2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-16 01:03:42 +02:00
|
|
|
// 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).
|
2013-10-16 23:59:46 +02:00
|
|
|
//
|
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 "db/log_reader.h"
|
|
|
|
#include "db/log_writer.h"
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
#include "env/composite_env_wrapper.h"
|
2019-09-16 19:31:27 +02:00
|
|
|
#include "file/sequence_file_reader.h"
|
|
|
|
#include "file/writable_file_writer.h"
|
2013-08-23 17:38:13 +02:00
|
|
|
#include "rocksdb/env.h"
|
2019-05-31 02:39:43 +02:00
|
|
|
#include "test_util/testharness.h"
|
|
|
|
#include "test_util/testutil.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "util/coding.h"
|
|
|
|
#include "util/crc32c.h"
|
|
|
|
#include "util/random.h"
|
|
|
|
|
2020-02-20 21:07:53 +01:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
2011-03-18 23:37:00 +01:00
|
|
|
namespace log {
|
|
|
|
|
|
|
|
// Construct a string of the specified length made out of the supplied
|
|
|
|
// partial string.
|
|
|
|
static std::string BigString(const std::string& partial_string, size_t n) {
|
|
|
|
std::string result;
|
|
|
|
while (result.size() < n) {
|
|
|
|
result.append(partial_string);
|
|
|
|
}
|
|
|
|
result.resize(n);
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Construct a string from a number
|
|
|
|
static std::string NumberString(int n) {
|
|
|
|
char buf[50];
|
|
|
|
snprintf(buf, sizeof(buf), "%d.", n);
|
|
|
|
return std::string(buf);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Return a skewed potentially long string
|
|
|
|
static std::string RandomSkewedString(int i, Random* rnd) {
|
|
|
|
return BigString(NumberString(i), rnd->Skewed(17));
|
|
|
|
}
|
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
// Param type is tuple<int, bool>
|
|
|
|
// get<0>(tuple): non-zero if recycling log, zero if regular log
|
|
|
|
// get<1>(tuple): true if allow retry after read EOF, false otherwise
|
|
|
|
class LogTest : public ::testing::TestWithParam<std::tuple<int, bool>> {
|
2011-03-18 23:37:00 +01:00
|
|
|
private:
|
|
|
|
class StringSource : public SequentialFile {
|
|
|
|
public:
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
Slice& contents_;
|
2011-03-18 23:37:00 +01:00
|
|
|
bool force_error_;
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
size_t force_error_position_;
|
|
|
|
bool force_eof_;
|
|
|
|
size_t force_eof_position_;
|
2011-03-18 23:37:00 +01:00
|
|
|
bool returned_partial_;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool fail_after_read_partial_;
|
|
|
|
explicit StringSource(Slice& contents, bool fail_after_read_partial)
|
|
|
|
: contents_(contents),
|
|
|
|
force_error_(false),
|
|
|
|
force_error_position_(0),
|
|
|
|
force_eof_(false),
|
|
|
|
force_eof_position_(0),
|
|
|
|
returned_partial_(false),
|
|
|
|
fail_after_read_partial_(fail_after_read_partial) {}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2019-02-14 22:52:47 +01:00
|
|
|
Status Read(size_t n, Slice* result, char* scratch) override {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (fail_after_read_partial_) {
|
|
|
|
EXPECT_TRUE(!returned_partial_) << "must not Read() after eof/error";
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
if (force_error_) {
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
if (force_error_position_ >= n) {
|
|
|
|
force_error_position_ -= n;
|
|
|
|
} else {
|
|
|
|
*result = Slice(contents_.data(), force_error_position_);
|
|
|
|
contents_.remove_prefix(force_error_position_);
|
|
|
|
force_error_ = false;
|
|
|
|
returned_partial_ = true;
|
|
|
|
return Status::Corruption("read error");
