rocksdb/table/sst_file_dumper.cc
Andrew Kryczka af54c4092a fix SstFileWriter with dictionary compression (#7323)
Summary:
In block-based table builder, the cut-over from buffered to unbuffered
mode involves sampling the buffered blocks and generating a dictionary.
There was a bug where `SstFileWriter` passed zero as the `target_file_size`
causing the cutover to happen immediately, so there were no samples
available for generating the dictionary.

This PR changes the meaning of `target_file_size == 0` to mean buffer
the whole file before cutting over. It also adds dictionary compression
support to `sst_dump --command=recompress` for easy evaluation.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/7323

Reviewed By: cheng-chang

Differential Revision: D23412158

Pulled By: ajkr

fbshipit-source-id: 3b232050e70ef3c2ee85a4b5f6fadb139c569873
2020-09-03 15:49:57 -07:00

485 lines
18 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).
//
#ifndef ROCKSDB_LITE
#include "table/sst_file_dumper.h"
#include <chrono>
#include <cinttypes>
#include <iostream>
#include <map>
#include <memory>
#include <sstream>
#include <vector>
#include "db/blob/blob_index.h"
#include "db/memtable.h"
#include "db/write_batch_internal.h"
#include "env/composite_env_wrapper.h"
#include "options/cf_options.h"
#include "port/port.h"
#include "rocksdb/db.h"
#include "rocksdb/env.h"
#include "rocksdb/iterator.h"
#include "rocksdb/slice_transform.h"
#include "rocksdb/status.h"
#include "rocksdb/table_properties.h"
#include "rocksdb/utilities/ldb_cmd.h"
#include "table/block_based/block.h"
#include "table/block_based/block_based_table_builder.h"
#include "table/block_based/block_based_table_factory.h"
#include "table/block_based/block_builder.h"
#include "table/format.h"
#include "table/meta_blocks.h"
#include "table/plain/plain_table_factory.h"
#include "table/table_reader.h"
#include "util/compression.h"
#include "util/random.h"
namespace ROCKSDB_NAMESPACE {
SstFileDumper::SstFileDumper(const Options& options,
const std::string& file_path,
size_t readahead_size, bool verify_checksum,
bool output_hex, bool decode_blob_index,
const EnvOptions& soptions, bool silent)
: file_name_(file_path),
read_num_(0),
output_hex_(output_hex),
decode_blob_index_(decode_blob_index),
soptions_(soptions),
silent_(silent),
options_(options),
ioptions_(options_),
moptions_(ColumnFamilyOptions(options_)),
read_options_(verify_checksum, false),
internal_comparator_(BytewiseComparator()) {
read_options_.readahead_size = readahead_size;
if (!silent_) {
fprintf(stdout, "Process %s\n", file_path.c_str());
}
init_result_ = GetTableReader(file_name_);
}
extern const uint64_t kBlockBasedTableMagicNumber;
extern const uint64_t kLegacyBlockBasedTableMagicNumber;
extern const uint64_t kPlainTableMagicNumber;
extern const uint64_t kLegacyPlainTableMagicNumber;
const char* testFileName = "test_file_name";
Status SstFileDumper::GetTableReader(const std::string& file_path) {
// Warning about 'magic_number' being uninitialized shows up only in UBsan
// builds. Though access is guarded by 's.ok()' checks, fix the issue to
// avoid any warnings.
