
};
// When writing to a block that requires entries to be sorted by
// `BytewiseComparator`, we can buffer the content to `BytewiseSortedMap`
// before writng to store.
typedef std::map<std::string, std::string, BytewiseLessThan> BytewiseSortedMap;
void AddStats(BytewiseSortedMap& stats, std::string name, uint64_t val) {
assert(stats.find(name) == stats.end());
std::string dst;
PutVarint64(&dst, val);
stats.insert(
std::make_pair(name, dst)
);
}
static bool GoodCompressionRatio(size_t compressed_size, size_t raw_size) {
// Check to see if compressed less than 12.5%
return compressed_size < raw_size - (raw_size / 8u);
}
} // anonymous namespace
struct TableBuilder::Rep {
Options options;
Options index_block_options;
WritableFile* file;
uint64_t offset = 0;
Status status;
BlockBuilder data_block;
BlockBuilder index_block;
std::string last_key;
// Whether enable compression in this table.
bool enable_compression;
uint64_t num_entries = 0;
uint64_t num_data_blocks = 0;
uint64_t raw_key_size = 0;
uint64_t raw_value_size = 0;
uint64_t data_size = 0;
bool closed = false; // Either Finish() or Abandon() has been called.
FilterBlockBuilder* filter_block;
// We do not emit the index entry for a block until we have seen the
// first key for the next data block. This allows us to use shorter
// keys in the index block. For example, consider a block boundary
// between the keys "the quick brown fox" and "the who". We can use
// "the r" as the key for the index block entry since it is >= all
// entries in the first block and < all entries in subsequent
// blocks.
//
// Invariant: r->pending_index_entry is true only if data_block is empty.
bool pending_index_entry;
BlockHandle pending_handle; // Handle to add to index block
std::string compressed_output;
Rep(const Options& opt, WritableFile* f, bool enable_compression)
: options(opt),
index_block_options(opt),
file(f),
data_block(&options),
index_block(1, index_block_options.comparator),
enable_compression(enable_compression),
filter_block(opt.filter_policy == nullptr ? nullptr
: new FilterBlockBuilder(opt)),
pending_index_entry(false) {
}
};
Allow having different compression algorithms on different levels. Summary: The leveldb API is enhanced to support different compression algorithms at different levels. This adds the option min_level_to_compress to db_bench that specifies the minimum level for which compression should be done when compression is enabled. This can be used to disable compression for levels 0 and 1 which are likely to suffer from stalls because of the CPU load for memtable flushes and (L0,L1) compaction. Level 0 is special as it gets frequent memtable flushes. Level 1 is special as it frequently gets all:all file compactions between it and level 0. But all other levels could be the same. For any level N where N > 1, the rate of sequential IO for that level should be the same. The last level is the exception because it might not be full and because files from it are not read to compact with the next larger level. The same amount of time will be spent doing compaction at any level N excluding N=0, 1 or the last level. By this standard all of those levels should use the same compression. The difference is that the loss (using more disk space) from a faster compression algorithm is less significant for N=2 than for N=3. So we might be willing to trade disk space for faster write rates with no compression for L0 and L1, snappy for L2, zlib for L3. Using a faster compression algorithm for the mid levels also allows us to reclaim some cpu without trading off much loss in disk space overhead. Also note that little is to be gained by compressing levels 0 and 1. For a 4-level tree they account for 10% of the data. For a 5-level tree they account for 1% of the data. With compression enabled: * memtable flush rate is ~18MB/second * (L0,L1) compaction rate is ~30MB/second With compression enabled but min_level_to_compress=2 * memtable flush rate is ~320MB/second * (L0,L1) compaction rate is ~560MB/second This practicaly takes the same code from https://reviews.facebook.net/D6225 but makes the leveldb api more general purpose with a few additional lines of code. Test Plan: make check Differential Revision: https://reviews.facebook.net/D6261
2012-10-28 07:13:17 +01:00
TableBuilder::TableBuilder(const Options& options, WritableFile* file,
int level, const bool enable_compression)
: rep_(new Rep(options, file, enable_compression)), level_(level) {
if (rep_->filter_block != nullptr) {
rep_->filter_block->StartBlock(0);
}
}
TableBuilder::~TableBuilder() {
assert(rep_->closed); // Catch errors where caller forgot to call Finish()
delete rep_->filter_block;
delete rep_;
}
Status TableBuilder::ChangeOptions(const Options& options) {
// Note: if more fields are added to Options, update
// this function to catch changes that should not be allowed to
// change in the middle of building a Table.
if (options.comparator != rep_->options.comparator) {
return Status::InvalidArgument("changing comparator while building table");
}
// Note that any live BlockBuilders point to rep_->options and therefore
// will automatically pick up the updated options.