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (contents_.size() < n) {
|
|
|
|
n = contents_.size();
|
|
|
|
returned_partial_ = true;
|
|
|
|
}
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
|
|
|
|
if (force_eof_) {
|
|
|
|
if (force_eof_position_ >= n) {
|
|
|
|
force_eof_position_ -= n;
|
|
|
|
} else {
|
|
|
|
force_eof_ = false;
|
|
|
|
n = force_eof_position_;
|
|
|
|
returned_partial_ = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// By using scratch we ensure that caller has control over the
|
|
|
|
// lifetime of result.data()
|
|
|
|
memcpy(scratch, contents_.data(), n);
|
|
|
|
*result = Slice(scratch, n);
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
contents_.remove_prefix(n);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
2011-05-21 04:17:43 +02:00
|
|
|
|
2019-02-14 22:52:47 +01:00
|
|
|
Status Skip(uint64_t n) override {
|
2011-05-21 04:17:43 +02:00
|
|
|
if (n > contents_.size()) {
|
|
|
|
contents_.clear();
|
|
|
|
return Status::NotFound("in-memory file skipepd past end");
|
|
|
|
}
|
|
|
|
|
|
|
|
contents_.remove_prefix(n);
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
};
|
|
|
|
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
inline StringSource* GetStringSourceFromLegacyReader(
|
|
|
|
SequentialFileReader* reader) {
|
|
|
|
LegacySequentialFileWrapper* file =
|
|
|
|
static_cast<LegacySequentialFileWrapper*>(reader->file());
|
|
|
|
return static_cast<StringSource*>(file->target());
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
class ReportCollector : public Reader::Reporter {
|
|
|
|
public:
|
|
|
|
size_t dropped_bytes_;
|
|
|
|
std::string message_;
|
|
|
|
|
|
|
|
ReportCollector() : dropped_bytes_(0) { }
|
2019-02-14 22:52:47 +01:00
|
|
|
void Corruption(size_t bytes, const Status& status) override {
|
2011-03-18 23:37:00 +01:00
|
|
|
dropped_bytes_ += bytes;
|
|
|
|
message_.append(status.ToString());
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2013-01-20 11:07:13 +01:00
|
|
|
std::string& dest_contents() {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
|
2013-01-20 11:07:13 +01:00
|
|
|
assert(dest);
|
|
|
|
return dest->contents_;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::string& dest_contents() const {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
|
2013-01-20 11:07:13 +01:00
|
|
|
assert(dest);
|
|
|
|
return dest->contents_;
|
|
|
|
}
|
|
|
|
|
|
|
|
void reset_source_contents() {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto src = GetStringSourceFromLegacyReader(reader_->file());
|
2013-01-20 11:07:13 +01:00
|
|
|
assert(src);
|
|
|
|
src->contents_ = dest_contents();
|
|
|
|
}
|
|
|
|
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
Slice reader_contents_;
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<WritableFileWriter> dest_holder_;
|
|
|
|
std::unique_ptr<SequentialFileReader> source_holder_;
|
2011-03-18 23:37:00 +01:00
|
|
|
ReportCollector report_;
|
|
|
|
Writer writer_;
|
2019-03-27 00:41:31 +01:00
|
|
|
std::unique_ptr<Reader> reader_;
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
protected:
|
|
|
|
bool allow_retry_read_;
|
2011-05-21 04:17:43 +02:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
public:
|
Move rate_limiter, write buffering, most perf context instrumentation and most random kill out of Env
Summary: We want to keep Env a think layer for better portability. Less platform dependent codes should be moved out of Env. In this patch, I create a wrapper of file readers and writers, and put rate limiting, write buffering, as well as most perf context instrumentation and random kill out of Env. It will make it easier to maintain multiple Env in the future.
Test Plan: Run all existing unit tests.
