2013-10-16 23:59:46 +02:00
|
|
|
// 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.
|
|
|
|
//
|
2011-03-18 23:37:00 +01:00
|
|
|
// 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/format.h"
|
|
|
|
|
2014-02-05 01:21:47 +01:00
|
|
|
#include <string>
|
2014-02-08 04:26:49 +01:00
|
|
|
#include <inttypes.h>
|
2014-02-05 01:21:47 +01:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "port/port.h"
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
#include "rocksdb/env.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "table/block.h"
|
|
|
|
#include "util/coding.h"
|
|
|
|
#include "util/crc32c.h"
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
#include "util/perf_context_imp.h"
|
2014-05-01 20:09:32 +02:00
|
|
|
#include "util/xxhash.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
namespace rocksdb {
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2014-05-01 20:09:32 +02:00
|
|
|
extern const uint64_t kLegacyBlockBasedTableMagicNumber;
|
|
|
|
extern const uint64_t kBlockBasedTableMagicNumber;
|
2014-05-08 02:45:27 +02:00
|
|
|
|
|
|
|
#ifndef ROCKSDB_LITE
|
|
|
|
extern const uint64_t kLegacyPlainTableMagicNumber;
|
2014-05-01 20:09:32 +02:00
|
|
|
extern const uint64_t kPlainTableMagicNumber;
|
2014-05-08 02:45:27 +02:00
|
|
|
#else
|
|
|
|
// ROCKSDB_LITE doesn't have plain table
|
|
|
|
const uint64_t kLegacyPlainTableMagicNumber = 0;
|
|
|
|
const uint64_t kPlainTableMagicNumber = 0;
|
|
|
|
#endif
|
2014-07-31 08:11:59 +02:00
|
|
|
const uint32_t DefaultStackBufferSize = 5000;
|
2014-05-01 20:09:32 +02:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
void BlockHandle::EncodeTo(std::string* dst) const {
|
|
|
|
// Sanity check that all fields have been set
|
|
|
|
assert(offset_ != ~static_cast<uint64_t>(0));
|
|
|
|
assert(size_ != ~static_cast<uint64_t>(0));
|
|
|
|
PutVarint64(dst, offset_);
|
|
|
|
PutVarint64(dst, size_);
|
|
|
|
}
|
|
|
|
|
|
|
|
Status BlockHandle::DecodeFrom(Slice* input) {
|
|
|
|
if (GetVarint64(input, &offset_) &&
|
|
|
|
GetVarint64(input, &size_)) {
|
|
|
|
return Status::OK();
|
|
|
|
} else {
|
|
|
|
return Status::Corruption("bad block handle");
|
|
|
|
}
|
|
|
|
}
|
2013-12-05 00:43:09 +01:00
|
|
|
const BlockHandle BlockHandle::kNullBlockHandle(0, 0);
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2014-05-01 20:09:32 +02:00
|
|
|
// legacy footer format:
|
|
|
|
// metaindex handle (varint64 offset, varint64 size)
|
|
|
|
// index handle (varint64 offset, varint64 size)
|
|
|
|
// <padding> to make the total size 2 * BlockHandle::kMaxEncodedLength
|
|
|
|
// table_magic_number (8 bytes)
|
|
|
|
// new footer format:
|
|
|
|
// checksum (char, 1 byte)
|
|
|
|
// metaindex handle (varint64 offset, varint64 size)
|
|
|
|
// index handle (varint64 offset, varint64 size)
|
|
|
|
// <padding> to make the total size 2 * BlockHandle::kMaxEncodedLength + 1
|
|
|
|
// footer version (4 bytes)
|
|
|
|
// table_magic_number (8 bytes)
|
2011-03-18 23:37:00 +01:00
|
|
|
void Footer::EncodeTo(std::string* dst) const {
|
2014-05-01 20:09:32 +02:00
|
|
|
if (version() == kLegacyFooter) {
|
|
|
|
// has to be default checksum with legacy footer
|
|
|
|
assert(checksum_ == kCRC32c);
|
|
|
|
const size_t original_size = dst->size();
|
|
|
|
metaindex_handle_.EncodeTo(dst);
|
|
|
|
index_handle_.EncodeTo(dst);
|
|
|
|
dst->resize(original_size + 2 * BlockHandle::kMaxEncodedLength); // Padding
|
|
|
|
PutFixed32(dst, static_cast<uint32_t>(table_magic_number() & 0xffffffffu));
|
|
|
|
PutFixed32(dst, static_cast<uint32_t>(table_magic_number() >> 32));
