You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/db/log_reader.cc

859 lines
30 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/log_reader.h"
#include <stdio.h>
#include "file/sequence_file_reader.h"
#include "port/lang.h"
#include "rocksdb/env.h"
#include "test_util/sync_point.h"
#include "util/coding.h"
#include "util/crc32c.h"
namespace ROCKSDB_NAMESPACE {
namespace log {
Reader::Reporter::~Reporter() {
}
Reader::Reader(std::shared_ptr<Logger> info_log,
std::unique_ptr<SequentialFileReader>&& _file,
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
Reporter* reporter, bool checksum, uint64_t log_num)
: info_log_(info_log),
file_(std::move(_file)),
reporter_(reporter),
checksum_(checksum),
backing_store_(new char[kBlockSize]),
buffer_(),
eof_(false),
read_error_(false),
eof_offset_(0),
last_record_offset_(0),
end_of_buffer_offset_(0),
log_number_(log_num),
recycled_(false),
first_record_read_(false),
compression_type_(kNoCompression),
compression_type_record_read_(false),
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
uncompress_(nullptr),
hash_state_(nullptr),
uncompress_hash_state_(nullptr){};
Reader::~Reader() {
delete[] backing_store_;
if (uncompress_) {
delete uncompress_;
}
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
if (hash_state_) {
XXH3_freeState(hash_state_);
}
if (uncompress_hash_state_) {
XXH3_freeState(uncompress_hash_state_);
}
}
// For kAbsoluteConsistency, on clean shutdown we don't expect any error
// in the log files. For other modes, we can ignore only incomplete records
// in the last log file, which are presumably due to a write in progress
// during restart (or from log recycling).
//
// TODO krad: Evaluate if we need to move to a more strict mode where we
// restrict the inconsistency to only the last log
bool Reader::ReadRecord(Slice* record, std::string* scratch,
WALRecoveryMode wal_recovery_mode,
uint64_t* record_checksum) {
scratch->clear();
record->clear();
if (record_checksum != nullptr) {
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
if (hash_state_ == nullptr) {
hash_state_ = XXH3_createState();
}
XXH3_64bits_reset(hash_state_);
}
if (uncompress_) {
uncompress_->Reset();
}
bool in_fragmented_record = false;
// Record offset of the logical record that we're reading
// 0 is a dummy value to make compilers happy
uint64_t prospective_record_offset = 0;
Slice fragment;
while (true) {
uint64_t physical_record_offset = end_of_buffer_offset_ - buffer_.size();
size_t drop_size = 0;
const unsigned int record_type =
ReadPhysicalRecord(&fragment, &drop_size, record_checksum);
switch (record_type) {
case kFullType:
case kRecyclableFullType:
rocksdb: Fixed 'Dead assignment' and 'Dead initialization' scan-build warnings Summary: This diff contains trivial fixes for 6 scan-build warnings: **db/c_test.c** `db` variable is never read. Removed assignment. scan-build report: http://home.fburl.com/~sugak/latest20/report-9b77d2.html#EndPath **db/db_iter.cc** `skipping` local variable is assigned to false. Then in the next switch block the only "non return" case assign `skipping` to true, the rest cases don't use it and all do return. scan-build report: http://home.fburl.com/~sugak/latest20/report-13fca7.html#EndPath **db/log_reader.cc** In `bool Reader::SkipToInitialBlock()` `offset_in_block` local variable is assigned to 0 `if (offset_in_block > kBlockSize - 6)` and then never used. Removed the assignment and renamed it to `initial_offset_in_block` to avoid confusion. scan-build report: http://home.fburl.com/~sugak/latest20/report-a618dd.html#EndPath In `bool Reader::ReadRecord(Slice* record, std::string* scratch)` local variable `in_fragmented_record` in switch case `kFullType` block is assigned to false and then does `return` without use. In the other switch case `kFirstType` block the same `in_fragmented_record` is assigned to false, but later assigned to true without prior use. Removed assignment for both cases. scan-build reprots: http://home.fburl.com/~sugak/latest20/report-bb86b0.html#EndPath http://home.fburl.com/~sugak/latest20/report-a975be.html#EndPath **table/plain_table_key_coding.cc** Local variable `user_key_size` is assigned when declared. But then in both places where it is used assigned to `static_cast<uint32_t>(key.size() - 8)`. Changed to initialize the variable to the proper value in declaration. scan-build report: http://home.fburl.com/~sugak/latest20/report-9e6b86.html#EndPath **tools/db_stress.cc** Missing `break` in switch case block. This seems to be a bug. Added missing `break`. Test Plan: Make sure all tests are passing and scan-build does not report 'Dead assignment' and 'Dead initialization' bugs. ```lang=bash % make check % make analyze ``` Reviewers: meyering, igor, kradhakrishnan, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33795
10 years ago
if (in_fragmented_record && !scratch->empty()) {
// Handle bug in earlier versions of log::Writer where
// it could emit an empty kFirstType record at the tail end
// of a block followed by a kFullType or kFirstType record
// at the beginning of the next block.