Reviewers: anthony, kradhakrishnan, IslamAbdelRahman, yhchiang, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D42321
2015-07-18 01:16:11 +02:00
|
|
|
LogTest()
|
|
|
|
: reader_contents_(),
|
2015-10-08 19:07:15 +02:00
|
|
|
dest_holder_(test::GetWritableFileWriter(
|
2018-08-23 19:04:10 +02:00
|
|
|
new test::StringSink(&reader_contents_), "" /* don't care */)),
|
2018-06-21 17:34:24 +02:00
|
|
|
source_holder_(test::GetSequentialFileReader(
|
2019-03-27 00:41:31 +01:00
|
|
|
new StringSource(reader_contents_, !std::get<1>(GetParam())),
|
|
|
|
"" /* file name */)),
|
|
|
|
writer_(std::move(dest_holder_), 123, std::get<0>(GetParam())),
|
|
|
|
allow_retry_read_(std::get<1>(GetParam())) {
|
|
|
|
if (allow_retry_read_) {
|
|
|
|
reader_.reset(new FragmentBufferedReader(
|
|
|
|
nullptr, std::move(source_holder_), &report_, true /* checksum */,
|
|
|
|
123 /* log_number */));
|
|
|
|
} else {
|
|
|
|
reader_.reset(new Reader(nullptr, std::move(source_holder_), &report_,
|
|
|
|
true /* checksum */, 123 /* log_number */));
|
|
|
|
}
|
2015-10-19 23:24:05 +02:00
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2015-12-11 20:12:03 +01:00
|
|
|
Slice* get_reader_contents() { return &reader_contents_; }
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
void Write(const std::string& msg) {
|
|
|
|
writer_.AddRecord(Slice(msg));
|
|
|
|
}
|
|
|
|
|
|
|
|
size_t WrittenBytes() const {
|
2013-01-20 11:07:13 +01:00
|
|
|
return dest_contents().size();
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-02 21:53:59 +02:00
|
|
|
std::string Read(const WALRecoveryMode wal_recovery_mode =
|
|
|
|
WALRecoveryMode::kTolerateCorruptedTailRecords) {
|
2011-03-18 23:37:00 +01:00
|
|
|
std::string scratch;
|
|
|
|
Slice record;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool ret = false;
|
|
|
|
ret = reader_->ReadRecord(&record, &scratch, wal_recovery_mode);
|
|
|
|
if (ret) {
|
2011-03-18 23:37:00 +01:00
|
|
|
return record.ToString();
|
|
|
|
} else {
|
|
|
|
return "EOF";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-10-19 19:48:47 +02:00
|
|
|
void IncrementByte(int offset, char delta) {
|
2013-01-20 11:07:13 +01:00
|
|
|
dest_contents()[offset] += delta;
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void SetByte(int offset, char new_byte) {
|
2013-01-20 11:07:13 +01:00
|
|
|
dest_contents()[offset] = new_byte;
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void ShrinkSize(int bytes) {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto dest = test::GetStringSinkFromLegacyWriter(writer_.file());
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
assert(dest);
|
|
|
|
dest->Drop(bytes);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-19 23:24:05 +02:00
|
|
|
void FixChecksum(int header_offset, int len, bool recyclable) {
|
2011-03-18 23:37:00 +01:00
|
|
|
// Compute crc of type/len/data
|
2015-10-19 23:24:05 +02:00
|
|
|
int header_size = recyclable ? kRecyclableHeaderSize : kHeaderSize;
|
|
|
|
uint32_t crc = crc32c::Value(&dest_contents()[header_offset + 6],
|
|
|
|
header_size - 6 + len);
|
2011-03-18 23:37:00 +01:00
|
|
|
crc = crc32c::Mask(crc);
|
2013-01-20 11:07:13 +01:00
|
|
|
EncodeFixed32(&dest_contents()[header_offset], crc);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
void ForceError(size_t position = 0) {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto src = GetStringSourceFromLegacyReader(reader_->file());
|
2013-01-20 11:07:13 +01:00
|
|
|
src->force_error_ = true;
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
src->force_error_position_ = position;
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
size_t DroppedBytes() const {
|
|
|
|
return report_.dropped_bytes_;
|
|
|
|
}
|
|
|
|
|
2011-05-21 04:17:43 +02:00
|
|
|
std::string ReportMessage() const {
|
|
|
|
return report_.message_;
|
|
|
|
}
|
|
|
|
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
void ForceEOF(size_t position = 0) {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto src = GetStringSourceFromLegacyReader(reader_->file());
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
src->force_eof_ = true;
|
|
|
|
src->force_eof_position_ = position;
|
|
|
|
}
|
|
|
|
|
|
|
|