rep_->options = options;
rep_->index_block_options = options;
rep_->index_block_options.block_restart_interval = 1;
return Status::OK();
}
void TableBuilder::Add(const Slice& key, const Slice& value) {
Rep* r = rep_;
assert(!r->closed);
if (!ok()) return;
if (r->num_entries > 0) {
assert(r->options.comparator->Compare(key, Slice(r->last_key)) > 0);
}
const size_t curr_size = r->data_block.CurrentSizeEstimate();
const size_t estimated_size_after = r->data_block.EstimateSizeAfterKV(key,
value);
// Do flush if one of the below two conditions is true:
// 1) if the current estimated size already exceeds the block size,
// 2) block_size_deviation is set and the estimated size after appending
// the kv will exceed the block size and the current size is under the
// the deviation.
if (curr_size >= r->options.block_size ||
(estimated_size_after > r->options.block_size &&
r->options.block_size_deviation > 0 &&
(curr_size * 100) >
r->options.block_size * (100 - r->options.block_size_deviation))) {
Flush();
}
if (r->pending_index_entry) {
assert(r->data_block.empty());
r->options.comparator->FindShortestSeparator(&r->last_key, key);
std::string handle_encoding;
r->pending_handle.EncodeTo(&handle_encoding);
r->index_block.Add(r->last_key, Slice(handle_encoding));
r->pending_index_entry = false;
}
if (r->filter_block != nullptr) {
r->filter_block->AddKey(key);
}
r->last_key.assign(key.data(), key.size());
r->data_block.Add(key, value);
r->num_entries++;
r->raw_key_size += key.size();
r->raw_value_size += value.size();
}
void TableBuilder::Flush() {
Rep* r = rep_;
assert(!r->closed);
if (!ok()) return;
if (r->data_block.empty()) return;
assert(!r->pending_index_entry);
WriteBlock(&r->data_block, &r->pending_handle);
if (ok()) {
r->pending_index_entry = true;
r->status = r->file->Flush();
}
if (r->filter_block != nullptr) {
r->filter_block->StartBlock(r->offset);
}
r->data_size = r->offset;
++r->num_data_blocks;
}
void TableBuilder::WriteBlock(BlockBuilder* block, BlockHandle* handle) {
// File format contains a sequence of blocks where each block has:
// block_data: uint8[n]
// type: uint8
// crc: uint32
assert(ok());
Rep* r = rep_;
Slice raw = block->Finish();
Slice block_contents;
std::string* compressed = &r->compressed_output;
Allow having different compression algorithms on different levels. Summary: The leveldb API is enhanced to support different compression algorithms at different levels. This adds the option min_level_to_compress to db_bench that specifies the minimum level for which compression should be done when compression is enabled. This can be used to disable compression for levels 0 and 1 which are likely to suffer from stalls because of the CPU load for memtable flushes and (L0,L1) compaction. Level 0 is special as it gets frequent memtable flushes. Level 1 is special as it frequently gets all:all file compactions between it and level 0. But all other levels could be the same. For any level N where N > 1, the rate of sequential IO for that level should be the same. The last level is the exception because it might not be full and because files from it are not read to compact with the next larger level. The same amount of time will be spent doing compaction at any level N excluding N=0, 1 or the last level. By this standard all of those levels should use the same compression. The difference is that the loss (using more disk space) from a faster compression algorithm is less significant for N=2 than for N=3. So we might be willing to trade disk space for faster write rates with no compression for L0 and L1, snappy for L2, zlib for L3. Using a faster compression algorithm for the mid levels also allows us to reclaim some cpu without trading off much loss in disk space overhead. Also note that little is to be gained by compressing levels 0 and 1. For a 4-level tree they account for 10% of the data. For a 5-level tree they account for 1% of the data. With compression enabled: * memtable flush rate is ~18MB/second * (L0,L1) compaction rate is ~30MB/second With compression enabled but min_level_to_compress=2 * memtable flush rate is ~320MB/second * (L0,L1) compaction rate is ~560MB/second This practicaly takes the same code from https://reviews.facebook.net/D6225 but makes the leveldb api more general purpose with a few additional lines of code. Test Plan: make check Differential Revision: https://reviews.facebook.net/D6261
2012-10-28 07:13:17 +01:00
CompressionType type;
if (!r->enable_compression) {
// disable compression
type = kNoCompression;
Allow having different compression algorithms on different levels. Summary: The leveldb API is enhanced to support different compression algorithms at different levels. This adds the option min_level_to_compress to db_bench that specifies the minimum level for which compression should be done when compression is enabled. This can be used to disable compression for levels 0 and 1 which are likely to suffer from stalls because of the CPU load for memtable flushes and (L0,L1) compaction. Level 0 is special as it gets frequent memtable flushes. Level 1 is special as it frequently gets all:all file compactions between it and level 0. But all other levels could be the same. For any level N where N > 1, the rate of sequential IO for that level should be the same. The last level is the exception because it might not be full and because files from it are not read to compact with the next larger level. The same amount of time will be spent doing compaction at any level N excluding N=0, 1 or the last level. By this standard all of those levels should use the same compression. The difference is that the loss (using more disk space) from a faster compression algorithm is less significant for N=2 than for N=3. So we might be willing to trade disk space for faster write rates with no compression for L0 and L1, snappy for L2, zlib for L3. Using a faster compression algorithm for the mid levels also allows us to reclaim some cpu without trading off much loss in disk space overhead. Also note that little is to be gained by compressing levels 0 and 1. For a 4-level tree they account for 10% of the data. For a 5-level tree they account for 1% of the data. With compression enabled: * memtable flush rate is ~18MB/second * (L0,L1) compaction rate is ~30MB/second With compression enabled but min_level_to_compress=2 * memtable flush rate is ~320MB/second * (L0,L1) compaction rate is ~560MB/second This practicaly takes the same code from https://reviews.facebook.net/D6225 but makes the leveldb api more general purpose with a few additional lines of code. Test Plan: make check Differential Revision: https://reviews.facebook.net/D6261