|
|
|
|
assert(dst->size() == original_size + kVersion0EncodedLength);
|
|
|
|
} else {
|
|
|
|
const size_t original_size = dst->size();
|
|
|
|
dst->push_back(static_cast<char>(checksum_));
|
|
|
|
metaindex_handle_.EncodeTo(dst);
|
|
|
|
index_handle_.EncodeTo(dst);
|
|
|
|
dst->resize(original_size + kVersion1EncodedLength - 12); // Padding
|
|
|
|
PutFixed32(dst, kFooterVersion);
|
|
|
|
PutFixed32(dst, static_cast<uint32_t>(table_magic_number() & 0xffffffffu));
|
|
|
|
PutFixed32(dst, static_cast<uint32_t>(table_magic_number() >> 32));
|
|
|
|
assert(dst->size() == original_size + kVersion1EncodedLength);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
inline bool IsLegacyFooterFormat(uint64_t magic_number) {
|
|
|
|
return magic_number == kLegacyBlockBasedTableMagicNumber ||
|
|
|
|
magic_number == kLegacyPlainTableMagicNumber;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline uint64_t UpconvertLegacyFooterFormat(uint64_t magic_number) {
|
|
|
|
if (magic_number == kLegacyBlockBasedTableMagicNumber) {
|
|
|
|
return kBlockBasedTableMagicNumber;
|
|
|
|
}
|
|
|
|
if (magic_number == kLegacyPlainTableMagicNumber) {
|
|
|
|
return kPlainTableMagicNumber;
|
|
|
|
}
|
|
|
|
assert(false);
|
2014-05-01 21:41:49 +02:00
|
|
|
return 0;
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
2014-05-01 20:09:32 +02:00
|
|
|
} // namespace
|
|
|
|
|
|
|
|
Footer::Footer(uint64_t table_magic_number)
|
|
|
|
: version_(IsLegacyFooterFormat(table_magic_number) ? kLegacyFooter
|
|
|
|
: kFooterVersion),
|
|
|
|
checksum_(kCRC32c),
|
|
|
|
table_magic_number_(table_magic_number) {}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
Status Footer::DecodeFrom(Slice* input) {
|
2013-03-01 03:04:58 +01:00
|
|
|
assert(input != nullptr);
|
2014-05-01 20:09:32 +02:00
|
|
|
assert(input->size() >= kMinEncodedLength);
|
2013-01-09 19:44:30 +01:00
|
|
|
|
2014-05-01 20:09:32 +02:00
|
|
|
const char *magic_ptr =
|
|
|
|
input->data() + input->size() - kMagicNumberLengthByte;
|
2011-03-18 23:37:00 +01:00
|
|
|
const uint32_t magic_lo = DecodeFixed32(magic_ptr);
|
|
|
|
const uint32_t magic_hi = DecodeFixed32(magic_ptr + 4);
|
2014-05-01 20:09:32 +02:00
|
|
|
uint64_t magic = ((static_cast<uint64_t>(magic_hi) << 32) |
|
|
|
|
(static_cast<uint64_t>(magic_lo)));
|
|
|
|
|
|
|
|
// We check for legacy formats here and silently upconvert them
|
|
|
|
bool legacy = IsLegacyFooterFormat(magic);
|
|
|
|
if (legacy) {
|
|
|
|
magic = UpconvertLegacyFooterFormat(magic);
|
|
|
|
}
|
2014-02-05 01:21:47 +01:00
|
|
|
if (HasInitializedTableMagicNumber()) {
|
|
|
|
if (magic != table_magic_number()) {
|
|
|
|
char buffer[80];
|
|
|
|
snprintf(buffer, sizeof(buffer) - 1,
|
2014-02-08 04:47:48 +01:00
|
|
|
"not an sstable (bad magic number --- %lx)",
|
|
|
|
(long)magic);
|
2014-02-05 01:21:47 +01:00
|
|
|
return Status::InvalidArgument(buffer);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
set_table_magic_number(magic);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2014-05-01 20:09:32 +02:00
|
|
|
if (legacy) {
|
|
|
|
// The size is already asserted to be at least kMinEncodedLength
|
|
|
|
// at the beginning of the function
|
|
|
|
input->remove_prefix(input->size() - kVersion0EncodedLength);
|
|
|
|
version_ = kLegacyFooter;
|
|
|
|
checksum_ = kCRC32c;
|
|
|
|
} else {
|
|
|
|
version_ = DecodeFixed32(magic_ptr - 4);
|
|
|
|
if (version_ != kFooterVersion) {
|
|
|
|
return Status::Corruption("bad footer version");
|
|
|
|
}
|
|
|
|
// Footer version 1 will always occupy exactly this many bytes.