rocksdb: Fixed 'Dead assignment' and 'Dead initialization' scan-build warnings Summary: This diff contains trivial fixes for 6 scan-build warnings: **db/c_test.c** `db` variable is never read. Removed assignment. scan-build report: http://home.fburl.com/~sugak/latest20/report-9b77d2.html#EndPath **db/db_iter.cc** `skipping` local variable is assigned to false. Then in the next switch block the only "non return" case assign `skipping` to true, the rest cases don't use it and all do return. scan-build report: http://home.fburl.com/~sugak/latest20/report-13fca7.html#EndPath **db/log_reader.cc** In `bool Reader::SkipToInitialBlock()` `offset_in_block` local variable is assigned to 0 `if (offset_in_block > kBlockSize - 6)` and then never used. Removed the assignment and renamed it to `initial_offset_in_block` to avoid confusion. scan-build report: http://home.fburl.com/~sugak/latest20/report-a618dd.html#EndPath In `bool Reader::ReadRecord(Slice* record, std::string* scratch)` local variable `in_fragmented_record` in switch case `kFullType` block is assigned to false and then does `return` without use. In the other switch case `kFirstType` block the same `in_fragmented_record` is assigned to false, but later assigned to true without prior use. Removed assignment for both cases. scan-build reprots: http://home.fburl.com/~sugak/latest20/report-bb86b0.html#EndPath http://home.fburl.com/~sugak/latest20/report-a975be.html#EndPath **table/plain_table_key_coding.cc** Local variable `user_key_size` is assigned when declared. But then in both places where it is used assigned to `static_cast<uint32_t>(key.size() - 8)`. Changed to initialize the variable to the proper value in declaration. scan-build report: http://home.fburl.com/~sugak/latest20/report-9e6b86.html#EndPath **tools/db_stress.cc** Missing `break` in switch case block. This seems to be a bug. Added missing `break`. Test Plan: Make sure all tests are passing and scan-build does not report 'Dead assignment' and 'Dead initialization' bugs. ```lang=bash % make check % make analyze ``` Reviewers: meyering, igor, kradhakrishnan, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33795
10 years ago
ReportCorruption(scratch->size(), "partial record without end(1)");
}
// No need to compute record_checksum since the record
// consists of a single fragment and the checksum is computed
// in ReadPhysicalRecord() if WAL compression is enabled
if (record_checksum != nullptr && uncompress_ == nullptr) {
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
// No need to stream since the record is a single fragment
*record_checksum = XXH3_64bits(fragment.data(), fragment.size());
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
}
prospective_record_offset = physical_record_offset;
scratch->clear();
*record = fragment;
last_record_offset_ = prospective_record_offset;
first_record_read_ = true;
return true;
case kFirstType:
case kRecyclableFirstType:
rocksdb: Fixed 'Dead assignment' and 'Dead initialization' scan-build warnings Summary: This diff contains trivial fixes for 6 scan-build warnings: **db/c_test.c** `db` variable is never read. Removed assignment. scan-build report: http://home.fburl.com/~sugak/latest20/report-9b77d2.html#EndPath **db/db_iter.cc** `skipping` local variable is assigned to false. Then in the next switch block the only "non return" case assign `skipping` to true, the rest cases don't use it and all do return. scan-build report: http://home.fburl.com/~sugak/latest20/report-13fca7.html#EndPath **db/log_reader.cc** In `bool Reader::SkipToInitialBlock()` `offset_in_block` local variable is assigned to 0 `if (offset_in_block > kBlockSize - 6)` and then never used. Removed the assignment and renamed it to `initial_offset_in_block` to avoid confusion. scan-build report: http://home.fburl.com/~sugak/latest20/report-a618dd.html#EndPath In `bool Reader::ReadRecord(Slice* record, std::string* scratch)` local variable `in_fragmented_record` in switch case `kFullType` block is assigned to false and then does `return` without use. In the other switch case `kFirstType` block the same `in_fragmented_record` is assigned to false, but later assigned to true without prior use. Removed assignment for both cases. scan-build reprots: http://home.fburl.com/~sugak/latest20/report-bb86b0.html#EndPath http://home.fburl.com/~sugak/latest20/report-a975be.html#EndPath **table/plain_table_key_coding.cc** Local variable `user_key_size` is assigned when declared. But then in both places where it is used assigned to `static_cast<uint32_t>(key.size() - 8)`. Changed to initialize the variable to the proper value in declaration. scan-build report: http://home.fburl.com/~sugak/latest20/report-9e6b86.html#EndPath **tools/db_stress.cc** Missing `break` in switch case block. This seems to be a bug. Added missing `break`. Test Plan: Make sure all tests are passing and scan-build does not report 'Dead assignment' and 'Dead initialization' bugs. ```lang=bash % make check % make analyze ``` Reviewers: meyering, igor, kradhakrishnan, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33795
10 years ago
if (in_fragmented_record && !scratch->empty()) {
// Handle bug in earlier versions of log::Writer where
// it could emit an empty kFirstType record at the tail end
// of a block followed by a kFullType or kFirstType record
// at the beginning of the next block.