uint64_t magic_number = Footer::kInvalidTableMagicNumber;
// read table magic number
Footer footer;
std::unique_ptr<RandomAccessFile> file;
uint64_t file_size = 0;
Status s = options_.env->NewRandomAccessFile(file_path, &file, soptions_);
if (s.ok()) {
s = options_.env->GetFileSize(file_path, &file_size);
}
// check empty file
// if true, skip further processing of this file
if (file_size == 0) {
return Status::Aborted(file_path, "Empty file");
}
file_.reset(new RandomAccessFileReader(NewLegacyRandomAccessFileWrapper(file),
file_path));
FilePrefetchBuffer prefetch_buffer(nullptr, 0, 0, true /* enable */,
false /* track_min_offset */);
if (s.ok()) {
const uint64_t kSstDumpTailPrefetchSize = 512 * 1024;
uint64_t prefetch_size = (file_size > kSstDumpTailPrefetchSize)
? kSstDumpTailPrefetchSize
: file_size;
uint64_t prefetch_off = file_size - prefetch_size;
IOOptions opts;
prefetch_buffer.Prefetch(opts, file_.get(), prefetch_off,
static_cast<size_t>(prefetch_size));
s = ReadFooterFromFile(opts, file_.get(), &prefetch_buffer, file_size,
&footer);
}
if (s.ok()) {
magic_number = footer.table_magic_number();
}
if (s.ok()) {
if (magic_number == kPlainTableMagicNumber ||
magic_number == kLegacyPlainTableMagicNumber) {
soptions_.use_mmap_reads = true;
options_.env->NewRandomAccessFile(file_path, &file, soptions_);
file_.reset(new RandomAccessFileReader(
NewLegacyRandomAccessFileWrapper(file), file_path));
}
options_.comparator = &internal_comparator_;
// For old sst format, ReadTableProperties might fail but file can be read
if (ReadTableProperties(magic_number, file_.get(), file_size,
(magic_number == kBlockBasedTableMagicNumber)
? &prefetch_buffer
: nullptr)
.ok()) {
SetTableOptionsByMagicNumber(magic_number);
} else {
SetOldTableOptions();
}
}
if (s.ok()) {
s = NewTableReader(ioptions_, soptions_, internal_comparator_, file_size,
&table_reader_);
}
return s;
}
Status SstFileDumper::NewTableReader(
const ImmutableCFOptions& /*ioptions*/, const EnvOptions& /*soptions*/,
const InternalKeyComparator& /*internal_comparator*/, uint64_t file_size,
std::unique_ptr<TableReader>* /*table_reader*/) {
auto t_opt =
TableReaderOptions(ioptions_, moptions_.prefix_extractor.get(), soptions_,
internal_comparator_, false /* skip_filters */,
false /* imortal */, true /* force_direct_prefetch */);
// Allow open file with global sequence number for backward compatibility.
t_opt.largest_seqno = kMaxSequenceNumber;
// We need to turn off pre-fetching of index and filter nodes for
// BlockBasedTable
if (BlockBasedTableFactory::kName == options_.table_factory->Name()) {
return options_.table_factory->NewTableReader(t_opt, std::move(file_),
file_size, &table_reader_,
/*enable_prefetch=*/false);
}
// For all other factory implementation
return options_.table_factory->NewTableReader(t_opt, std::move(file_),
file_size, &table_reader_);
}
Status SstFileDumper::VerifyChecksum() {
// We could pass specific readahead setting into read options if needed.
return table_reader_->VerifyChecksum(read_options_,
TableReaderCaller::kSSTDumpTool);
}
Status SstFileDumper::DumpTable(const std::string& out_filename) {
std::unique_ptr<WritableFile> out_file;
Env* env = options_.env;
env->NewWritableFile(out_filename, &out_file, soptions_);
Status s = table_reader_->DumpTable(out_file.get());
out_file->Close();
return s;
}
uint64_t SstFileDumper::CalculateCompressedTableSize(
const TableBuilderOptions& tb_options, size_t block_size,
uint64_t* num_data_blocks) {
std::unique_ptr<WritableFile> out_file;
std::unique_ptr<Env> env(NewMemEnv(options_.env));
env->NewWritableFile(testFileName, &out_file, soptions_);
std::unique_ptr<WritableFileWriter> dest_writer;
dest_writer.reset(
new WritableFileWriter(NewLegacyWritableFileWrapper(std::move(out_file)),
testFileName, soptions_));
BlockBasedTableOptions table_options;
table_options.block_size = block_size;
BlockBasedTableFactory block_based_tf(table_options);
std::unique_ptr<TableBuilder> table_builder;
table_builder.reset(block_based_tf.NewTableBuilder(
tb_options,
TablePropertiesCollectorFactory::Context::kUnknownColumnFamily,
dest_writer.get()));
std::unique_ptr<InternalIterator> iter(table_reader_->NewIterator(
read_options_, moptions_.prefix_extractor.get(), /*arena=*/nullptr,
/*skip_filters=*/false, TableReaderCaller::kSSTDumpTool));
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
table_builder->Add(iter->key(), iter->value());
}
if (!iter->status().ok()) {
fputs(iter->status().ToString().c_str(), stderr);
exit(1);
}
Status s = table_builder->Finish();
if (!s.ok()) {
fputs(s.ToString().c_str(), stderr);
exit(1);
}
uint64_t size = table_builder->FileSize();
assert(num_data_blocks != nullptr);