|
|
|
|
// It consists of the checksum type, two block handles, padding,
|
|
|
|
// a version number, and a magic number
|
|
|
|
if (input->size() < kVersion1EncodedLength) {
|
|
|
|
return Status::InvalidArgument("input is too short to be an sstable");
|
|
|
|
} else {
|
|
|
|
input->remove_prefix(input->size() - kVersion1EncodedLength);
|
|
|
|
}
|
|
|
|
uint32_t checksum;
|
|
|
|
if (!GetVarint32(input, &checksum)) {
|
|
|
|
return Status::Corruption("bad checksum type");
|
|
|
|
}
|
|
|
|
checksum_ = static_cast<ChecksumType>(checksum);
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
Status result = metaindex_handle_.DecodeFrom(input);
|
|
|
|
if (result.ok()) {
|
|
|
|
result = index_handle_.DecodeFrom(input);
|
|
|
|
}
|
|
|
|
if (result.ok()) {
|
|
|
|
// We skip over any leftover data (just padding for now) in "input"
|
2014-05-01 20:09:32 +02:00
|
|
|
const char* end = magic_ptr + kMagicNumberLengthByte;
|
2011-03-18 23:37:00 +01:00
|
|
|
*input = Slice(end, input->data() + input->size() - end);
|
|
|
|
}
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2013-12-05 01:35:48 +01:00
|
|
|
Status ReadFooterFromFile(RandomAccessFile* file,
|
|
|
|
uint64_t file_size,
|
|
|
|
Footer* footer) {
|
2014-05-01 20:09:32 +02:00
|
|
|
if (file_size < Footer::kMinEncodedLength) {
|
2013-12-05 01:35:48 +01:00
|
|
|
return Status::InvalidArgument("file is too short to be an sstable");
|
|
|
|
}
|
|
|
|
|
2014-05-01 20:09:32 +02:00
|
|
|
char footer_space[Footer::kMaxEncodedLength];
|
2013-12-05 01:35:48 +01:00
|
|
|
Slice footer_input;
|
2014-05-01 20:09:32 +02:00
|
|
|
size_t read_offset = (file_size > Footer::kMaxEncodedLength)
|
|
|
|
? (file_size - Footer::kMaxEncodedLength)
|
|
|
|
: 0;
|
|
|
|
Status s = file->Read(read_offset, Footer::kMaxEncodedLength, &footer_input,
|
2013-12-05 01:35:48 +01:00
|
|
|
footer_space);
|
|
|
|
if (!s.ok()) return s;
|
|
|
|
|
|
|
|
// Check that we actually read the whole footer from the file. It may be
|
|
|
|
// that size isn't correct.
|
2014-05-01 20:09:32 +02:00
|
|
|
if (footer_input.size() < Footer::kMinEncodedLength) {
|
2013-12-05 01:35:48 +01:00
|
|
|
return Status::InvalidArgument("file is too short to be an sstable");
|
|
|
|
}
|
|
|
|
|
|
|
|
return footer->DecodeFrom(&footer_input);
|
|
|
|
}
|
|
|
|
|
2014-07-31 08:11:59 +02:00
|
|
|
// Read a block and check its CRC
|
|
|
|
// contents is the result of reading.
