Digest ZSTD compression dictionary once per SST file (#4251)

Summary:
In RocksDB, for a given SST file, all data blocks are compressed with the same dictionary. When we compress a block using the dictionary's raw bytes, the compression library first has to digest the dictionary to get it into a usable form. This digestion work is redundant and ideally should be done once per file.

ZSTD offers APIs for the caller to create and reuse a digested dictionary object (`ZSTD_CDict`). In this PR, we call `ZSTD_createCDict` once per file to digest the raw bytes. Then we use `ZSTD_compress_usingCDict` to compress each data block using the pre-digested dictionary. Once the file's created `ZSTD_freeCDict` releases the resources held by the digested dictionary.

There are a couple other changes included in this PR:

- Changed the parameter object for (un)compression functions from `CompressionContext`/`UncompressionContext` to `CompressionInfo`/`UncompressionInfo`. This avoids the previous pattern, where `CompressionContext`/`UncompressionContext` had to be mutated before calling a (un)compression function depending on whether dictionary should be used. I felt that mutation was error-prone so eliminated it.
- Added support for digested uncompression dictionaries (`ZSTD_DDict`) as well. However, this PR does not support reusing them across uncompression calls for the same file. That work is deferred to a later PR when we will store the `ZSTD_DDict` objects in block cache.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4251

Differential Revision: D9257078

Pulled By: ajkr

fbshipit-source-id: 21b8cb6bbdd48e459f1c62343780ab66c0a64438
This commit is contained in:
Andrew Kryczka 2018-08-23 19:19:16 -07:00 committed by Facebook Github Bot
parent ee234e83e3
commit 6c40806e51
11 changed files with 391 additions and 180 deletions

View File

@ -103,19 +103,19 @@ bool GoodCompressionRatio(size_t compressed_size, size_t raw_size) {
} // namespace
// format_version is the block format as defined in include/rocksdb/table.h
Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
Slice CompressBlock(const Slice& raw, const CompressionInfo& compression_info,
CompressionType* type, uint32_t format_version,
std::string* compressed_output) {
*type = compression_ctx.type();
if (compression_ctx.type() == kNoCompression) {
*type = compression_info.type();
if (compression_info.type() == kNoCompression) {
return raw;
}
// Will return compressed block contents if (1) the compression method is
// supported in this platform and (2) the compression rate is "good enough".
switch (compression_ctx.type()) {
switch (compression_info.type()) {
case kSnappyCompression:
if (Snappy_Compress(compression_ctx, raw.data(), raw.size(),
if (Snappy_Compress(compression_info, raw.data(), raw.size(),
compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
return *compressed_output;
@ -123,7 +123,7 @@ Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
break; // fall back to no compression.
case kZlibCompression:
if (Zlib_Compress(
compression_ctx,
compression_info,
GetCompressFormatForVersion(kZlibCompression, format_version),
raw.data(), raw.size(), compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
@ -132,7 +132,7 @@ Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
break; // fall back to no compression.
case kBZip2Compression:
if (BZip2_Compress(
compression_ctx,
compression_info,
GetCompressFormatForVersion(kBZip2Compression, format_version),
raw.data(), raw.size(), compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
@ -141,7 +141,7 @@ Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
break; // fall back to no compression.
case kLZ4Compression:
if (LZ4_Compress(
compression_ctx,
compression_info,
GetCompressFormatForVersion(kLZ4Compression, format_version),
raw.data(), raw.size(), compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
@ -150,7 +150,7 @@ Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
break; // fall back to no compression.
case kLZ4HCCompression:
if (LZ4HC_Compress(
compression_ctx,
compression_info,
GetCompressFormatForVersion(kLZ4HCCompression, format_version),
raw.data(), raw.size(), compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
@ -166,7 +166,7 @@ Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
break;
case kZSTD:
case kZSTDNotFinalCompression:
if (ZSTD_Compress(compression_ctx, raw.data(), raw.size(),
if (ZSTD_Compress(compression_info, raw.data(), raw.size(),
compressed_output) &&
GoodCompressionRatio(compressed_output->size(), raw.size())) {
return *compressed_output;
@ -260,8 +260,9 @@ struct BlockBasedTableBuilder::Rep {
PartitionedIndexBuilder* p_index_builder_ = nullptr;
std::string last_key;
// Compression dictionary or nullptr
const std::string* compression_dict;
CompressionType compression_type;
CompressionOptions compression_opts;
CompressionDict compression_dict;
CompressionContext compression_ctx;
std::unique_ptr<UncompressionContext> verify_ctx;
TableProperties props;
@ -312,8 +313,9 @@ struct BlockBasedTableBuilder::Rep {
table_options.data_block_hash_table_util_ratio),
range_del_block(1 /* block_restart_interval */),
internal_prefix_transform(_moptions.prefix_extractor.get()),
compression_dict(_compression_dict),
compression_ctx(_compression_type, _compression_opts),
compression_type(_compression_type),
compression_opts(_compression_opts),
compression_ctx(_compression_type),
use_delta_encoding_for_index_values(table_opt.format_version >= 4 &&
!table_opt.block_align),
compressed_cache_key_prefix_size(0),
@ -324,6 +326,15 @@ struct BlockBasedTableBuilder::Rep {
column_family_name(_column_family_name),
creation_time(_creation_time),
oldest_key_time(_oldest_key_time) {
if (_compression_dict != nullptr) {
compression_dict.Init(*_compression_dict,
CompressionDict::Mode::kCompression,
_compression_type, _compression_opts.level);
} else {
compression_dict.Init(Slice() /* dict */,
CompressionDict::Mode::kEmpty,
_compression_type, _compression_opts.level);
}
if (table_options.index_type ==
BlockBasedTableOptions::kTwoLevelIndexSearch) {
p_index_builder_ = PartitionedIndexBuilder::CreateIndexBuilder(
@ -354,7 +365,7 @@ struct BlockBasedTableBuilder::Rep {
_moptions.prefix_extractor != nullptr));
if (table_options.verify_compression) {
verify_ctx.reset(new UncompressionContext(UncompressionContext::NoCache(),
compression_ctx.type()));
compression_type));
}
}
@ -498,7 +509,7 @@ void BlockBasedTableBuilder::WriteBlock(const Slice& raw_block_contents,
assert(ok());
Rep* r = rep_;
auto type = r->compression_ctx.type();
auto type = r->compression_type;
Slice block_contents;
bool abort_compression = false;
@ -506,24 +517,12 @@ void BlockBasedTableBuilder::WriteBlock(const Slice& raw_block_contents,
ShouldReportDetailedTime(r->ioptions.env, r->ioptions.statistics));
if (raw_block_contents.size() < kCompressionSizeLimit) {
Slice compression_dict;
if (is_data_block && r->compression_dict && r->compression_dict->size()) {
r->compression_ctx.dict() = *r->compression_dict;
if (r->table_options.verify_compression) {
assert(r->verify_ctx != nullptr);
r->verify_ctx->dict() = *r->compression_dict;
}
} else {
// Clear dictionary
r->compression_ctx.dict() = Slice();
if (r->table_options.verify_compression) {
assert(r->verify_ctx != nullptr);
r->verify_ctx->dict() = Slice();
}
}
CompressionInfo compression_info(
r->compression_opts, r->compression_ctx,
is_data_block ? r->compression_dict : CompressionDict::GetEmptyDict(),
r->compression_type);
block_contents =
CompressBlock(raw_block_contents, r->compression_ctx, &type,
CompressBlock(raw_block_contents, compression_info, &type,
r->table_options.format_version, &r->compressed_output);
// Some of the compression algorithms are known to be unreliable. If
@ -532,8 +531,12 @@ void BlockBasedTableBuilder::WriteBlock(const Slice& raw_block_contents,
if (type != kNoCompression && r->table_options.verify_compression) {
// Retrieve the uncompressed contents into a new buffer
BlockContents contents;
UncompressionInfo uncompression_info(
*r->verify_ctx,
is_data_block ? r->compression_dict : CompressionDict::GetEmptyDict(),
r->compression_type);
Status stat = UncompressBlockContentsForCompressionType(
*r->verify_ctx, block_contents.data(), block_contents.size(),
uncompression_info, block_contents.data(), block_contents.size(),
&contents, r->table_options.format_version, r->ioptions);
if (stat.ok()) {
@ -780,7 +783,7 @@ void BlockBasedTableBuilder::WritePropertiesBlock(
? rep_->ioptions.merge_operator->Name()
: "nullptr";
rep_->props.compression_name =
CompressionTypeToString(rep_->compression_ctx.type());
CompressionTypeToString(rep_->compression_type);
rep_->props.prefix_extractor_name =
rep_->moptions.prefix_extractor != nullptr
? rep_->moptions.prefix_extractor->Name()
@ -829,10 +832,10 @@ void BlockBasedTableBuilder::WritePropertiesBlock(
void BlockBasedTableBuilder::WriteCompressionDictBlock(
MetaIndexBuilder* meta_index_builder) {
if (rep_->compression_dict && rep_->compression_dict->size()) {
if (rep_->compression_dict.GetRawDict().size()) {
BlockHandle compression_dict_block_handle;
if (ok()) {
WriteRawBlock(*rep_->compression_dict, kNoCompression,
WriteRawBlock(rep_->compression_dict.GetRawDict(), kNoCompression,
&compression_dict_block_handle);
}
if (ok()) {