*num_data_blocks = table_builder->GetTableProperties().num_data_blocks;
env->DeleteFile(testFileName);
return size;
}
int SstFileDumper::ShowAllCompressionSizes(
size_t block_size,
const std::vector<std::pair<CompressionType, const char*>>&
compression_types,
int32_t compress_level_from, int32_t compress_level_to,
uint32_t max_dict_bytes, uint32_t zstd_max_train_bytes) {
fprintf(stdout, "Block Size: %" ROCKSDB_PRIszt "\n", block_size);
for (auto& i : compression_types) {
if (CompressionTypeSupported(i.first)) {
fprintf(stdout, "Compression: %-24s\n", i.second);
CompressionOptions compress_opt;
compress_opt.max_dict_bytes = max_dict_bytes;
compress_opt.zstd_max_train_bytes = zstd_max_train_bytes;
for (int32_t j = compress_level_from; j <= compress_level_to; j++) {
fprintf(stdout, "Compression level: %d", j);
compress_opt.level = j;
ShowCompressionSize(block_size, i.first, compress_opt);
}
} else {
fprintf(stdout, "Unsupported compression type: %s.\n", i.second);
}
}
return 0;
}
int SstFileDumper::ShowCompressionSize(size_t block_size,
CompressionType compress_type,
const CompressionOptions& compress_opt) {
Options opts;
opts.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
opts.statistics->set_stats_level(StatsLevel::kAll);
const ImmutableCFOptions imoptions(opts);
const ColumnFamilyOptions cfo(opts);
const MutableCFOptions moptions(cfo);
ROCKSDB_NAMESPACE::InternalKeyComparator ikc(opts.comparator);
std::vector<std::unique_ptr<IntTblPropCollectorFactory>>
block_based_table_factories;
std::string column_family_name;
int unknown_level = -1;
TableBuilderOptions tb_opts(
imoptions, moptions, ikc, &block_based_table_factories, compress_type,
0 /* sample_for_compression */, compress_opt, false /* skip_filters */,
column_family_name, unknown_level);
uint64_t num_data_blocks = 0;
std::chrono::steady_clock::time_point start =
std::chrono::steady_clock::now();
uint64_t file_size =
CalculateCompressedTableSize(tb_opts, block_size, &num_data_blocks);
std::chrono::steady_clock::time_point end = std::chrono::steady_clock::now();
fprintf(stdout, " Size: %10" PRIu64, file_size);
fprintf(stdout, " Blocks: %6" PRIu64, num_data_blocks);
fprintf(stdout, " Time Taken: %10s microsecs",
std::to_string(
std::chrono::duration_cast<std::chrono::microseconds>(end - start)
.count())
.c_str());
const uint64_t compressed_blocks =
opts.statistics->getAndResetTickerCount(NUMBER_BLOCK_COMPRESSED);
const uint64_t not_compressed_blocks =
opts.statistics->getAndResetTickerCount(NUMBER_BLOCK_NOT_COMPRESSED);
// When the option enable_index_compression is true,
// NUMBER_BLOCK_COMPRESSED is incremented for index block(s).
if ((compressed_blocks + not_compressed_blocks) > num_data_blocks) {
num_data_blocks = compressed_blocks + not_compressed_blocks;
}
const uint64_t ratio_not_compressed_blocks =
(num_data_blocks - compressed_blocks) - not_compressed_blocks;
const double compressed_pcnt =
(0 == num_data_blocks) ? 0.0
: ((static_cast<double>(compressed_blocks) /
static_cast<double>(num_data_blocks)) *
100.0);
const double ratio_not_compressed_pcnt =
(0 == num_data_blocks)
? 0.0
: ((static_cast<double>(ratio_not_compressed_blocks) /
static_cast<double>(num_data_blocks)) *
100.0);
const double not_compressed_pcnt =
(0 == num_data_blocks) ? 0.0
: ((static_cast<double>(not_compressed_blocks) /
static_cast<double>(num_data_blocks)) *
100.0);
fprintf(stdout, " Compressed: %6" PRIu64 " (%5.1f%%)", compressed_blocks,
compressed_pcnt);
fprintf(stdout, " Not compressed (ratio): %6" PRIu64 " (%5.1f%%)",
ratio_not_compressed_blocks, ratio_not_compressed_pcnt);
fprintf(stdout, " Not compressed (abort): %6" PRIu64 " (%5.1f%%)\n",
not_compressed_blocks, not_compressed_pcnt);
return 0;
}
Status SstFileDumper::ReadTableProperties(uint64_t table_magic_number,
RandomAccessFileReader* file,
uint64_t file_size,
FilePrefetchBuffer* prefetch_buffer) {
TableProperties* table_properties = nullptr;
Status s = ROCKSDB_NAMESPACE::ReadTableProperties(
file, file_size, table_magic_number, ioptions_, &table_properties,
/* compression_type_missing= */ false,
/* memory_allocator= */ nullptr, prefetch_buffer);
if (s.ok()) {
table_properties_.reset(table_properties);
} else {
if (!silent_) {
fprintf(stdout, "Not able to read table properties\n");
}
}
return s;
}
Status SstFileDumper::SetTableOptionsByMagicNumber(
uint64_t table_magic_number) {
assert(table_properties_);
if (table_magic_number == kBlockBasedTableMagicNumber ||
table_magic_number == kLegacyBlockBasedTableMagicNumber) {
BlockBasedTableFactory* bbtf = new BlockBasedTableFactory();
// To force tail prefetching, we fake reporting two useful reads of 512KB
// from the tail.