|
|
|
|
// According to the implementation of file->Read, contents may not point to buf
|
|
|
|
Status ReadBlock(RandomAccessFile* file, const Footer& footer,
|
2014-08-04 20:25:42 +02:00
|
|
|
const ReadOptions& options, const BlockHandle& handle,
|
|
|
|
Slice* contents, /* result of reading */ char* buf) {
|
2011-04-21 00:48:11 +02:00
|
|
|
size_t n = static_cast<size_t>(handle.size());
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
|
2014-04-08 19:58:07 +02:00
|
|
|
PERF_TIMER_AUTO(block_read_time);
|
2014-07-31 08:11:59 +02:00
|
|
|
Status s = file->Read(handle.offset(), n + kBlockTrailerSize, contents, buf);
|
2014-04-08 19:58:07 +02:00
|
|
|
PERF_TIMER_MEASURE(block_read_time);
|
|
|
|
PERF_COUNTER_ADD(block_read_count, 1);
|
|
|
|
PERF_COUNTER_ADD(block_read_byte, n + kBlockTrailerSize);
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
2014-07-31 08:11:59 +02:00
|
|
|
if (contents->size() != n + kBlockTrailerSize) {
|
2011-03-18 23:37:00 +01:00
|
|
|
return Status::Corruption("truncated block read");
|
|
|
|
}
|
|
|
|
|
|
|
|
// Check the crc of the type and the block contents
|
2014-07-31 08:11:59 +02:00
|
|
|
const char* data = contents->data(); // Pointer to where Read put the data
|
2011-03-18 23:37:00 +01:00
|
|
|
if (options.verify_checksums) {
|
2014-05-01 20:09:32 +02:00
|
|
|
uint32_t value = DecodeFixed32(data + n + 1);
|
2014-05-01 20:12:35 +02:00
|
|
|
uint32_t actual = 0;
|
2014-05-01 20:09:32 +02:00
|
|
|
switch (footer.checksum()) {
|
|
|
|
case kCRC32c:
|
|
|
|
value = crc32c::Unmask(value);
|
|
|
|
actual = crc32c::Value(data, n + 1);
|
|
|
|
break;
|
|
|
|
case kxxHash:
|
|
|
|
actual = XXH32(data, n + 1, 0);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
s = Status::Corruption("unknown checksum type");
|
|
|
|
}
|
|
|
|
if (s.ok() && actual != value) {
|
2011-03-18 23:37:00 +01:00
|
|
|
s = Status::Corruption("block checksum mismatch");
|
2014-05-01 20:09:32 +02:00
|
|
|
}
|
|
|
|
if (!s.ok()) {
|
2011-03-18 23:37:00 +01:00
|
|
|
return s;
|
|
|
|
}
|
2014-07-31 08:11:59 +02:00
|
|
|
PERF_TIMER_STOP(block_checksum_time);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
2014-07-31 08:11:59 +02:00
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Decompress a block according to params
|
|
|
|
// May need to malloc a space for cache usage
|
|
|
|
Status DecompressBlock(BlockContents* result, size_t block_size,
|
2014-08-04 20:25:42 +02:00
|
|
|
bool do_uncompress, const char* buf,
|
|
|
|
const Slice& contents, bool use_stack_buf) {
|
2014-07-31 08:11:59 +02:00
|
|
|
Status s;
|
|
|
|
size_t n = block_size;
|
|
|
|
const char* data = contents.data();
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2014-07-31 08:11:59 +02:00
|
|
|
result->data = Slice();
|
|
|
|
result->cachable = false;
|
|
|
|
result->heap_allocated = false;
|
|
|
|
|
|
|
|
PERF_TIMER_AUTO(block_decompress_time);
|
2014-05-01 20:09:32 +02:00
|
|
|
rocksdb::CompressionType compression_type =
|
|
|
|
static_cast<rocksdb::CompressionType>(data[n]);
|
2013-09-02 08:23:40 +02:00
|
|
|
// If the caller has requested that the block not be uncompressed
|
2014-05-01 20:09:32 +02:00
|
|
|
if (!do_uncompress || compression_type == kNoCompression) {
|
2013-09-02 08:23:40 +02:00
|
|
|
if (data != buf) {
|
|
|
|
// File implementation gave us pointer to some other data.
|
|
|
|
// Use it directly under the assumption that it will be live
|
|
|
|
// while the file is open.