View File

@ -131,7 +131,7 @@ class BlockBasedTableBuilder : public TableBuilder {
const uint64_t kCompressionSizeLimit = std::numeric_limits<int>::max();
};
Slice CompressBlock(const Slice& raw, const CompressionContext& compression_ctx,
Slice CompressBlock(const Slice& raw, const CompressionInfo& info,
CompressionType* type, uint32_t format_version,
std::string* compressed_output);

View File

@ -1226,9 +1226,12 @@ Status BlockBasedTable::GetDataBlockFromCache(
// Retrieve the uncompressed contents into a new buffer
BlockContents contents;
UncompressionContext uncompresssion_ctx(compressed_block->compression_type(),
compression_dict);
s = UncompressBlockContents(uncompresssion_ctx, compressed_block->data(),
UncompressionContext context(compressed_block->compression_type());
CompressionDict dict;
dict.Init(compression_dict, CompressionDict::Mode::kUncompression,
compressed_block->compression_type());
UncompressionInfo info(context, dict, compressed_block->compression_type());
s = UncompressBlockContents(info, compressed_block->data(),
compressed_block->size(), &contents,
format_version, ioptions);
@ -1301,11 +1304,13 @@ Status BlockBasedTable::PutDataBlockToCache(
BlockContents contents;
Statistics* statistics = ioptions.statistics;
if (raw_block->compression_type() != kNoCompression) {
UncompressionContext uncompression_ctx(raw_block->compression_type(),
compression_dict);
s = UncompressBlockContents(uncompression_ctx, raw_block->data(),
raw_block->size(), &contents, format_version,
ioptions);
UncompressionContext context(raw_block->compression_type());
CompressionDict dict;
dict.Init(compression_dict, CompressionDict::Mode::kUncompression,
raw_block->compression_type());
UncompressionInfo info(context, dict, raw_block->compression_type());
s = UncompressBlockContents(info, raw_block->data(), raw_block->size(),
&contents, format_version, ioptions);
}
if (!s.ok()) {
delete raw_block;

View File

@ -227,10 +227,13 @@ Status BlockFetcher::ReadBlockContents() {
if (do_uncompress_ && compression_type != kNoCompression) {
// compressed page, uncompress, update cache
UncompressionContext uncompression_ctx(compression_type, compression_dict_);
status_ =
UncompressBlockContents(uncompression_ctx, slice_.data(), block_size_,
contents_, footer_.version(), ioptions_);
UncompressionContext context(compression_type);
CompressionDict dict;
dict.Init(compression_dict_, CompressionDict::Mode::kUncompression,
compression_type);
UncompressionInfo info(context, dict, compression_type);
status_ = UncompressBlockContents(info, slice_.data(), block_size_,
contents_, footer_.version(), ioptions_);
} else {
GetBlockContents();
}