// It needs at least two data points to warm up the stats.
bbtf->tail_prefetch_stats()->RecordEffectiveSize(512 * 1024);
bbtf->tail_prefetch_stats()->RecordEffectiveSize(512 * 1024);
options_.table_factory.reset(bbtf);
if (!silent_) {
fprintf(stdout, "Sst file format: block-based\n");
}
auto& props = table_properties_->user_collected_properties;
auto pos = props.find(BlockBasedTablePropertyNames::kIndexType);
if (pos != props.end()) {
auto index_type_on_file = static_cast<BlockBasedTableOptions::IndexType>(
DecodeFixed32(pos->second.c_str()));
if (index_type_on_file ==
BlockBasedTableOptions::IndexType::kHashSearch) {
options_.prefix_extractor.reset(NewNoopTransform());
}
}
} else if (table_magic_number == kPlainTableMagicNumber ||
table_magic_number == kLegacyPlainTableMagicNumber) {
options_.allow_mmap_reads = true;
PlainTableOptions plain_table_options;
plain_table_options.user_key_len = kPlainTableVariableLength;
plain_table_options.bloom_bits_per_key = 0;
plain_table_options.hash_table_ratio = 0;
plain_table_options.index_sparseness = 1;
plain_table_options.huge_page_tlb_size = 0;
plain_table_options.encoding_type = kPlain;
plain_table_options.full_scan_mode = true;
options_.table_factory.reset(NewPlainTableFactory(plain_table_options));
if (!silent_) {
fprintf(stdout, "Sst file format: plain table\n");
}
} else {
char error_msg_buffer[80];
snprintf(error_msg_buffer, sizeof(error_msg_buffer) - 1,
"Unsupported table magic number --- %lx",
(long)table_magic_number);
return Status::InvalidArgument(error_msg_buffer);
}
return Status::OK();
}
Status SstFileDumper::SetOldTableOptions() {
assert(table_properties_ == nullptr);
options_.table_factory = std::make_shared<BlockBasedTableFactory>();
if (!silent_) {
fprintf(stdout, "Sst file format: block-based(old version)\n");
}
return Status::OK();
}
Status SstFileDumper::ReadSequential(bool print_kv, uint64_t read_num,
bool has_from, const std::string& from_key,
bool has_to, const std::string& to_key,
bool use_from_as_prefix) {
if (!table_reader_) {
return init_result_;
}
InternalIterator* iter = table_reader_->NewIterator(
read_options_, moptions_.prefix_extractor.get(),
/*arena=*/nullptr, /*skip_filters=*/false,
TableReaderCaller::kSSTDumpTool);
uint64_t i = 0;
if (has_from) {
InternalKey ikey;
ikey.SetMinPossibleForUserKey(from_key);
iter->Seek(ikey.Encode());
} else {
iter->SeekToFirst();
}
for (; iter->Valid(); iter->Next()) {
Slice key = iter->key();
Slice value = iter->value();
++i;
if (read_num > 0 && i > read_num) break;
ParsedInternalKey ikey;
if (!ParseInternalKey(key, &ikey)) {
std::cerr << "Internal Key [" << key.ToString(true /* in hex*/)
<< "] parse error!\n";
continue;
}
// the key returned is not prefixed with out 'from' key
if (use_from_as_prefix && !ikey.user_key.starts_with(from_key)) {
break;
}
// If end marker was specified, we stop before it
if (has_to && BytewiseComparator()->Compare(ikey.user_key, to_key) >= 0) {
break;
}
if (print_kv) {
if (!decode_blob_index_ || ikey.type != kTypeBlobIndex) {
fprintf(stdout, "%s => %s\n", ikey.DebugString(output_hex_).c_str(),
value.ToString(output_hex_).c_str());
} else {
BlobIndex blob_index;
const Status s = blob_index.DecodeFrom(value);
if (!s.ok()) {
fprintf(stderr, "%s => error decoding blob index\n",
ikey.DebugString(output_hex_).c_str());
continue;
}
fprintf(stdout, "%s => %s\n", ikey.DebugString(output_hex_).c_str(),
blob_index.DebugString(output_hex_).c_str());
}
}
}
read_num_ += i;
Status ret = iter->status();
delete iter;
return ret;
}
Status SstFileDumper::ReadTableProperties(
std::shared_ptr<const TableProperties>* table_properties) {
if (!table_reader_) {
return init_result_;
}
*table_properties = table_reader_->GetTableProperties();
return init_result_;
}
} // namespace ROCKSDB_NAMESPACE
#endif // ROCKSDB_LITE