2012-10-28 07:13:17 +01:00
} else {
// If the use has specified a different compression level for each level,
// then pick the compresison for that level.
if (!r->options.compression_per_level.empty()) {
const int n = r->options.compression_per_level.size();
// It is possible for level_ to be -1; in that case, we use level
// 0's compression. This occurs mostly in backwards compatibility
// situations when the builder doesn't know what level the file
// belongs to. Likewise, if level_ is beyond the end of the
// specified compression levels, use the last value.
type = r->options.compression_per_level[std::max(0,
std::min(level_, n))];
} else {
type = r->options.compression;
}
Allow having different compression algorithms on different levels. Summary: The leveldb API is enhanced to support different compression algorithms at different levels. This adds the option min_level_to_compress to db_bench that specifies the minimum level for which compression should be done when compression is enabled. This can be used to disable compression for levels 0 and 1 which are likely to suffer from stalls because of the CPU load for memtable flushes and (L0,L1) compaction. Level 0 is special as it gets frequent memtable flushes. Level 1 is special as it frequently gets all:all file compactions between it and level 0. But all other levels could be the same. For any level N where N > 1, the rate of sequential IO for that level should be the same. The last level is the exception because it might not be full and because files from it are not read to compact with the next larger level. The same amount of time will be spent doing compaction at any level N excluding N=0, 1 or the last level. By this standard all of those levels should use the same compression. The difference is that the loss (using more disk space) from a faster compression algorithm is less significant for N=2 than for N=3. So we might be willing to trade disk space for faster write rates with no compression for L0 and L1, snappy for L2, zlib for L3. Using a faster compression algorithm for the mid levels also allows us to reclaim some cpu without trading off much loss in disk space overhead. Also note that little is to be gained by compressing levels 0 and 1. For a 4-level tree they account for 10% of the data. For a 5-level tree they account for 1% of the data. With compression enabled: * memtable flush rate is ~18MB/second * (L0,L1) compaction rate is ~30MB/second With compression enabled but min_level_to_compress=2 * memtable flush rate is ~320MB/second * (L0,L1) compaction rate is ~560MB/second This practicaly takes the same code from https://reviews.facebook.net/D6225 but makes the leveldb api more general purpose with a few additional lines of code. Test Plan: make check Differential Revision: https://reviews.facebook.net/D6261
2012-10-28 07:13:17 +01:00
}
switch (type) {
case kNoCompression:
block_contents = raw;
break;
case kSnappyCompression: {
std::string* compressed = &r->compressed_output;
if (port::Snappy_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {
// Snappy not supported, or not good compression ratio, so just
// store uncompressed form
block_contents = raw;
type = kNoCompression;
}
break;
}
case kZlibCompression:
if (port::Zlib_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {
// Zlib not supported, or not good compression ratio, so just
// store uncompressed form
block_contents = raw;
type = kNoCompression;
}
break;
case kBZip2Compression:
if (port::BZip2_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {
// BZip not supported, or not good compression ratio, so just
// store uncompressed form
block_contents = raw;
type = kNoCompression;
}
break;
}
WriteRawBlock(block_contents, type, handle);
r->compressed_output.clear();
block->Reset();
}
void TableBuilder::WriteRawBlock(const Slice& block_contents,
CompressionType type,
BlockHandle* handle) {
Rep* r = rep_;
StopWatch sw(r->options.env, r->options.statistics, WRITE_RAW_BLOCK_MICROS);
handle->set_offset(r->offset);
handle->set_size(block_contents.size());
r->status = r->file->Append(block_contents);
if (r->status.ok()) {
char trailer[kBlockTrailerSize];
trailer[0] = type;
uint32_t crc = crc32c::Value(block_contents.data(), block_contents.size());
crc = crc32c::Extend(crc, trailer, 1); // Extend crc to cover block type
EncodeFixed32(trailer+1, crc32c::Mask(crc));
r->status = r->file->Append(Slice(trailer, kBlockTrailerSize));
if (r->status.ok()) {
r->offset += block_contents.size() + kBlockTrailerSize;
}
}
}
Status TableBuilder::status() const {
return rep_->status;
}
Status TableBuilder::Finish() {
Rep* r = rep_;
Flush();
assert(!r->closed);
r->closed = true;
BlockHandle filter_block_handle, metaindex_block_handle, index_block_handle;
// Write filter block
if (ok() && r->filter_block != nullptr) {
WriteRawBlock(r->filter_block->Finish(), kNoCompression,
&filter_block_handle);
}
// To make sure stats block is able to keep the accurate size of index
// block, we will finish writing all index entries here and flush them
// to storage after metaindex block is written.