View File

@ -284,18 +284,18 @@ Status ReadFooterFromFile(RandomAccessFileReader* file,
}
Status UncompressBlockContentsForCompressionType(
const UncompressionContext& uncompression_ctx, const char* data, size_t n,
const UncompressionInfo& uncompression_info, const char* data, size_t n,
BlockContents* contents, uint32_t format_version,
const ImmutableCFOptions& ioptions) {
std::unique_ptr<char[]> ubuf;
assert(uncompression_ctx.type() != kNoCompression &&
assert(uncompression_info.type() != kNoCompression &&
"Invalid compression type");
StopWatchNano timer(ioptions.env,
ShouldReportDetailedTime(ioptions.env, ioptions.statistics));
int decompress_size = 0;
switch (uncompression_ctx.type()) {
switch (uncompression_info.type()) {
case kSnappyCompression: {
size_t ulength = 0;
static char snappy_corrupt_msg[] =
@ -312,7 +312,7 @@ Status UncompressBlockContentsForCompressionType(
}
case kZlibCompression:
ubuf.reset(Zlib_Uncompress(
uncompression_ctx, data, n, &decompress_size,
uncompression_info, data, n, &decompress_size,
GetCompressFormatForVersion(kZlibCompression, format_version)));
if (!ubuf) {
static char zlib_corrupt_msg[] =
@ -336,7 +336,7 @@ Status UncompressBlockContentsForCompressionType(
break;
case kLZ4Compression:
ubuf.reset(LZ4_Uncompress(
uncompression_ctx, data, n, &decompress_size,
uncompression_info, data, n, &decompress_size,
GetCompressFormatForVersion(kLZ4Compression, format_version)));
if (!ubuf) {
static char lz4_corrupt_msg[] =
@ -348,7 +348,7 @@ Status UncompressBlockContentsForCompressionType(
break;
case kLZ4HCCompression:
ubuf.reset(LZ4_Uncompress(
uncompression_ctx, data, n, &decompress_size,
uncompression_info, data, n, &decompress_size,
GetCompressFormatForVersion(kLZ4HCCompression, format_version)));
if (!ubuf) {
static char lz4hc_corrupt_msg[] =
@ -370,7 +370,8 @@ Status UncompressBlockContentsForCompressionType(
break;
case kZSTD:
case kZSTDNotFinalCompression:
ubuf.reset(ZSTD_Uncompress(uncompression_ctx, data, n, &decompress_size));
ubuf.reset(
ZSTD_Uncompress(uncompression_info, data, n, &decompress_size));
if (!ubuf) {
static char zstd_corrupt_msg[] =
"ZSTD not supported or corrupted ZSTD compressed block contents";
@ -400,14 +401,14 @@ Status UncompressBlockContentsForCompressionType(
// buffer is returned via 'result' and it is upto the caller to
// free this buffer.
// format_version is the block format as defined in include/rocksdb/table.h
Status UncompressBlockContents(const UncompressionContext& uncompression_ctx,
Status UncompressBlockContents(const UncompressionInfo& uncompression_info,
const char* data, size_t n,
BlockContents* contents, uint32_t format_version,
const ImmutableCFOptions& ioptions) {
assert(data[n] != kNoCompression);
assert(data[n] == uncompression_ctx.type());
assert(data[n] == uncompression_info.type());
return UncompressBlockContentsForCompressionType(
uncompression_ctx, data, n, contents, format_version, ioptions);
uncompression_info, data, n, contents, format_version, ioptions);
}
} // namespace rocksdb

View File

@ -250,16 +250,17 @@ extern Status ReadBlockContents(
// free this buffer.
// For description of compress_format_version and possible values, see
// util/compression.h
extern Status UncompressBlockContents(
const UncompressionContext& uncompression_ctx, const char* data, size_t n,
BlockContents* contents, uint32_t compress_format_version,
const ImmutableCFOptions& ioptions);
extern Status UncompressBlockContents(const UncompressionInfo& info,
const char* data, size_t n,
BlockContents* contents,
uint32_t compress_format_version,
const ImmutableCFOptions& ioptions);
// This is an extension to UncompressBlockContents that accepts
// a specific compression type. This is used by un-wrapped blocks
// with no compression header.
extern Status UncompressBlockContentsForCompressionType(
const UncompressionContext& uncompression_ctx, const char* data, size_t n,
const UncompressionInfo& info, const char* data, size_t n,
BlockContents* contents, uint32_t compress_format_version,
const ImmutableCFOptions& ioptions);