void UnmarkEOF() {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto src = GetStringSourceFromLegacyReader(reader_->file());
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
src->returned_partial_ = false;
|
2019-03-27 00:41:31 +01:00
|
|
|
reader_->UnmarkEOF();
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
}
|
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
bool IsEOF() { return reader_->IsEOF(); }
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
// Returns OK iff recorded error message contains "msg"
|
|
|
|
std::string MatchError(const std::string& msg) const {
|
|
|
|
if (report_.message_.find(msg) == std::string::npos) {
|
|
|
|
return report_.message_;
|
|
|
|
} else {
|
|
|
|
return "OK";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, Empty) { ASSERT_EQ("EOF", Read()); }
|
2011-05-21 04:17:43 +02:00
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ReadWrite) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
|
|
|
Write("");
|
|
|
|
Write("xxxx");
|
|
|
|
ASSERT_EQ("foo", Read());
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("", Read());
|
|
|
|
ASSERT_EQ("xxxx", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
ASSERT_EQ("EOF", Read()); // Make sure reads at eof work
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ManyBlocks) {
|
2011-03-18 23:37:00 +01:00
|
|
|
for (int i = 0; i < 100000; i++) {
|
|
|
|
Write(NumberString(i));
|
|
|
|
}
|
|
|
|
for (int i = 0; i < 100000; i++) {
|
|
|
|
ASSERT_EQ(NumberString(i), Read());
|
|
|
|
}
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, Fragmentation) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("small");
|
|
|
|
Write(BigString("medium", 50000));
|
|
|
|
Write(BigString("large", 100000));
|
|
|
|
ASSERT_EQ("small", Read());
|
|
|
|
ASSERT_EQ(BigString("medium", 50000), Read());
|
|
|
|
ASSERT_EQ(BigString("large", 100000), Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, MarginalTrailer) {
|
2011-03-18 23:37:00 +01:00
|
|
|
// Make a trailer that is exactly the same length as an empty record.
|
2019-03-27 00:41:31 +01:00
|
|
|
int header_size =
|
|
|
|
std::get<0>(GetParam()) ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
const int n = kBlockSize - 2 * header_size;
|
2011-03-18 23:37:00 +01:00
|
|
|
Write(BigString("foo", n));
|
2015-10-19 23:24:05 +02:00
|
|
|
ASSERT_EQ((unsigned int)(kBlockSize - header_size), WrittenBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("");
|
|
|
|
Write("bar");
|
|
|
|
ASSERT_EQ(BigString("foo", n), Read());
|
|
|
|
ASSERT_EQ("", Read());
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, MarginalTrailer2) {
|
2011-05-21 04:17:43 +02:00
|
|
|
// Make a trailer that is exactly the same length as an empty record.
|
2019-03-27 00:41:31 +01:00
|
|
|
int header_size =
|
|
|
|
std::get<0>(GetParam()) ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
const int n = kBlockSize - 2 * header_size;
|
2011-05-21 04:17:43 +02:00
|
|
|
Write(BigString("foo", n));
|
2015-10-19 23:24:05 +02:00
|
|
|
ASSERT_EQ((unsigned int)(kBlockSize - header_size), WrittenBytes());
|
2011-05-21 04:17:43 +02:00
|
|
|
Write("bar");
|
|
|
|
ASSERT_EQ(BigString("foo", n), Read());
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
2011-05-21 04:17:43 +02:00
|
|
|
ASSERT_EQ("", ReportMessage());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ShortTrailer) {
|
2019-03-27 00:41:31 +01:00
|
|
|
int header_size =
|
|
|
|
std::get<0>(GetParam()) ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
const int n = kBlockSize - 2 * header_size + 4;
|
2011-03-18 23:37:00 +01:00
|
|
|
Write(BigString("foo", n));
|
2015-10-19 23:24:05 +02:00
|
|
|
ASSERT_EQ((unsigned int)(kBlockSize - header_size + 4), WrittenBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("");
|
|
|
|
Write("bar");
|
|
|
|
ASSERT_EQ(BigString("foo", n), Read());
|
|
|
|
ASSERT_EQ("", Read());
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, AlignedEof) {
|
2019-03-27 00:41:31 +01:00
|
|
|
int header_size =
|
|
|
|
std::get<0>(GetParam()) ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
const int n = kBlockSize - 2 * header_size + 4;
|
2011-03-18 23:37:00 +01:00
|
|
|
Write(BigString("foo", n));
|
2015-10-19 23:24:05 +02:00
|
|
|