|
|
|
|
result->data = Slice(data, n);
|
|
|
|
result->heap_allocated = false;
|
|
|
|
result->cachable = false; // Do not double-cache
|
|
|
|
} else {
|
2014-07-31 08:11:59 +02:00
|
|
|
if (use_stack_buf) {
|
|
|
|
// Need to allocate space in heap for cache usage
|
|
|
|
char* new_buf = new char[n];
|
|
|
|
memcpy(new_buf, buf, n);
|
|
|
|
result->data = Slice(new_buf, n);
|
|
|
|
} else {
|
|
|
|
result->data = Slice(buf, n);
|
|
|
|
}
|
|
|
|
|
2013-09-02 08:23:40 +02:00
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
|
|
|
}
|
2014-05-01 20:09:32 +02:00
|
|
|
result->compression_type = compression_type;
|
|
|
|
s = Status::OK();
|
2013-09-02 08:23:40 +02:00
|
|
|
} else {
|
2013-11-06 23:16:22 +01:00
|
|
|
s = UncompressBlockContents(data, n, result);
|
2013-09-02 08:23:40 +02:00
|
|
|
}
|
2014-04-08 19:58:07 +02:00
|
|
|
PERF_TIMER_STOP(block_decompress_time);
|
2013-09-02 08:23:40 +02:00
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2014-07-31 08:11:59 +02:00
|
|
|
// Read and Decompress block
|
|
|
|
// Use buf in stack as temp reading buffer
|
|
|
|
Status ReadAndDecompressFast(RandomAccessFile* file, const Footer& footer,
|
|
|
|
const ReadOptions& options,
|
|
|
|
const BlockHandle& handle, BlockContents* result,
|
|
|
|
Env* env, bool do_uncompress) {
|
|
|
|
Status s;
|
|
|
|
Slice contents;
|
|
|
|
size_t n = static_cast<size_t>(handle.size());
|
|
|
|
char buf[DefaultStackBufferSize];
|
|
|
|
|
|
|
|
s = ReadBlock(file, footer, options, handle, &contents, buf);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
s = DecompressBlock(result, n, do_uncompress, buf, contents, true);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Read and Decompress block
|
|
|
|
// Use buf in heap as temp reading buffer
|
|
|
|
Status ReadAndDecompress(RandomAccessFile* file, const Footer& footer,
|
|
|
|
const ReadOptions& options, const BlockHandle& handle,
|
|
|
|
BlockContents* result, Env* env, bool do_uncompress) {
|
|
|
|
Status s;
|
|
|
|
Slice contents;
|
|
|
|
size_t n = static_cast<size_t>(handle.size());
|
|
|
|
char* buf = new char[n + kBlockTrailerSize];
|
|
|
|
|
|
|
|
s = ReadBlock(file, footer, options, handle, &contents, buf);
|
|
|
|
if (!s.ok()) {
|
|
|
|
delete[] buf;
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
s = DecompressBlock(result, n, do_uncompress, buf, contents, false);
|
|
|
|
if (!s.ok()) {
|
|
|
|
delete[] buf;
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (result->data.data() != buf) {
|
|
|
|
delete[] buf;
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status ReadBlockContents(RandomAccessFile* file, const Footer& footer,
|
|
|
|
const ReadOptions& options, const BlockHandle& handle,
|
|
|
|
BlockContents* result, Env* env, bool do_uncompress) {
|
|
|
|
size_t n = static_cast<size_t>(handle.size());
|
|
|
|
if (do_uncompress && n + kBlockTrailerSize < DefaultStackBufferSize) {
|
|
|
|
return ReadAndDecompressFast(file, footer, options, handle, result, env,
|
|
|
|
do_uncompress);
|
|
|
|
} else {
|
|
|
|
return ReadAndDecompress(file, footer, options, handle, result, env,
|
|
|
|
do_uncompress);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-09-02 08:23:40 +02:00
|
|
|
//
|
|
|
|
// The 'data' points to the raw block contents that was read in from file.
|
|
|
|
// This method allocates a new heap buffer and the raw block
|
|
|
|
// contents are uncompresed into this buffer. This
|
|
|
|
// buffer is returned via 'result' and it is upto the caller to
|
|
|
|
// free this buffer.