rocksdb: Fixed 'Dead assignment' and 'Dead initialization' scan-build warnings Summary: This diff contains trivial fixes for 6 scan-build warnings: **db/c_test.c** `db` variable is never read. Removed assignment. scan-build report: http://home.fburl.com/~sugak/latest20/report-9b77d2.html#EndPath **db/db_iter.cc** `skipping` local variable is assigned to false. Then in the next switch block the only "non return" case assign `skipping` to true, the rest cases don't use it and all do return. scan-build report: http://home.fburl.com/~sugak/latest20/report-13fca7.html#EndPath **db/log_reader.cc** In `bool Reader::SkipToInitialBlock()` `offset_in_block` local variable is assigned to 0 `if (offset_in_block > kBlockSize - 6)` and then never used. Removed the assignment and renamed it to `initial_offset_in_block` to avoid confusion. scan-build report: http://home.fburl.com/~sugak/latest20/report-a618dd.html#EndPath In `bool Reader::ReadRecord(Slice* record, std::string* scratch)` local variable `in_fragmented_record` in switch case `kFullType` block is assigned to false and then does `return` without use. In the other switch case `kFirstType` block the same `in_fragmented_record` is assigned to false, but later assigned to true without prior use. Removed assignment for both cases. scan-build reprots: http://home.fburl.com/~sugak/latest20/report-bb86b0.html#EndPath http://home.fburl.com/~sugak/latest20/report-a975be.html#EndPath **table/plain_table_key_coding.cc** Local variable `user_key_size` is assigned when declared. But then in both places where it is used assigned to `static_cast<uint32_t>(key.size() - 8)`. Changed to initialize the variable to the proper value in declaration. scan-build report: http://home.fburl.com/~sugak/latest20/report-9e6b86.html#EndPath **tools/db_stress.cc** Missing `break` in switch case block. This seems to be a bug. Added missing `break`. Test Plan: Make sure all tests are passing and scan-build does not report 'Dead assignment' and 'Dead initialization' bugs. ```lang=bash % make check % make analyze ``` Reviewers: meyering, igor, kradhakrishnan, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33795
10 years ago
ReportCorruption(scratch->size(), "partial record without end(2)");
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
XXH3_64bits_reset(hash_state_);
}
if (record_checksum != nullptr) {
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
XXH3_64bits_update(hash_state_, fragment.data(), fragment.size());
}
prospective_record_offset = physical_record_offset;
scratch->assign(fragment.data(), fragment.size());
in_fragmented_record = true;
break;
case kMiddleType:
case kRecyclableMiddleType:
if (!in_fragmented_record) {
ReportCorruption(fragment.size(),
"missing start of fragmented record(1)");
} else {
if (record_checksum != nullptr) {
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
XXH3_64bits_update(hash_state_, fragment.data(), fragment.size());
}
scratch->append(fragment.data(), fragment.size());
}
break;
case kLastType:
case kRecyclableLastType:
if (!in_fragmented_record) {
ReportCorruption(fragment.size(),
"missing start of fragmented record(2)");
} else {
if (record_checksum != nullptr) {
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
XXH3_64bits_update(hash_state_, fragment.data(), fragment.size());
*record_checksum = XXH3_64bits_digest(hash_state_);
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
}
scratch->append(fragment.data(), fragment.size());
*record = Slice(*scratch);
last_record_offset_ = prospective_record_offset;
first_record_read_ = true;
return true;
}
break;
case kBadHeader:
if (wal_recovery_mode == WALRecoveryMode::kAbsoluteConsistency ||
wal_recovery_mode == WALRecoveryMode::kPointInTimeRecovery) {
// In clean shutdown we don't expect any error in the log files.
// In point-in-time recovery an incomplete record at the end could
// produce a hole in the recovered data. Report an error here, which
// higher layers can choose to ignore when it's provable there is no
// hole.
ReportCorruption(drop_size, "truncated header");
}
FALLTHROUGH_INTENDED;
case kEof:
if (in_fragmented_record) {
if (wal_recovery_mode == WALRecoveryMode::kAbsoluteConsistency ||
wal_recovery_mode == WALRecoveryMode::kPointInTimeRecovery) {
// In clean shutdown we don't expect any error in the log files.
// In point-in-time recovery an incomplete record at the end could
// produce a hole in the recovered data. Report an error here, which
// higher layers can choose to ignore when it's provable there is no
// hole.
ReportCorruption(scratch->size(), "error reading trailing data");
}
// This can be caused by the writer dying immediately after
// writing a physical record but before completing the next; don't
// treat it as a corruption, just ignore the entire logical record.
scratch->clear();
}
return false;
case kOldRecord:
if (wal_recovery_mode != WALRecoveryMode::kSkipAnyCorruptedRecords) {
// Treat a record from a previous instance of the log as EOF.
if (in_fragmented_record) {
if (wal_recovery_mode == WALRecoveryMode::kAbsoluteConsistency ||
wal_recovery_mode == WALRecoveryMode::kPointInTimeRecovery) {
// In clean shutdown we don't expect any error in the log files.
// In point-in-time recovery an incomplete record at the end could
// produce a hole in the recovered data. Report an error here,
// which higher layers can choose to ignore when it's provable
// there is no hole.
ReportCorruption(scratch->size(), "error reading trailing data");
}
// This can be caused by the writer dying immediately after
// writing a physical record but before completing the next; don't
// treat it as a corruption, just ignore the entire logical record.
scratch->clear();
}
return false;
}
FALLTHROUGH_INTENDED;
case kBadRecord:
if (in_fragmented_record) {
ReportCorruption(scratch->size(), "error in middle of record");
in_fragmented_record = false;
scratch->clear();
}
break;
case kBadRecordLen:
if (eof_) {
if (wal_recovery_mode == WALRecoveryMode::kAbsoluteConsistency ||
wal_recovery_mode == WALRecoveryMode::kPointInTimeRecovery) {
// In clean shutdown we don't expect any error in the log files.
// In point-in-time recovery an incomplete record at the end could
// produce a hole in the recovered data. Report an error here, which
// higher layers can choose to ignore when it's provable there is no
// hole.