View File

@ -1991,28 +1991,28 @@ class Benchmark {
return true;
}
inline bool CompressSlice(const CompressionContext& compression_ctx,
inline bool CompressSlice(const CompressionInfo& compression_info,
const Slice& input, std::string* compressed) {
bool ok = true;
switch (FLAGS_compression_type_e) {
case rocksdb::kSnappyCompression:
ok = Snappy_Compress(compression_ctx, input.data(), input.size(),
ok = Snappy_Compress(compression_info, input.data(), input.size(),
compressed);
break;
case rocksdb::kZlibCompression:
ok = Zlib_Compress(compression_ctx, 2, input.data(), input.size(),
ok = Zlib_Compress(compression_info, 2, input.data(), input.size(),
compressed);
break;
case rocksdb::kBZip2Compression:
ok = BZip2_Compress(compression_ctx, 2, input.data(), input.size(),
ok = BZip2_Compress(compression_info, 2, input.data(), input.size(),
compressed);
break;
case rocksdb::kLZ4Compression:
ok = LZ4_Compress(compression_ctx, 2, input.data(), input.size(),
ok = LZ4_Compress(compression_info, 2, input.data(), input.size(),
compressed);
break;
case rocksdb::kLZ4HCCompression:
ok = LZ4HC_Compress(compression_ctx, 2, input.data(), input.size(),
ok = LZ4HC_Compress(compression_info, 2, input.data(), input.size(),
compressed);
break;
case rocksdb::kXpressCompression:
@ -2020,7 +2020,7 @@ class Benchmark {
input.size(), compressed);
break;
case rocksdb::kZSTD:
ok = ZSTD_Compress(compression_ctx, input.data(), input.size(),
ok = ZSTD_Compress(compression_info, input.data(), input.size(),
compressed);
break;
default:
@ -2103,10 +2103,11 @@ class Benchmark {
const int len = FLAGS_block_size;
std::string input_str(len, 'y');
std::string compressed;
CompressionContext compression_ctx(FLAGS_compression_type_e,
Options().compression_opts);
bool result =
CompressSlice(compression_ctx, Slice(input_str), &compressed);
CompressionOptions opts;
CompressionContext context(FLAGS_compression_type_e);
CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
FLAGS_compression_type_e);
bool result = CompressSlice(info, Slice(input_str), &compressed);
if (!result) {
fprintf(stdout, "WARNING: %s compression is not enabled\n",
@ -2956,13 +2957,14 @@ void VerifyDBFromDB(std::string& truth_db_name) {
int64_t produced = 0;
bool ok = true;
std::string compressed;
CompressionContext compression_ctx(FLAGS_compression_type_e,
Options().compression_opts);
CompressionOptions opts;
CompressionContext context(FLAGS_compression_type_e);
CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
FLAGS_compression_type_e);
// Compress 1G
while (ok && bytes < int64_t(1) << 30) {
compressed.clear();
ok = CompressSlice(compression_ctx, input, &compressed);
ok = CompressSlice(info, input, &compressed);
produced += compressed.size();
bytes += input.size();
thread->stats.FinishedOps(nullptr, nullptr, 1, kCompress);
@ -2984,11 +2986,17 @@ void VerifyDBFromDB(std::string& truth_db_name) {
Slice input = gen.Generate(FLAGS_block_size);
std::string compressed;
CompressionContext compression_ctx(FLAGS_compression_type_e);
CompressionOptions compression_opts;
CompressionInfo compression_info(compression_opts, compression_ctx,
CompressionDict::GetEmptyDict(),
FLAGS_compression_type_e);
UncompressionContext uncompression_ctx(FLAGS_compression_type_e);
CompressionContext compression_ctx(FLAGS_compression_type_e,
Options().compression_opts);
UncompressionInfo uncompression_info(uncompression_ctx,
CompressionDict::GetEmptyDict(),
FLAGS_compression_type_e);
bool ok = CompressSlice(compression_ctx, input, &compressed);
bool ok = CompressSlice(compression_info, input, &compressed);
int64_t bytes = 0;
int decompress_size;
while (ok && bytes < 1024 * 1048576) {
@ -3008,7 +3016,7 @@ void VerifyDBFromDB(std::string& truth_db_name) {
break;
}
case rocksdb::kZlibCompression:
uncompressed = Zlib_Uncompress(uncompression_ctx, compressed.data(),
uncompressed = Zlib_Uncompress(uncompression_info, compressed.data(),
compressed.size(), &decompress_size, 2);
ok = uncompressed != nullptr;
break;
@ -3018,12 +3026,12 @@ void VerifyDBFromDB(std::string& truth_db_name) {
ok = uncompressed != nullptr;
break;
case rocksdb::kLZ4Compression:
uncompressed = LZ4_Uncompress(uncompression_ctx, compressed.data(),
uncompressed = LZ4_Uncompress(uncompression_info, compressed.data(),
compressed.size(), &decompress_size, 2);
ok = uncompressed != nullptr;
break;
case rocksdb::kLZ4HCCompression:
uncompressed = LZ4_Uncompress(uncompression_ctx, compressed.data(),
uncompressed = LZ4_Uncompress(uncompression_info, compressed.data(),
compressed.size(), &decompress_size, 2);
ok = uncompressed != nullptr;
break;
@ -3033,7 +3041,7 @@ void VerifyDBFromDB(std::string& truth_db_name) {
ok = uncompressed != nullptr;
break;
case rocksdb::kZSTD:
uncompressed = ZSTD_Uncompress(uncompression_ctx, compressed.data(),
uncompressed = ZSTD_Uncompress(uncompression_info, compressed.data(),
compressed.size(), &decompress_size);
ok = uncompressed != nullptr;
break;