|
|
|
|
Status UncompressBlockContents(const char* data, size_t n,
|
|
|
|
BlockContents* result) {
|
2013-03-01 03:04:58 +01:00
|
|
|
char* ubuf = nullptr;
|
2012-06-29 04:26:43 +02:00
|
|
|
int decompress_size = 0;
|
2013-09-02 08:23:40 +02:00
|
|
|
assert(data[n] != kNoCompression);
|
2011-03-18 23:37:00 +01:00
|
|
|
switch (data[n]) {
|
2011-03-23 00:24:02 +01:00
|
|
|
case kSnappyCompression: {
|
2011-07-21 04:40:18 +02:00
|
|
|
size_t ulength = 0;
|
2012-12-20 23:25:06 +01:00
|
|
|
static char snappy_corrupt_msg[] =
|
|
|
|
"Snappy not supported or corrupted Snappy compressed block contents";
|
2011-07-21 04:40:18 +02:00
|
|
|
if (!port::Snappy_GetUncompressedLength(data, n, &ulength)) {
|
2012-12-20 23:25:06 +01:00
|
|
|
return Status::Corruption(snappy_corrupt_msg);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
2012-06-28 08:41:33 +02:00
|
|
|
ubuf = new char[ulength];
|
2011-07-21 04:40:18 +02:00
|
|
|
if (!port::Snappy_Uncompress(data, n, ubuf)) {
|
|
|
|
delete[] ubuf;
|
2012-12-20 23:25:06 +01:00
|
|
|
return Status::Corruption(snappy_corrupt_msg);
|
2011-07-21 04:40:18 +02:00
|
|
|
}
|
2012-04-17 17:36:46 +02:00
|
|
|
result->data = Slice(ubuf, ulength);
|
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
2011-03-18 23:37:00 +01:00
|
|
|
break;
|
|
|
|
}
|
2012-06-28 08:41:33 +02:00
|
|
|
case kZlibCompression:
|
|
|
|
ubuf = port::Zlib_Uncompress(data, n, &decompress_size);
|
2012-12-20 23:25:06 +01:00
|
|
|
static char zlib_corrupt_msg[] =
|
|
|
|
"Zlib not supported or corrupted Zlib compressed block contents";
|
2012-06-28 08:41:33 +02:00
|
|
|
if (!ubuf) {
|
2012-12-20 23:25:06 +01:00
|
|
|
return Status::Corruption(zlib_corrupt_msg);
|
2012-06-28 08:41:33 +02:00
|
|
|
}
|
|
|
|
result->data = Slice(ubuf, decompress_size);
|
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
|
|
|
break;
|
2012-06-29 04:26:43 +02:00
|
|
|
case kBZip2Compression:
|
|
|
|
ubuf = port::BZip2_Uncompress(data, n, &decompress_size);
|
2012-12-20 23:25:06 +01:00
|
|
|
static char bzip2_corrupt_msg[] =
|
|
|
|
"Bzip2 not supported or corrupted Bzip2 compressed block contents";
|
2012-06-29 04:26:43 +02:00
|
|
|
if (!ubuf) {
|
2012-12-20 23:25:06 +01:00
|
|
|
return Status::Corruption(bzip2_corrupt_msg);
|
2012-06-29 04:26:43 +02:00
|
|
|
}
|
|
|
|
result->data = Slice(ubuf, decompress_size);
|
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
|
|
|
break;
|
2014-02-08 03:12:30 +01:00
|
|
|
case kLZ4Compression:
|
|
|
|
ubuf = port::LZ4_Uncompress(data, n, &decompress_size);
|
|
|
|
static char lz4_corrupt_msg[] =
|
|
|
|
"LZ4 not supported or corrupted LZ4 compressed block contents";
|
|
|
|
if (!ubuf) {
|
|
|
|
return Status::Corruption(lz4_corrupt_msg);
|
|
|
|
}
|
|
|
|
result->data = Slice(ubuf, decompress_size);
|
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
|
|
|
break;
|
|
|
|
case kLZ4HCCompression:
|
|
|
|
ubuf = port::LZ4_Uncompress(data, n, &decompress_size);
|
|
|
|
static char lz4hc_corrupt_msg[] =
|
|
|
|
"LZ4HC not supported or corrupted LZ4HC compressed block contents";
|
|
|
|
if (!ubuf) {
|
|
|
|
return Status::Corruption(lz4hc_corrupt_msg);
|
|
|
|
}
|
|
|
|
result->data = Slice(ubuf, decompress_size);
|
|
|
|
result->heap_allocated = true;
|
|
|
|
result->cachable = true;
|
|
|
|
break;
|
2011-03-18 23:37:00 +01:00
|
|
|
default:
|
|
|
|
return Status::Corruption("bad block type");
|
|
|
|
}
|
2014-02-05 01:21:47 +01:00
|
|
|
result->compression_type = kNoCompression; // not compressed any more
|
2011-03-18 23:37:00 +01:00
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
} // namespace rocksdb
|