ReportCorruption(drop_size, "truncated record body");
}
return false;
}
FALLTHROUGH_INTENDED;
case kBadRecordChecksum:
if (recycled_ &&
wal_recovery_mode ==
WALRecoveryMode::kTolerateCorruptedTailRecords) {
scratch->clear();
return false;
}
if (record_type == kBadRecordLen) {
ReportCorruption(drop_size, "bad record length");
} else {
ReportCorruption(drop_size, "checksum mismatch");
}
if (in_fragmented_record) {
ReportCorruption(scratch->size(), "error in middle of record");
in_fragmented_record = false;
scratch->clear();
}
break;
case kSetCompressionType: {
if (compression_type_record_read_) {
ReportCorruption(fragment.size(),
"read multiple SetCompressionType records");
}
if (first_record_read_) {
ReportCorruption(fragment.size(),
"SetCompressionType not the first record");
}
prospective_record_offset = physical_record_offset;
scratch->clear();
last_record_offset_ = prospective_record_offset;
CompressionTypeRecord compression_record(kNoCompression);
Status s = compression_record.DecodeFrom(&fragment);
if (!s.ok()) {
ReportCorruption(fragment.size(),
"could not decode SetCompressionType record");
} else {
InitCompression(compression_record);
}
break;
}
default: {
char buf[40];
snprintf(buf, sizeof(buf), "unknown record type %u", record_type);
ReportCorruption(
(fragment.size() + (in_fragmented_record ? scratch->size() : 0)),
buf);
in_fragmented_record = false;
scratch->clear();
break;
}
}
}
return false;
}
uint64_t Reader::LastRecordOffset() {
return last_record_offset_;
}
uint64_t Reader::LastRecordEnd() {
return end_of_buffer_offset_ - buffer_.size();
}
void Reader::UnmarkEOF() {
if (read_error_) {
return;
}
eof_ = false;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
if (eof_offset_ == 0) {
return;
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
UnmarkEOFInternal();
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
void Reader::UnmarkEOFInternal() {
// If the EOF was in the middle of a block (a partial block was read) we have
// to read the rest of the block as ReadPhysicalRecord can only read full
// blocks and expects the file position indicator to be aligned to the start
// of a block.
//
// consumed_bytes + buffer_size() + remaining == kBlockSize
size_t consumed_bytes = eof_offset_ - buffer_.size();
size_t remaining = kBlockSize - eof_offset_;
// backing_store_ is used to concatenate what is left in buffer_ and
// the remainder of the block. If buffer_ already uses backing_store_,
// we just append the new data.
if (buffer_.data() != backing_store_ + consumed_bytes) {
// Buffer_ does not use backing_store_ for storage.
// Copy what is left in buffer_ to backing_store.
memmove(backing_store_ + consumed_bytes, buffer_.data(), buffer_.size());
}
Slice read_buffer;
Support read rate-limiting in SequentialFileReader (#9973) Summary: Added rate limiter and read rate-limiting support to SequentialFileReader. I've updated call sites to SequentialFileReader::Read with appropriate IO priority (or left a TODO and specified IO_TOTAL for now). The PR is separated into four commits: the first one added the rate-limiting support, but with some fixes in the unit test since the number of request bytes from rate limiter in SequentialFileReader are not accurate (there is overcharge at EOF). The second commit fixed this by allowing SequentialFileReader to check file size and determine how many bytes are left in the file to read. The third commit added benchmark related code. The fourth commit moved the logic of using file size to avoid overcharging the rate limiter into backup engine (the main user of SequentialFileReader). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9973 Test Plan: - `make check`, backup_engine_test covers usage of SequentialFileReader with rate limiter. - Run db_bench to check if rate limiting is throttling as expected: Verified that reads and writes are together throttled at 2MB/s, and at 0.2MB chunks that are 100ms apart. - Set up: `./db_bench --benchmarks=fillrandom -db=/dev/shm/test_rocksdb` - Benchmark: ``` strace -ttfe read,write ./db_bench --benchmarks=backup -db=/dev/shm/test_rocksdb --backup_rate_limit=2097152 --use_existing_db strace -ttfe read,write ./db_bench --benchmarks=restore -db=/dev/shm/test_rocksdb --restore_rate_limit=2097152 --use_existing_db ``` - db bench on backup and restore to ensure no performance regression. - backup (avg over 50 runs): pre-change: 1.90443e+06 micros/op; post-change: 1.8993e+06 micros/op (improve by 0.2%) - restore (avg over 50 runs): pre-change: 1.79105e+06 micros/op; post-change: 1.78192e+06 micros/op (improve by 0.5%) ``` # Set up ./db_bench --benchmarks=fillrandom -db=/tmp/test_rocksdb -num=10000000 # benchmark TEST_TMPDIR=/tmp/test_rocksdb NUM_RUN=50 for ((j=0;j<$NUM_RUN;j++)) do ./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=backup -use_existing_db | egrep 'backup' # Restore #./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=restore -use_existing_db done > rate_limit.txt && awk -v NUM_RUN=$NUM_RUN '{sum+=$3;sum_sqrt+=$3^2}END{print sum/NUM_RUN, sqrt(sum_sqrt/NUM_RUN-(sum/NUM_RUN)^2)}' rate_limit.txt >> rate_limit_2.txt ``` Reviewed By: hx235 Differential Revision: D36327418 Pulled By: cbi42 fbshipit-source-id: e75d4307cff815945482df5ba630c1e88d064691
2 years ago
// TODO: rate limit log reader with approriate priority.
// TODO: avoid overcharging rate limiter:
// Note that the Read here might overcharge SequentialFileReader's internal
// rate limiter if priority is not IO_TOTAL, e.g., when there is not enough
// content left until EOF to read.