View File

@ -133,13 +133,147 @@ class ZSTDUncompressCachedData {
namespace rocksdb {
// Instantiate this class and pass it to the uncompression API below
// Holds dictionary and related data, like ZSTD's digested dictionary.
struct CompressionDict {
enum class Mode {
kUninit,
kEmpty, // An empty one can be used for both compression and uncompression
kCompression,
kUncompression,
};
#if ZSTD_VERSION_NUMBER >= 700
union {
ZSTD_CDict* zstd_cdict_;
ZSTD_DDict* zstd_ddict_;
};
#endif // ZSTD_VERSION_NUMBER >= 700
Mode mode_ = Mode::kUninit;
Slice dict_;
public:
static const CompressionDict& GetEmptyDict() {
static CompressionDict empty_dict{};
static bool init = false;
if (!init) {
empty_dict.Init(Slice() /* dict */, Mode::kEmpty,
false /* use_zstd_trainer */);
init = true;
}
return empty_dict;
}
void Init(Slice dict, Mode mode, CompressionType type, int level = -1) {
return Init(dict, mode, type == kZSTD || type == kZSTDNotFinalCompression,
level);
}
private:
#if ZSTD_VERSION_NUMBER >= 700
void Init(Slice dict, Mode mode, bool use_zstd_trainer, int level = -1) {
#else // ZSTD_VERSION_NUMBER >= 700
void Init(Slice dict, Mode mode, bool /* use_zstd_trainer */,
int /*level*/ = -1) {
#endif // ZSTD_VERSION_NUMBER >= 700
assert(mode_ == Mode::kUninit);
dict_ = std::move(dict);
mode_ = mode;
switch (mode) {
case Mode::kUninit:
assert(false);
break;
case Mode::kEmpty:
break;
case Mode::kCompression:
#if ZSTD_VERSION_NUMBER >= 700
zstd_cdict_ = nullptr;
if (!dict_.empty() && use_zstd_trainer) {
if (level == CompressionOptions::kDefaultCompressionLevel) {
// 3 is the value of ZSTD_CLEVEL_DEFAULT (not exposed publicly), see
// https://github.com/facebook/zstd/issues/1148
level = 3;
}
// Should be safe (but slower) if below call fails as we'll use the
// raw dictionary to compress.
zstd_cdict_ = ZSTD_createCDict(dict_.data(), dict_.size(), level);
assert(zstd_cdict_ != nullptr);
}
#endif // ZSTD_VERSION_NUMBER >= 700
break;
case Mode::kUncompression:
#if ZSTD_VERSION_NUMBER >= 700
zstd_ddict_ = nullptr;
if (!dict_.empty() && use_zstd_trainer) {
zstd_ddict_ = ZSTD_createDDict(dict_.data(), dict_.size());
assert(zstd_ddict_ != nullptr);
}
#endif // ZSTD_VERSION_NUMBER >= 700
break;
}
}
public:
~CompressionDict() {
#if ZSTD_VERSION_NUMBER >= 700
size_t res = 0;
switch (mode_) {
case Mode::kUninit:
break;
case Mode::kEmpty:
break;
case Mode::kCompression:
if (zstd_cdict_ != nullptr) {
res = ZSTD_freeCDict(zstd_cdict_);
}
break;
case Mode::kUncompression:
if (zstd_ddict_ != nullptr) {
res = ZSTD_freeDDict(zstd_ddict_);
}
break;
}
assert(res == 0); // Last I checked they can't fail
(void)res; // prevent unused var warning
#endif // ZSTD_VERSION_NUMBER >= 700
}
#if ZSTD_VERSION_NUMBER >= 700
const ZSTD_CDict* GetDigestedZstdCDict() const {
assert(mode_ != Mode::kUninit);
if (mode_ == Mode::kEmpty) {
return nullptr;
}
assert(mode_ == Mode::kCompression);
return zstd_cdict_;
}
const ZSTD_DDict* GetDigestedZstdDDict() const {
assert(mode_ != Mode::kUninit);
if (mode_ == Mode::kEmpty) {
return nullptr;
}
assert(mode_ == Mode::kUncompression);
return zstd_ddict_;
}
#endif // ZSTD_VERSION_NUMBER >= 700
Slice GetRawDict() const {
assert(mode_ != Mode::kUninit);
assert(mode_ != Mode::kEmpty || dict_.empty());
return dict_;
}
CompressionDict() = default;
// Disable copy/move
CompressionDict(const CompressionDict&) = delete;
CompressionDict& operator=(const CompressionDict&) = delete;
CompressionDict(CompressionDict&&) = delete;
CompressionDict& operator=(CompressionDict&&) = delete;
};
class CompressionContext {
#if defined(ZSTD) && (ZSTD_VERSION_NUMBER >= 500)
private:
const CompressionType type_;
const CompressionOptions opts_;
Slice dict_;
#if defined(ZSTD) && (ZSTD_VERSION_NUMBER >= 500)
ZSTD_CCtx* zstd_ctx_ = nullptr;
void CreateNativeContext() {
if (type_ == kZSTD || type_ == kZSTDNotFinalCompression) {
@ -163,35 +297,45 @@ class CompressionContext {
assert(type_ == kZSTD || type_ == kZSTDNotFinalCompression);
return zstd_ctx_;
}
explicit CompressionContext(CompressionType comp_type) : type_(comp_type) {
CreateNativeContext();
}
#else // ZSTD && (ZSTD_VERSION_NUMBER >= 500)
public:
explicit CompressionContext(CompressionType /* comp_type */) {}
private:
void CreateNativeContext() {}
void DestroyNativeContext() {}
#endif // ZSTD && (ZSTD_VERSION_NUMBER >= 500)
public:
explicit CompressionContext(CompressionType comp_type) : type_(comp_type) {
CreateNativeContext();
}
CompressionContext(CompressionType comp_type, const CompressionOptions& opts,
const Slice& comp_dict = Slice())
: type_(comp_type), opts_(opts), dict_(comp_dict) {
CreateNativeContext();
}
~CompressionContext() { DestroyNativeContext(); }
CompressionContext(const CompressionContext&) = delete;
CompressionContext& operator=(const CompressionContext&) = delete;
const CompressionOptions& options() const { return opts_; }
CompressionType type() const { return type_; }
const Slice& dict() const { return dict_; }
Slice& dict() { return dict_; }
};
// Instantiate this class and pass it to the uncompression API below
class CompressionInfo {
const CompressionOptions& opts_;
const CompressionContext& context_;
const CompressionDict& dict_;
const CompressionType type_;
public:
CompressionInfo(const CompressionOptions& _opts,
const CompressionContext& _context,
const CompressionDict& _dict, CompressionType _type)
: opts_(_opts), context_(_context), dict_(_dict), type_(_type) {}
const CompressionOptions& options() const { return opts_; }
const CompressionContext& context() const { return context_; }
const CompressionDict& dict() const { return dict_; }
CompressionType type() const { return type_; }
};
class UncompressionContext {
private:
CompressionType type_;
Slice dict_;
const CompressionType type_;
CompressionContextCache* ctx_cache_ = nullptr;
ZSTDUncompressCachedData uncomp_cached_data_;
@ -199,10 +343,8 @@ class UncompressionContext {
struct NoCache {};
// Do not use context cache, used by TableBuilder
UncompressionContext(NoCache, CompressionType comp_type) : type_(comp_type) {}
explicit UncompressionContext(CompressionType comp_type)
: UncompressionContext(comp_type, Slice()) {}
UncompressionContext(CompressionType comp_type, const Slice& comp_dict)
: type_(comp_type), dict_(comp_dict) {
explicit UncompressionContext(CompressionType comp_type) : type_(comp_type) {
if (type_ == kZSTD || type_ == kZSTDNotFinalCompression) {
ctx_cache_ = CompressionContextCache::Instance();
uncomp_cached_data_ = ctx_cache_->GetCachedZSTDUncompressData();