if (ok() && (r->pending_index_entry)) {
r->options.comparator->FindShortSuccessor(&r->last_key);
std::string handle_encoding;
r->pending_handle.EncodeTo(&handle_encoding);
r->index_block.Add(r->last_key, Slice(handle_encoding));
r->pending_index_entry = false;
}
// Write meta blocks and metaindex block with the following order.
// 1. [meta block: filter]
// 2. [meta block: stats]
// 3. [metaindex block]
if (ok()) {
// We use `BytewiseComparator` as the comparator for meta block.
BlockBuilder meta_index_block(
r->options.block_restart_interval,
BytewiseComparator()
);
// Key: meta block name
// Value: block handle to that meta block
BytewiseSortedMap meta_block_handles;
// Write filter block.
if (r->filter_block != nullptr) {
// Add mapping from "<filter_block_prefix>.Name" to location
// of filter data.
std::string key = Table::kFilterBlockPrefix;
key.append(r->options.filter_policy->Name());
std::string handle_encoding;
filter_block_handle.EncodeTo(&handle_encoding);
meta_block_handles.insert(
std::make_pair(key, handle_encoding)
);
}
// Write stats block.
{
BlockBuilder stats_block(
r->options.block_restart_interval,
BytewiseComparator()
);
BytewiseSortedMap stats;
// Add basic stats
AddStats(stats, TableStatsNames::kRawKeySize, r->raw_key_size);
AddStats(stats, TableStatsNames::kRawValueSize, r->raw_value_size);
AddStats(stats, TableStatsNames::kDataSize, r->data_size);
AddStats(
stats,
TableStatsNames::kIndexSize,
r->index_block.CurrentSizeEstimate() + kBlockTrailerSize
);
AddStats(stats, TableStatsNames::kNumEntries, r->num_entries);
AddStats(stats, TableStatsNames::kNumDataBlocks, r->num_data_blocks);
if (r->filter_block != nullptr) {
stats.insert(std::make_pair(
TableStatsNames::kFilterPolicy,
r->options.filter_policy->Name()
));
}
for (const auto& stat : stats) {
stats_block.Add(stat.first, stat.second);
}
BlockHandle stats_block_handle;
WriteBlock(&stats_block, &stats_block_handle);
std::string handle_encoding;
stats_block_handle.EncodeTo(&handle_encoding);
meta_block_handles.insert(
std::make_pair(Table::kStatsBlock, handle_encoding)
);
} // end of stats block writing
for (const auto& metablock : meta_block_handles) {
meta_index_block.Add(metablock.first, metablock.second);
}
WriteBlock(&meta_index_block, &metaindex_block_handle);
} // meta blocks and metaindex block.
// Write index block
if (ok()) {
WriteBlock(&r->index_block, &index_block_handle);
}
// Write footer
if (ok()) {
Footer footer;
footer.set_metaindex_handle(metaindex_block_handle);
footer.set_index_handle(index_block_handle);
std::string footer_encoding;
footer.EncodeTo(&footer_encoding);
r->status = r->file->Append(footer_encoding);
if (r->status.ok()) {
r->offset += footer_encoding.size();
}
}
return r->status;
}
void TableBuilder::Abandon() {
Rep* r = rep_;
assert(!r->closed);
r->closed = true;
}
uint64_t TableBuilder::NumEntries() const {
return rep_->num_entries;
}
uint64_t TableBuilder::FileSize() const {
return rep_->offset;
}
} // namespace rocksdb