Status status =
file_->Read(remaining, &read_buffer, backing_store_ + eof_offset_,
Env::IO_TOTAL /* rate_limiter_priority */);
size_t added = read_buffer.size();
end_of_buffer_offset_ += added;
if (!status.ok()) {
if (added > 0) {
ReportDrop(added, status);
}
read_error_ = true;
return;
}
if (read_buffer.data() != backing_store_ + eof_offset_) {
// Read did not write to backing_store_
memmove(backing_store_ + eof_offset_, read_buffer.data(),
read_buffer.size());
}
buffer_ = Slice(backing_store_ + consumed_bytes,
eof_offset_ + added - consumed_bytes);
if (added < remaining) {
eof_ = true;
eof_offset_ += added;
} else {
eof_offset_ = 0;
}
}
void Reader::ReportCorruption(size_t bytes, const char* reason) {
ReportDrop(bytes, Status::Corruption(reason));
}
void Reader::ReportDrop(size_t bytes, const Status& reason) {
if (reporter_ != nullptr) {
reporter_->Corruption(bytes, reason);
}
}
bool Reader::ReadMore(size_t* drop_size, int *error) {
if (!eof_ && !read_error_) {
// Last read was a full read, so this is a trailer to skip
buffer_.clear();
Support read rate-limiting in SequentialFileReader (#9973) Summary: Added rate limiter and read rate-limiting support to SequentialFileReader. I've updated call sites to SequentialFileReader::Read with appropriate IO priority (or left a TODO and specified IO_TOTAL for now). The PR is separated into four commits: the first one added the rate-limiting support, but with some fixes in the unit test since the number of request bytes from rate limiter in SequentialFileReader are not accurate (there is overcharge at EOF). The second commit fixed this by allowing SequentialFileReader to check file size and determine how many bytes are left in the file to read. The third commit added benchmark related code. The fourth commit moved the logic of using file size to avoid overcharging the rate limiter into backup engine (the main user of SequentialFileReader). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9973 Test Plan: - `make check`, backup_engine_test covers usage of SequentialFileReader with rate limiter. - Run db_bench to check if rate limiting is throttling as expected: Verified that reads and writes are together throttled at 2MB/s, and at 0.2MB chunks that are 100ms apart. - Set up: `./db_bench --benchmarks=fillrandom -db=/dev/shm/test_rocksdb` - Benchmark: ``` strace -ttfe read,write ./db_bench --benchmarks=backup -db=/dev/shm/test_rocksdb --backup_rate_limit=2097152 --use_existing_db strace -ttfe read,write ./db_bench --benchmarks=restore -db=/dev/shm/test_rocksdb --restore_rate_limit=2097152 --use_existing_db ``` - db bench on backup and restore to ensure no performance regression. - backup (avg over 50 runs): pre-change: 1.90443e+06 micros/op; post-change: 1.8993e+06 micros/op (improve by 0.2%) - restore (avg over 50 runs): pre-change: 1.79105e+06 micros/op; post-change: 1.78192e+06 micros/op (improve by 0.5%) ``` # Set up ./db_bench --benchmarks=fillrandom -db=/tmp/test_rocksdb -num=10000000 # benchmark TEST_TMPDIR=/tmp/test_rocksdb NUM_RUN=50 for ((j=0;j<$NUM_RUN;j++)) do ./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=backup -use_existing_db | egrep 'backup' # Restore #./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=restore -use_existing_db done > rate_limit.txt && awk -v NUM_RUN=$NUM_RUN '{sum+=$3;sum_sqrt+=$3^2}END{print sum/NUM_RUN, sqrt(sum_sqrt/NUM_RUN-(sum/NUM_RUN)^2)}' rate_limit.txt >> rate_limit_2.txt ``` Reviewed By: hx235 Differential Revision: D36327418 Pulled By: cbi42 fbshipit-source-id: e75d4307cff815945482df5ba630c1e88d064691
2 years ago
// TODO: rate limit log reader with approriate priority.
// TODO: avoid overcharging rate limiter:
// Note that the Read here might overcharge SequentialFileReader's internal
// rate limiter if priority is not IO_TOTAL, e.g., when there is not enough
// content left until EOF to read.
Status status = file_->Read(kBlockSize, &buffer_, backing_store_,
Env::IO_TOTAL /* rate_limiter_priority */);
TEST_SYNC_POINT_CALLBACK("LogReader::ReadMore:AfterReadFile", &status);
end_of_buffer_offset_ += buffer_.size();
if (!status.ok()) {
buffer_.clear();
ReportDrop(kBlockSize, status);
read_error_ = true;
*error = kEof;
return false;
} else if (buffer_.size() < static_cast<size_t>(kBlockSize)) {
eof_ = true;
eof_offset_ = buffer_.size();
}
return true;
} else {
// Note that if buffer_ is non-empty, we have a truncated header at the
// end of the file, which can be caused by the writer crashing in the
// middle of writing the header. Unless explicitly requested we don't
// considering this an error, just report EOF.
if (buffer_.size()) {
*drop_size = buffer_.size();
buffer_.clear();
*error = kBadHeader;
return false;
}
buffer_.clear();
*error = kEof;
return false;
}
}
unsigned int Reader::ReadPhysicalRecord(Slice* result, size_t* drop_size,
uint64_t* fragment_checksum) {
while (true) {
// We need at least the minimum header size
if (buffer_.size() < static_cast<size_t>(kHeaderSize)) {
// the default value of r is meaningless because ReadMore will overwrite
// it if it returns false; in case it returns true, the return value will
// not be used anyway
int r = kEof;
if (!ReadMore(drop_size, &r)) {
return r;
}
continue;
}
// Parse the header
const char* header = buffer_.data();
const uint32_t a = static_cast<uint32_t>(header[4]) & 0xff;
const uint32_t b = static_cast<uint32_t>(header[5]) & 0xff;
const unsigned int type = header[6];
const uint32_t length = a | (b << 8);
int header_size = kHeaderSize;
if (type >= kRecyclableFullType && type <= kRecyclableLastType) {
if (end_of_buffer_offset_ - buffer_.size() == 0) {
recycled_ = true;
}
header_size = kRecyclableHeaderSize;
// We need enough for the larger header
if (buffer_.size() < static_cast<size_t>(kRecyclableHeaderSize)) {
int r = kEof;
if (!ReadMore(drop_size, &r)) {
return r;
}
continue;
}
const uint32_t log_num = DecodeFixed32(header + 7);
if (log_num != log_number_) {
return kOldRecord;
}
}
if (header_size + length > buffer_.size()) {
assert(buffer_.size() >= static_cast<size_t>(header_size));
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
*drop_size = buffer_.size();
buffer_.clear();
// If the end of the read has been reached without seeing
// `header_size + length` bytes of payload, report a corruption. The
// higher layers can decide how to handle it based on the recovery mode,
// whether this occurred at EOF, whether this is the final WAL, etc.
return kBadRecordLen;
}
if (type == kZeroType && length == 0) {
// Skip zero length record without reporting any drops since
// such records are produced by the mmap based writing code in
// env_posix.cc that preallocates file regions.