@ -222,9 +364,21 @@ class UncompressionContext {
ZSTDUncompressCachedData::ZSTDNativeContext GetZSTDContext() const {
return uncomp_cached_data_.Get();
}
};
class UncompressionInfo {
const UncompressionContext& context_;
const CompressionDict& dict_;
const CompressionType type_;
public:
UncompressionInfo(const UncompressionContext& _context,
const CompressionDict& _dict, CompressionType _type)
: context_(_context), dict_(_dict), type_(_type) {}
const UncompressionContext& context() const { return context_; }
const CompressionDict& dict() const { return dict_; }
CompressionType type() const { return type_; }
const Slice& dict() const { return dict_; }
Slice& dict() { return dict_; }
};
inline bool Snappy_Supported() {
@ -343,9 +497,8 @@ inline std::string CompressionTypeToString(CompressionType compression_type) {
// 2 -- Zlib, BZip2 and LZ4 encode decompressed size as Varint32 just before the
// start of compressed block. Snappy format is the same as version 1.
inline bool Snappy_Compress(const CompressionContext& /*ctx*/,
const char* input, size_t length,
::std::string* output) {
inline bool Snappy_Compress(const CompressionInfo& /*info*/, const char* input,
size_t length, ::std::string* output) {
#ifdef SNAPPY
output->resize(snappy::MaxCompressedLength(length));
size_t outlen;
@ -410,7 +563,7 @@ inline bool GetDecompressedSizeInfo(const char** input_data,
// header in varint32 format
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline bool Zlib_Compress(const CompressionContext& ctx,
inline bool Zlib_Compress(const CompressionInfo& info,
uint32_t compress_format_version, const char* input,
size_t length, ::std::string* output) {
#ifdef ZLIB
@ -435,24 +588,25 @@ inline bool Zlib_Compress(const CompressionContext& ctx,
// The default value is 8. See zconf.h for more details.
static const int memLevel = 8;
int level;
if (ctx.options().level == CompressionOptions::kDefaultCompressionLevel) {
if (info.options().level == CompressionOptions::kDefaultCompressionLevel) {
level = Z_DEFAULT_COMPRESSION;
} else {
level = ctx.options().level;
level = info.options().level;
}
z_stream _stream;
memset(&_stream, 0, sizeof(z_stream));
int st = deflateInit2(&_stream, level, Z_DEFLATED, ctx.options().window_bits,
memLevel, ctx.options().strategy);
int st = deflateInit2(&_stream, level, Z_DEFLATED, info.options().window_bits,
memLevel, info.options().strategy);
if (st != Z_OK) {
return false;
}
if (ctx.dict().size()) {
Slice compression_dict = info.dict().GetRawDict();
if (compression_dict.size()) {
// Initialize the compression library's dictionary
st = deflateSetDictionary(&_stream,
reinterpret_cast<const Bytef*>(ctx.dict().data()),
static_cast<unsigned int>(ctx.dict().size()));
st = deflateSetDictionary(
&_stream, reinterpret_cast<const Bytef*>(compression_dict.data()),
static_cast<unsigned int>(compression_dict.size()));
if (st != Z_OK) {
deflateEnd(&_stream);
return false;
@ -480,7 +634,7 @@ inline bool Zlib_Compress(const CompressionContext& ctx,
deflateEnd(&_stream);
return compressed;
#else
(void)ctx;
(void)info;
(void)compress_format_version;
(void)input;
(void)length;
@ -495,7 +649,7 @@ inline bool Zlib_Compress(const CompressionContext& ctx,
// header in varint32 format
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline char* Zlib_Uncompress(const UncompressionContext& ctx,
inline char* Zlib_Uncompress(const UncompressionInfo& info,
const char* input_data, size_t input_length,
int* decompress_size,
uint32_t compress_format_version,
@ -528,11 +682,12 @@ inline char* Zlib_Uncompress(const UncompressionContext& ctx,
return nullptr;
}
if (ctx.dict().size()) {
Slice compression_dict = info.dict().GetRawDict();
if (compression_dict.size()) {
// Initialize the compression library's dictionary
st = inflateSetDictionary(&_stream,
reinterpret_cast<const Bytef*>(ctx.dict().data()),
static_cast<unsigned int>(ctx.dict().size()));
st = inflateSetDictionary(
&_stream, reinterpret_cast<const Bytef*>(compression_dict.data()),
static_cast<unsigned int>(compression_dict.size()));
if (st != Z_OK) {
return nullptr;
}
@ -585,7 +740,7 @@ inline char* Zlib_Uncompress(const UncompressionContext& ctx,
inflateEnd(&_stream);
return output;
#else
(void)ctx;
(void)info;
(void)input_data;
(void)input_length;
(void)decompress_size;
@ -599,7 +754,7 @@ inline char* Zlib_Uncompress(const UncompressionContext& ctx,
// block header
// compress_format_version == 2 -- decompressed size is included in the block
// header in varint32 format
inline bool BZip2_Compress(const CompressionContext& /*ctx*/,
inline bool BZip2_Compress(const CompressionInfo& /*info*/,
uint32_t compress_format_version, const char* input,
size_t length, ::std::string* output) {
#ifdef BZIP2
@ -746,7 +901,7 @@ inline char* BZip2_Uncompress(const char* input_data, size_t input_length,
// header in varint32 format
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline bool LZ4_Compress(const CompressionContext& ctx,
inline bool LZ4_Compress(const CompressionInfo& info,
uint32_t compress_format_version, const char* input,
size_t length, ::std::string* output) {
#ifdef LZ4
@ -774,9 +929,10 @@ inline bool LZ4_Compress(const CompressionContext& ctx,
int outlen;
#if LZ4_VERSION_NUMBER >= 10400 // r124+
LZ4_stream_t* stream = LZ4_createStream();
if (ctx.dict().size()) {
LZ4_loadDict(stream, ctx.dict().data(),
static_cast<int>(ctx.dict().size()));
Slice compression_dict = info.dict().GetRawDict();
if (compression_dict.size()) {
LZ4_loadDict(stream, compression_dict.data(),
static_cast<int>(compression_dict.size()));
}
#if LZ4_VERSION_NUMBER >= 10700 // r129+
outlen =
@ -799,7 +955,7 @@ inline bool LZ4_Compress(const CompressionContext& ctx,
output->resize(static_cast<size_t>(output_header_len + outlen));
return true;
#else // LZ4
(void)ctx;
(void)info;
(void)compress_format_version;
(void)input;
(void)length;
@ -814,7 +970,7 @@ inline bool LZ4_Compress(const CompressionContext& ctx,
// header in varint32 format
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline char* LZ4_Uncompress(const UncompressionContext& ctx,
inline char* LZ4_Uncompress(const UncompressionInfo& info,
const char* input_data, size_t input_length,
int* decompress_size,
uint32_t compress_format_version) {
@ -840,9 +996,10 @@ inline char* LZ4_Uncompress(const UncompressionContext& ctx,
char* output = new char[output_len];
#if LZ4_VERSION_NUMBER >= 10400 // r124+
LZ4_streamDecode_t* stream = LZ4_createStreamDecode();
if (ctx.dict().size()) {