ASSERT_EQ((unsigned int)(kBlockSize - header_size + 4), WrittenBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ(BigString("foo", n), Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, RandomRead) {
|
2011-03-18 23:37:00 +01:00
|
|
|
const int N = 500;
|
|
|
|
Random write_rnd(301);
|
|
|
|
for (int i = 0; i < N; i++) {
|
|
|
|
Write(RandomSkewedString(i, &write_rnd));
|
|
|
|
}
|
|
|
|
Random read_rnd(301);
|
|
|
|
for (int i = 0; i < N; i++) {
|
|
|
|
ASSERT_EQ(RandomSkewedString(i, &read_rnd), Read());
|
|
|
|
}
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Tests of all the error paths in log_reader.cc follow:
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ReadError) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
ForceError();
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ((unsigned int)kBlockSize, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("read error"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, BadRecordType) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
// Type is stored in header[6]
|
|
|
|
IncrementByte(6, 100);
|
2015-10-19 23:24:05 +02:00
|
|
|
FixChecksum(0, 3, false);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("unknown record type"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, TruncatedTrailingRecordIsIgnored) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
ShrinkSize(4); // Drop all payload as well as a header byte
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2014-02-28 22:19:47 +01:00
|
|
|
// Truncated last record is ignored, not treated as an error
|
2014-03-14 23:44:35 +01:00
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
2014-02-28 22:19:47 +01:00
|
|
|
ASSERT_EQ("", ReportMessage());
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, TruncatedTrailingRecordIsNotIgnored) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then truncated trailing record should not
|
|
|
|
// raise an error.
|
|
|
|
return;
|
|
|
|
}
|
2015-06-15 21:03:13 +02:00
|
|
|
Write("foo");
|
|
|
|
ShrinkSize(4); // Drop all payload as well as a header byte
|
2015-10-02 21:53:59 +02:00
|
|
|
ASSERT_EQ("EOF", Read(WALRecoveryMode::kAbsoluteConsistency));
|
2015-06-15 21:03:13 +02:00
|
|
|
// Truncated last record is ignored, not treated as an error
|
|
|
|
ASSERT_GT(DroppedBytes(), 0U);
|
|
|
|
ASSERT_EQ("OK", MatchError("Corruption: truncated header"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, BadLength) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then we should not raise an error when the
|
|
|
|
// record length specified in header is longer than data currently
|
|
|
|
// available. It's possible that the body of the record is not written yet.
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
int header_size = recyclable_log ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
const int kPayloadSize = kBlockSize - header_size;
|
2014-02-28 22:19:47 +01:00
|
|
|
Write(BigString("bar", kPayloadSize));
|
|
|
|
Write("foo");
|
|
|
|
// Least significant size byte is stored in header[4].
|
|
|
|
IncrementByte(4, 1);
|
2019-03-27 00:41:31 +01:00
|
|
|
if (!recyclable_log) {
|
2015-12-11 23:17:43 +01:00
|
|
|
ASSERT_EQ("foo", Read());
|
|
|
|
ASSERT_EQ(kBlockSize, DroppedBytes());
|
|
|
|
ASSERT_EQ("OK", MatchError("bad record length"));
|
|
|
|
} else {
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
2014-02-28 22:19:47 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, BadLengthAtEndIsIgnored) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then we should not raise an error when the
|
|
|
|
// record length specified in header is longer than data currently
|
|
|
|
// available. It's possible that the body of the record is not written yet.
|
|
|
|
return;
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
ShrinkSize(1);
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2014-03-14 23:44:35 +01:00
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
2014-02-28 22:19:47 +01:00
|
|
|
ASSERT_EQ("", ReportMessage());
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, BadLengthAtEndIsNotIgnored) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then we should not raise an error when the
|
|
|
|
// record length specified in header is longer than data currently
|
|
|
|
// available. It's possible that the body of the record is not written yet.