// NOTE: this should never happen in DB written by new RocksDB versions,
// since we turn off mmap writes to manifest and log files
buffer_.clear();
return kBadRecord;
}
// Check crc
if (checksum_) {
uint32_t expected_crc = crc32c::Unmask(DecodeFixed32(header));
uint32_t actual_crc = crc32c::Value(header + 6, length + header_size - 6);
if (actual_crc != expected_crc) {
// Drop the rest of the buffer since "length" itself may have
// been corrupted and if we trust it, we could find some
// fragment of a real log record that just happens to look
// like a valid log record.
*drop_size = buffer_.size();
buffer_.clear();
return kBadRecordChecksum;
}
}
buffer_.remove_prefix(header_size + length);
if (!uncompress_ || type == kSetCompressionType) {
*result = Slice(header + header_size, length);
return type;
} else {
// Uncompress compressed records
uncompressed_record_.clear();
if (fragment_checksum != nullptr) {
if (uncompress_hash_state_ == nullptr) {
uncompress_hash_state_ = XXH3_createState();
}
XXH3_64bits_reset(uncompress_hash_state_);
}
size_t uncompressed_size = 0;
int remaining = 0;
do {
remaining = uncompress_->Uncompress(header + header_size, length,
uncompressed_buffer_.get(),
&uncompressed_size);
if (remaining < 0) {
buffer_.clear();
return kBadRecord;
}
if (uncompressed_size > 0) {
if (fragment_checksum != nullptr) {
XXH3_64bits_update(uncompress_hash_state_,
uncompressed_buffer_.get(), uncompressed_size);
}
uncompressed_record_.append(uncompressed_buffer_.get(),
uncompressed_size);
}
} while (remaining > 0 || uncompressed_size == kBlockSize);
if (fragment_checksum != nullptr) {
// We can remove this check by updating hash_state_ directly,
// but that requires resetting hash_state_ for full and first types
// for edge cases like consecutive fist type records.
// Leaving the check as is since it is cleaner and can revert to the
// above approach if it causes performance impact.
*fragment_checksum = XXH3_64bits_digest(uncompress_hash_state_);
uint64_t actual_checksum = XXH3_64bits(uncompressed_record_.data(),
uncompressed_record_.size());
if (*fragment_checksum != actual_checksum) {
// uncompressed_record_ contains bad content that does not match
// actual decompressed content
return kBadRecord;
}
}
*result = Slice(uncompressed_record_);
return type;
}
}
}
// Initialize uncompress related fields
void Reader::InitCompression(const CompressionTypeRecord& compression_record) {
compression_type_ = compression_record.GetCompressionType();
compression_type_record_read_ = true;
constexpr uint32_t compression_format_version = 2;
uncompress_ = StreamingUncompress::Create(
compression_type_, compression_format_version, kBlockSize);
assert(uncompress_ != nullptr);
uncompressed_buffer_ = std::unique_ptr<char[]>(new char[kBlockSize]);
assert(uncompressed_buffer_);
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
bool FragmentBufferedReader::ReadRecord(Slice* record, std::string* scratch,
Handoff checksum during WAL replay (#10212) Summary: Added checksum protection for write batch content read from WAL to when per key-value checksum is computed on the write batch. This gives full coverage on write batch integrity of WAL replay to memtable. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10212 Test Plan: - Added unit test and the existing tests (replay code path covers the change in this PR): `make -j32 check` - Stress test: ran `db_stress` for 30min. - Perf regression: ``` # setup TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ ./db_bench -benchmarks=fillrandom -write_buffer_size=1048576000 # benchmark db open time TEST_TMPDIR=/dev/shm/100MB_WAL_DB/ /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=overwrite -write_buffer_size=1048576000 -writes=1 -report_open_timing=true For 20 runs, pre-PR avg: 3734.31ms, post-PR avg: 3790.06 ms (~1.5% regression). Pre-PR OpenDb: 3714.36 milliseconds OpenDb: 3622.71 milliseconds OpenDb: 3591.17 milliseconds OpenDb: 3674.7 milliseconds OpenDb: 3615.79 milliseconds OpenDb: 3982.83 milliseconds OpenDb: 3650.6 milliseconds OpenDb: 3809.26 milliseconds OpenDb: 3576.44 milliseconds OpenDb: 3638.12 milliseconds OpenDb: 3845.68 milliseconds OpenDb: 3677.32 milliseconds OpenDb: 3659.64 milliseconds OpenDb: 3837.55 milliseconds OpenDb: 3899.64 milliseconds OpenDb: 3840.72 milliseconds OpenDb: 3802.71 milliseconds OpenDb: 3573.27 milliseconds OpenDb: 3895.76 milliseconds OpenDb: 3778.02 milliseconds Post-PR: OpenDb: 3880.46 milliseconds OpenDb: 3709.02 milliseconds OpenDb: 3954.67 milliseconds OpenDb: 3955.64 milliseconds OpenDb: 3958.64 milliseconds OpenDb: 3631.28 milliseconds OpenDb: 3721 milliseconds OpenDb: 3729.89 milliseconds OpenDb: 3730.55 milliseconds OpenDb: 3966.32 milliseconds OpenDb: 3685.54 milliseconds OpenDb: 3573.17 milliseconds OpenDb: 3703.75 milliseconds OpenDb: 3873.62 milliseconds OpenDb: 3704.4 milliseconds OpenDb: 3820.98 milliseconds OpenDb: 3721.62 milliseconds OpenDb: 3770.86 milliseconds OpenDb: 3949.78 milliseconds OpenDb: 3760.07 milliseconds ``` Reviewed By: ajkr Differential Revision: D37302092 Pulled By: cbi42 fbshipit-source-id: 7346e625f453ce4c0e5d708776cd1fb2af6b068b
2 years ago
WALRecoveryMode /*unused*/,
uint64_t* /* checksum */) {
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
assert(record != nullptr);
assert(scratch != nullptr);
record->clear();
scratch->clear();
if (uncompress_) {
uncompress_->Reset();
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
uint64_t prospective_record_offset = 0;
uint64_t physical_record_offset = end_of_buffer_offset_ - buffer_.size();
size_t drop_size = 0;
unsigned int fragment_type_or_err = 0; // Initialize to make compiler happy
Slice fragment;
while (TryReadFragment(&fragment, &drop_size, &fragment_type_or_err)) {
switch (fragment_type_or_err) {
case kFullType:
case kRecyclableFullType:
if (in_fragmented_record_ && !fragments_.empty()) {
ReportCorruption(fragments_.size(), "partial record without end(1)");
}
fragments_.clear();
*record = fragment;
prospective_record_offset = physical_record_offset;
last_record_offset_ = prospective_record_offset;
first_record_read_ = true;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
in_fragmented_record_ = false;
return true;
case kFirstType:
case kRecyclableFirstType:
if (in_fragmented_record_ || !fragments_.empty()) {
ReportCorruption(fragments_.size(), "partial record without end(2)");
}
prospective_record_offset = physical_record_offset;
fragments_.assign(fragment.data(), fragment.size());
in_fragmented_record_ = true;
break;
case kMiddleType:
case kRecyclableMiddleType:
if (!in_fragmented_record_) {
ReportCorruption(fragment.size(),
"missing start of fragmented record(1)");
} else {
fragments_.append(fragment.data(), fragment.size());
}
break;
case kLastType:
case kRecyclableLastType:
if (!in_fragmented_record_) {
ReportCorruption(fragment.size(),
"missing start of fragmented record(2)");
} else {
fragments_.append(fragment.data(), fragment.size());
scratch->assign(fragments_.data(), fragments_.size());
fragments_.clear();
*record = Slice(*scratch);
last_record_offset_ = prospective_record_offset;
first_record_read_ = true;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
in_fragmented_record_ = false;
return true;
}
break;
case kBadHeader:
case kBadRecord:
case kEof:
case kOldRecord:
if (in_fragmented_record_) {
ReportCorruption(fragments_.size(), "error in middle of record");
in_fragmented_record_ = false;
fragments_.clear();
}
break;
case kBadRecordChecksum:
if (recycled_) {
fragments_.clear();
return false;
}
ReportCorruption(drop_size, "checksum mismatch");
if (in_fragmented_record_) {
ReportCorruption(fragments_.size(), "error in middle of record");
in_fragmented_record_ = false;
fragments_.clear();
}
break;
case kSetCompressionType: {
if (compression_type_record_read_) {
ReportCorruption(fragment.size(),
"read multiple SetCompressionType records");
}
if (first_record_read_) {
ReportCorruption(fragment.size(),
"SetCompressionType not the first record");
}
fragments_.clear();
prospective_record_offset = physical_record_offset;
last_record_offset_ = prospective_record_offset;
in_fragmented_record_ = false;
CompressionTypeRecord compression_record(kNoCompression);
Status s = compression_record.DecodeFrom(&fragment);
if (!s.ok()) {
ReportCorruption(fragment.size(),
"could not decode SetCompressionType record");
} else {
InitCompression(compression_record);
}
break;
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
default: {
char buf[40];
snprintf(buf, sizeof(buf), "unknown record type %u",
fragment_type_or_err);
ReportCorruption(
fragment.size() + (in_fragmented_record_ ? fragments_.size() : 0),
buf);
in_fragmented_record_ = false;
fragments_.clear();
break;
}
}
}
return false;
}
void FragmentBufferedReader::UnmarkEOF() {
if (read_error_) {
return;
}
eof_ = false;
UnmarkEOFInternal();
}
bool FragmentBufferedReader::TryReadMore(size_t* drop_size, int* error) {
if (!eof_ && !read_error_) {
// Last read was a full read, so this is a trailer to skip
buffer_.clear();
Support read rate-limiting in SequentialFileReader (#9973) Summary: Added rate limiter and read rate-limiting support to SequentialFileReader. I've updated call sites to SequentialFileReader::Read with appropriate IO priority (or left a TODO and specified IO_TOTAL for now). The PR is separated into four commits: the first one added the rate-limiting support, but with some fixes in the unit test since the number of request bytes from rate limiter in SequentialFileReader are not accurate (there is overcharge at EOF). The second commit fixed this by allowing SequentialFileReader to check file size and determine how many bytes are left in the file to read. The third commit added benchmark related code. The fourth commit moved the logic of using file size to avoid overcharging the rate limiter into backup engine (the main user of SequentialFileReader). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9973 Test Plan: - `make check`, backup_engine_test covers usage of SequentialFileReader with rate limiter. - Run db_bench to check if rate limiting is throttling as expected: Verified that reads and writes are together throttled at 2MB/s, and at 0.2MB chunks that are 100ms apart. - Set up: `./db_bench --benchmarks=fillrandom -db=/dev/shm/test_rocksdb` - Benchmark: ``` strace -ttfe read,write ./db_bench --benchmarks=backup -db=/dev/shm/test_rocksdb --backup_rate_limit=2097152 --use_existing_db strace -ttfe read,write ./db_bench --benchmarks=restore -db=/dev/shm/test_rocksdb --restore_rate_limit=2097152 --use_existing_db ``` - db bench on backup and restore to ensure no performance regression. - backup (avg over 50 runs): pre-change: 1.90443e+06 micros/op; post-change: 1.8993e+06 micros/op (improve by 0.2%) - restore (avg over 50 runs): pre-change: 1.79105e+06 micros/op; post-change: 1.78192e+06 micros/op (improve by 0.5%) ``` # Set up ./db_bench --benchmarks=fillrandom -db=/tmp/test_rocksdb -num=10000000 # benchmark TEST_TMPDIR=/tmp/test_rocksdb NUM_RUN=50 for ((j=0;j<$NUM_RUN;j++)) do ./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=backup -use_existing_db | egrep 'backup' # Restore #./db_bench -db=$TEST_TMPDIR -num=10000000 -benchmarks=restore -use_existing_db done > rate_limit.txt && awk -v NUM_RUN=$NUM_RUN '{sum+=$3;sum_sqrt+=$3^2}END{print sum/NUM_RUN, sqrt(sum_sqrt/NUM_RUN-(sum/NUM_RUN)^2)}' rate_limit.txt >> rate_limit_2.txt ``` Reviewed By: hx235 Differential Revision: D36327418 Pulled By: cbi42 fbshipit-source-id: e75d4307cff815945482df5ba630c1e88d064691
2 years ago
// TODO: rate limit log reader with approriate priority.