LZ4_setStreamDecode(stream, ctx.dict().data(),
static_cast<int>(ctx.dict().size()));
Slice compression_dict = info.dict().GetRawDict();
if (compression_dict.size()) {
LZ4_setStreamDecode(stream, compression_dict.data(),
static_cast<int>(compression_dict.size()));
}
*decompress_size = LZ4_decompress_safe_continue(
stream, input_data, output, static_cast<int>(input_length),
@ -861,7 +1018,7 @@ inline char* LZ4_Uncompress(const UncompressionContext& ctx,
assert(*decompress_size == static_cast<int>(output_len));
return output;
#else // LZ4
(void)ctx;
(void)info;
(void)input_data;
(void)input_length;
(void)decompress_size;
@ -876,7 +1033,7 @@ inline char* LZ4_Uncompress(const UncompressionContext& ctx,
// header in varint32 format
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline bool LZ4HC_Compress(const CompressionContext& ctx,
inline bool LZ4HC_Compress(const CompressionInfo& info,
uint32_t compress_format_version, const char* input,
size_t length, ::std::string* output) {
#ifdef LZ4
@ -903,17 +1060,18 @@ inline bool LZ4HC_Compress(const CompressionContext& ctx,
int outlen;
int level;
if (ctx.options().level == CompressionOptions::kDefaultCompressionLevel) {
if (info.options().level == CompressionOptions::kDefaultCompressionLevel) {
level = 0; // lz4hc.h says any value < 1 will be sanitized to default
} else {
level = ctx.options().level;
level = info.options().level;
}
#if LZ4_VERSION_NUMBER >= 10400 // r124+
LZ4_streamHC_t* stream = LZ4_createStreamHC();
LZ4_resetStreamHC(stream, level);
Slice compression_dict = info.dict().GetRawDict();
const char* compression_dict_data =
ctx.dict().size() > 0 ? ctx.dict().data() : nullptr;
size_t compression_dict_size = ctx.dict().size();
compression_dict.size() > 0 ? compression_dict.data() : nullptr;
size_t compression_dict_size = compression_dict.size();
LZ4_loadDictHC(stream, compression_dict_data,
static_cast<int>(compression_dict_size));
@ -944,7 +1102,7 @@ inline bool LZ4HC_Compress(const CompressionContext& ctx,
output->resize(static_cast<size_t>(output_header_len + outlen));
return true;
#else // LZ4
(void)ctx;
(void)info;
(void)compress_format_version;
(void)input;
(void)length;
@ -978,9 +1136,7 @@ inline char* XPRESS_Uncompress(const char* /*input_data*/,
}
#endif
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline bool ZSTD_Compress(const CompressionContext& ctx, const char* input,
inline bool ZSTD_Compress(const CompressionInfo& info, const char* input,
size_t length, ::std::string* output) {
#ifdef ZSTD
if (length > std::numeric_limits<uint32_t>::max()) {
@ -995,19 +1151,29 @@ inline bool ZSTD_Compress(const CompressionContext& ctx, const char* input,
output->resize(static_cast<size_t>(output_header_len + compressBound));
size_t outlen = 0;
int level;
if (ctx.options().level == CompressionOptions::kDefaultCompressionLevel) {
if (info.options().level == CompressionOptions::kDefaultCompressionLevel) {
// 3 is the value of ZSTD_CLEVEL_DEFAULT (not exposed publicly), see
// https://github.com/facebook/zstd/issues/1148
level = 3;
} else {
level = ctx.options().level;
level = info.options().level;
}
#if ZSTD_VERSION_NUMBER >= 500 // v0.5.0+
ZSTD_CCtx* context = ctx.ZSTDPreallocCtx();
ZSTD_CCtx* context = info.context().ZSTDPreallocCtx();
assert(context != nullptr);
outlen = ZSTD_compress_usingDict(context, &(*output)[output_header_len],
compressBound, input, length,
ctx.dict().data(), ctx.dict().size(), level);
#if ZSTD_VERSION_NUMBER >= 700 // v0.7.0+
if (info.dict().GetDigestedZstdCDict() != nullptr) {
outlen = ZSTD_compress_usingCDict(context, &(*output)[output_header_len],
compressBound, input, length,
info.dict().GetDigestedZstdCDict());
}
#endif // ZSTD_VERSION_NUMBER >= 700
if (outlen == 0) {
outlen = ZSTD_compress_usingDict(context, &(*output)[output_header_len],
compressBound, input, length,
info.dict().GetRawDict().data(),
info.dict().GetRawDict().size(), level);
}
#else // up to v0.4.x
outlen = ZSTD_compress(&(*output)[output_header_len], compressBound, input,
length, level);
@ -1018,7 +1184,7 @@ inline bool ZSTD_Compress(const CompressionContext& ctx, const char* input,
output->resize(output_header_len + outlen);
return true;
#else // ZSTD
(void)ctx;
(void)info;
(void)input;
(void)length;
(void)output;
@ -1028,7 +1194,7 @@ inline bool ZSTD_Compress(const CompressionContext& ctx, const char* input,
// @param compression_dict Data for presetting the compression library's
// dictionary.
inline char* ZSTD_Uncompress(const UncompressionContext& ctx,
inline char* ZSTD_Uncompress(const UncompressionInfo& info,
const char* input_data, size_t input_length,
int* decompress_size) {
#ifdef ZSTD
@ -1039,14 +1205,24 @@ inline char* ZSTD_Uncompress(const UncompressionContext& ctx,
}
char* output = new char[output_len];
size_t actual_output_length;
size_t actual_output_length = 0;
#if ZSTD_VERSION_NUMBER >= 500 // v0.5.0+
ZSTD_DCtx* context = ctx.GetZSTDContext();
ZSTD_DCtx* context = info.context().GetZSTDContext();
assert(context != nullptr);
actual_output_length = ZSTD_decompress_usingDict(
context, output, output_len, input_data, input_length, ctx.dict().data(),
ctx.dict().size());
#if ZSTD_VERSION_NUMBER >= 700 // v0.7.0+
if (info.dict().GetDigestedZstdDDict() != nullptr) {
actual_output_length = ZSTD_decompress_usingDDict(
context, output, output_len, input_data, input_length,
info.dict().GetDigestedZstdDDict());
}
#endif // ZSTD_VERSION_NUMBER >= 700
if (actual_output_length == 0) {
actual_output_length = ZSTD_decompress_usingDict(
context, output, output_len, input_data, input_length,
info.dict().GetRawDict().data(), info.dict().GetRawDict().size());
}
#else // up to v0.4.x
(void) info;
actual_output_length =
ZSTD_decompress(output, output_len, input_data, input_length);
#endif // ZSTD_VERSION_NUMBER >= 500
@ -1054,7 +1230,7 @@ inline char* ZSTD_Uncompress(const UncompressionContext& ctx,
*decompress_size = static_cast<int>(actual_output_length);
return output;
#else // ZSTD
(void)ctx;
(void)info;
(void)input_data;
(void)input_length;
(void)decompress_size;
@ -1069,6 +1245,10 @@ inline std::string ZSTD_TrainDictionary(const std::string& samples,
// available for dynamic linking until v1.1.3. For now we enable the feature
// in v1.1.3+ only.
#if ZSTD_VERSION_NUMBER >= 10103 // v1.1.3+
assert(samples.empty() == sample_lens.empty());
if (samples.empty()) {
return "";
}
std::string dict_data(max_dict_bytes, '\0');
size_t dict_len = ZDICT_trainFromBuffer(
&dict_data[0], max_dict_bytes, &samples[0], &sample_lens[0],