|
|
|
|
return;
|
|
|
|
}
|
2015-06-15 21:03:13 +02:00
|
|
|
Write("foo");
|
|
|
|
ShrinkSize(1);
|
2015-10-02 21:53:59 +02:00
|
|
|
ASSERT_EQ("EOF", Read(WALRecoveryMode::kAbsoluteConsistency));
|
2015-06-15 21:03:13 +02:00
|
|
|
ASSERT_GT(DroppedBytes(), 0U);
|
2020-12-01 03:10:18 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("Corruption: truncated record body"));
|
2015-06-15 21:03:13 +02:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ChecksumMismatch) {
|
2015-10-19 23:24:05 +02:00
|
|
|
Write("foooooo");
|
|
|
|
IncrementByte(0, 14);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("EOF", Read());
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
if (!recyclable_log) {
|
2015-12-11 23:17:43 +01:00
|
|
|
ASSERT_EQ(14U, DroppedBytes());
|
|
|
|
ASSERT_EQ("OK", MatchError("checksum mismatch"));
|
|
|
|
} else {
|
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
|
|
|
ASSERT_EQ("", ReportMessage());
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, UnexpectedMiddleType) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
SetByte(6, static_cast<char>(recyclable_log ? kRecyclableMiddleType
|
|
|
|
: kMiddleType));
|
|
|
|
FixChecksum(0, 3, !!recyclable_log);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("missing start"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, UnexpectedLastType) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
SetByte(6,
|
|
|
|
static_cast<char>(recyclable_log ? kRecyclableLastType : kLastType));
|
|
|
|
FixChecksum(0, 3, !!recyclable_log);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("missing start"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, UnexpectedFullType) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
SetByte(
|
|
|
|
6, static_cast<char>(recyclable_log ? kRecyclableFirstType : kFirstType));
|
|
|
|
FixChecksum(0, 3, !!recyclable_log);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("partial record without end"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, UnexpectedFirstType) {
|
2011-03-18 23:37:00 +01:00
|
|
|
Write("foo");
|
|
|
|
Write(BigString("bar", 100000));
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
SetByte(
|
|
|
|
6, static_cast<char>(recyclable_log ? kRecyclableFirstType : kFirstType));
|
|
|
|
FixChecksum(0, 3, !!recyclable_log);
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ(BigString("bar", 100000), Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
2012-11-06 21:02:18 +01:00
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
2011-03-18 23:37:00 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("partial record without end"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, MissingLastIsIgnored) {
|
2014-02-28 22:19:47 +01:00
|
|
|
Write(BigString("bar", kBlockSize));
|
|
|
|
// Remove the LAST block, including header.
|
|
|
|
ShrinkSize(14);
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
ASSERT_EQ("", ReportMessage());
|
2014-03-14 23:44:35 +01:00
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
2014-02-28 22:19:47 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, MissingLastIsNotIgnored) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then truncated trailing record should not
|
|
|
|
// raise an error.
|
|
|
|
return;
|
|
|
|
}
|
2015-06-15 21:03:13 +02:00
|
|
|
Write(BigString("bar", kBlockSize));
|
|
|
|
// Remove the LAST block, including header.
|
|
|
|
ShrinkSize(14);
|
2015-10-02 21:53:59 +02:00
|
|
|
ASSERT_EQ("EOF", Read(WALRecoveryMode::kAbsoluteConsistency));
|
2015-06-15 21:03:13 +02:00
|
|
|
ASSERT_GT(DroppedBytes(), 0U);
|
|
|
|
ASSERT_EQ("OK", MatchError("Corruption: error reading trailing data"));
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, PartialLastIsIgnored) {
|
2014-02-28 22:19:47 +01:00
|
|
|
Write(BigString("bar", kBlockSize));
|
|
|
|
// Cause a bad record length in the LAST block.
|
|
|
|
ShrinkSize(1);
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
ASSERT_EQ("", ReportMessage());
|
2014-03-14 23:44:35 +01:00
|
|
|
ASSERT_EQ(0U, DroppedBytes());
|
2014-02-28 22:19:47 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, PartialLastIsNotIgnored) {
|
2019-03-27 00:41:31 +01:00
|
|
|
if (allow_retry_read_) {
|
|
|
|
// If read retry is allowed, then truncated trailing record should not
|
|
|
|
// raise an error.
|
|
|
|
return;
|
|
|
|
}
|
2015-06-15 21:03:13 +02:00
|
|
|
Write(BigString("bar", kBlockSize));
|
|
|
|
// Cause a bad record length in the LAST block.