// TODO: avoid overcharging rate limiter:
// Note that the Read here might overcharge SequentialFileReader's internal
// rate limiter if priority is not IO_TOTAL, e.g., when there is not enough
// content left until EOF to read.
Status status = file_->Read(kBlockSize, &buffer_, backing_store_,
Env::IO_TOTAL /* rate_limiter_priority */);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
end_of_buffer_offset_ += buffer_.size();
if (!status.ok()) {
buffer_.clear();
ReportDrop(kBlockSize, status);
read_error_ = true;
*error = kEof;
return false;
} else if (buffer_.size() < static_cast<size_t>(kBlockSize)) {
eof_ = true;
eof_offset_ = buffer_.size();
TEST_SYNC_POINT_CALLBACK(
"FragmentBufferedLogReader::TryReadMore:FirstEOF", nullptr);
}
return true;
} else if (!read_error_) {
UnmarkEOF();
}
if (!read_error_) {
return true;
}
*error = kEof;
*drop_size = buffer_.size();
if (buffer_.size() > 0) {
*error = kBadHeader;
}
buffer_.clear();
return false;
}
// return true if the caller should process the fragment_type_or_err.
bool FragmentBufferedReader::TryReadFragment(
Slice* fragment, size_t* drop_size, unsigned int* fragment_type_or_err) {
assert(fragment != nullptr);
assert(drop_size != nullptr);
assert(fragment_type_or_err != nullptr);
while (buffer_.size() < static_cast<size_t>(kHeaderSize)) {
size_t old_size = buffer_.size();
int error = kEof;
if (!TryReadMore(drop_size, &error)) {
*fragment_type_or_err = error;
return false;
} else if (old_size == buffer_.size()) {
return false;
}
}
const char* header = buffer_.data();
const uint32_t a = static_cast<uint32_t>(header[4]) & 0xff;
const uint32_t b = static_cast<uint32_t>(header[5]) & 0xff;
const unsigned int type = header[6];
const uint32_t length = a | (b << 8);
int header_size = kHeaderSize;
if (type >= kRecyclableFullType && type <= kRecyclableLastType) {
if (end_of_buffer_offset_ - buffer_.size() == 0) {
recycled_ = true;
}
header_size = kRecyclableHeaderSize;
while (buffer_.size() < static_cast<size_t>(kRecyclableHeaderSize)) {
size_t old_size = buffer_.size();
int error = kEof;
if (!TryReadMore(drop_size, &error)) {
*fragment_type_or_err = error;
return false;
} else if (old_size == buffer_.size()) {
return false;
}
}
const uint32_t log_num = DecodeFixed32(header + 7);
if (log_num != log_number_) {
*fragment_type_or_err = kOldRecord;
return true;
}
}
while (header_size + length > buffer_.size()) {
size_t old_size = buffer_.size();
int error = kEof;
if (!TryReadMore(drop_size, &error)) {
*fragment_type_or_err = error;
return false;
} else if (old_size == buffer_.size()) {
return false;
}
}
if (type == kZeroType && length == 0) {
buffer_.clear();
*fragment_type_or_err = kBadRecord;
return true;
}
if (checksum_) {
uint32_t expected_crc = crc32c::Unmask(DecodeFixed32(header));
uint32_t actual_crc = crc32c::Value(header + 6, length + header_size - 6);
if (actual_crc != expected_crc) {
*drop_size = buffer_.size();
buffer_.clear();
*fragment_type_or_err = kBadRecordChecksum;
return true;
}
}
buffer_.remove_prefix(header_size + length);
if (!uncompress_ || type == kSetCompressionType) {
*fragment = Slice(header + header_size, length);
*fragment_type_or_err = type;
return true;
} else {
// Uncompress compressed records
uncompressed_record_.clear();
size_t uncompressed_size = 0;
int remaining = 0;
do {
remaining = uncompress_->Uncompress(header + header_size, length,
uncompressed_buffer_.get(),
&uncompressed_size);
if (remaining < 0) {
buffer_.clear();
*fragment_type_or_err = kBadRecord;
return true;
}
if (uncompressed_size > 0) {
uncompressed_record_.append(uncompressed_buffer_.get(),
uncompressed_size);
}
} while (remaining > 0 || uncompressed_size == kBlockSize);
*fragment = Slice(std::move(uncompressed_record_));
*fragment_type_or_err = type;
return true;
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
}
} // namespace log
} // namespace ROCKSDB_NAMESPACE