View File

@ -771,9 +771,11 @@ Slice BlobDBImpl::GetCompressedSlice(const Slice& raw,
return raw;
}
StopWatch compression_sw(env_, statistics_, BLOB_DB_COMPRESSION_MICROS);
CompressionType ct = bdb_options_.compression;
CompressionContext compression_ctx(ct);
CompressBlock(raw, compression_ctx, &ct, kBlockBasedTableVersionFormat,
CompressionType type = bdb_options_.compression;
CompressionOptions opts;
CompressionContext context(type);
CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(), type);
CompressBlock(raw, info, &type, kBlockBasedTableVersionFormat,
compression_output);
return *compression_output;
}
@ -1106,9 +1108,11 @@ Status BlobDBImpl::GetBlobValue(const Slice& key, const Slice& index_entry,
{
StopWatch decompression_sw(env_, statistics_,
BLOB_DB_DECOMPRESSION_MICROS);
UncompressionContext uncompression_ctx(bfile->compression());
UncompressionContext context(bfile->compression());
UncompressionInfo info(context, CompressionDict::GetEmptyDict(),
bfile->compression());
s = UncompressBlockContentsForCompressionType(
uncompression_ctx, blob_value.data(), blob_value.size(), &contents,
info, blob_value.data(), blob_value.size(), &contents,
kBlockBasedTableVersionFormat, *(cfh->cfd()->ioptions()));
}
value->PinSelf(contents.data);

View File

@ -208,9 +208,11 @@ Status BlobDumpTool::DumpRecord(DisplayType show_key, DisplayType show_blob,
if (compression != kNoCompression &&
(show_uncompressed_blob != DisplayType::kNone || show_summary)) {
BlockContents contents;
UncompressionContext uncompression_ctx(compression);
UncompressionContext context(compression);
UncompressionInfo info(context, CompressionDict::GetEmptyDict(),
compression);
s = UncompressBlockContentsForCompressionType(
uncompression_ctx, slice.data() + key_size, value_size, &contents,
info, slice.data() + key_size, value_size, &contents,
2 /*compress_format_version*/, ImmutableCFOptions(Options()));
if (!s.ok()) {
return s;

View File

@ -89,8 +89,9 @@ void ColumnAwareEncodingReader::DecodeBlocks(
CompressionType type =
(CompressionType)slice_final_with_bit[slice_final_with_bit.size() - 1];
if (type != kNoCompression) {
UncompressionContext uncompression_ctx(type);
UncompressBlockContents(uncompression_ctx, slice_final_with_bit.c_str(),
UncompressionContext context(type);
UncompressionInfo info(context, CompressionDict::GetEmptyDict(), type);
UncompressBlockContents(info, slice_final_with_bit.c_str(),
slice_final_with_bit.size() - 1, &contents,
format_version, ioptions);
content_ptr = contents.data.data();
@ -171,8 +172,9 @@ void ColumnAwareEncodingReader::DecodeBlocksFromRowFormat(
CompressionType type =
(CompressionType)slice_final_with_bit[slice_final_with_bit.size() - 1];
if (type != kNoCompression) {
UncompressionContext uncompression_ctx(type);
UncompressBlockContents(uncompression_ctx, slice_final_with_bit.c_str(),
UncompressionContext context(type);
UncompressionInfo info(context, CompressionDict::GetEmptyDict(), type);
UncompressBlockContents(info, slice_final_with_bit.c_str(),
slice_final_with_bit.size() - 1, &contents,
format_version, ioptions);
decoded_content = std::string(contents.data.data(), contents.data.size());
@ -243,10 +245,12 @@ namespace {
void CompressDataBlock(const std::string& output_content, Slice* slice_final,
CompressionType* type, std::string* compressed_output) {
CompressionContext compression_ctx(*type);
CompressionContext context(*type);
CompressionOptions opts;
CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(), *type);
uint32_t format_version = 2; // hard-coded version
*slice_final = CompressBlock(output_content, compression_ctx, type,
format_version, compressed_output);
*slice_final = CompressBlock(output_content, info, type, format_version,
compressed_output);
}
} // namespace