|
|
|
|
ShrinkSize(1);
|
2015-10-02 21:53:59 +02:00
|
|
|
ASSERT_EQ("EOF", Read(WALRecoveryMode::kAbsoluteConsistency));
|
2015-06-15 21:03:13 +02:00
|
|
|
ASSERT_GT(DroppedBytes(), 0U);
|
2020-12-01 03:10:18 +01:00
|
|
|
ASSERT_EQ("OK", MatchError("Corruption: truncated record body"));
|
2015-06-15 21:03:13 +02:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ErrorJoinsRecords) {
|
2011-03-18 23:37:00 +01:00
|
|
|
// Consider two fragmented records:
|
|
|
|
// first(R1) last(R1) first(R2) last(R2)
|
|
|
|
// where the middle two fragments disappear. We do not want
|
|
|
|
// first(R1),last(R2) to get joined and returned as a valid record.
|
|
|
|
|
|
|
|
// Write records that span two blocks
|
|
|
|
Write(BigString("foo", kBlockSize));
|
|
|
|
Write(BigString("bar", kBlockSize));
|
|
|
|
Write("correct");
|
|
|
|
|
|
|
|
// Wipe the middle block
|
2013-03-19 22:53:22 +01:00
|
|
|
for (unsigned int offset = kBlockSize; offset < 2*kBlockSize; offset++) {
|
2011-03-18 23:37:00 +01:00
|
|
|
SetByte(offset, 'x');
|
|
|
|
}
|
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
if (!recyclable_log) {
|
2015-12-11 23:17:43 +01:00
|
|
|
ASSERT_EQ("correct", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
size_t dropped = DroppedBytes();
|
|
|
|
ASSERT_LE(dropped, 2 * kBlockSize + 100);
|
|
|
|
ASSERT_GE(dropped, 2 * kBlockSize);
|
|
|
|
} else {
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ClearEofSingleBlock) {
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
int header_size = recyclable_log ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
ForceEOF(3 + header_size + 2);
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
ASSERT_EQ("foo", Read());
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
Write("xxx");
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("xxx", Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ClearEofMultiBlock) {
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
size_t num_full_blocks = 5;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
int header_size = recyclable_log ? kRecyclableHeaderSize : kHeaderSize;
|
2015-10-19 23:24:05 +02:00
|
|
|
size_t n = (kBlockSize - header_size) * num_full_blocks + 25;
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
Write(BigString("foo", n));
|
|
|
|
Write(BigString("bar", n));
|
2015-10-19 23:24:05 +02:00
|
|
|
ForceEOF(n + num_full_blocks * header_size + header_size + 3);
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
ASSERT_EQ(BigString("foo", n), Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ(BigString("bar", n), Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
Write(BigString("xxx", n));
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ(BigString("xxx", n), Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ClearEofError) {
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
// If an error occurs during Read() in UnmarkEOF(), the records contained
|
|
|
|
// in the buffer should be returned on subsequent calls of ReadRecord()
|
|
|
|
// until no more full records are left, whereafter ReadRecord() should return
|
|
|
|
// false to indicate that it cannot read any further.
|
|
|
|
|
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("foo", Read());
|
|
|
|
ASSERT_TRUE(IsEOF());
|
|
|
|
Write("xxx");
|
|
|
|
ForceError(0);
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2015-10-08 19:07:15 +02:00
|
|
|
TEST_P(LogTest, ClearEofError2) {
|
Fix UnmarkEOF for partial blocks
Summary:
Blocks in the transaction log are a fixed size, but the last block in the transaction log file is usually a partial block. When a new record is added after the reader hit the end of the file, a new physical record will be appended to the last block. ReadPhysicalRecord can only read full blocks and assumes that the file position indicator is aligned to the start of a block. If the reader is forced to read further by simply clearing the EOF flag, ReadPhysicalRecord will read a full block starting from somewhere in the middle of a real block, causing it to lose alignment and to have a partial physical record at the end of the read buffer. This will result in length mismatches and checksum failures. When the log file is tailed for replication this will cause the log iterator to become invalid, necessitating the creation of a new iterator which will have to read the log file from scratch.
This diff fixes this issue by reading the remaining portion of the last block we read from. This is done when the reader is forced to read further (UnmarkEOF is called).
Test Plan:
- Added unit tests
- Stress test (with replication). Check dbdir/LOG file for corruptions.
- Test on test tier
Reviewers: emayanke, haobo, dhruba
Reviewed By: haobo
CC: vamsi, sheki, dhruba, kailiu, igor
Differential Revision: https://reviews.facebook.net/D15249
2014-01-27 23:49:10 +01:00
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("foo", Read());
|
|
|
|
Write("xxx");
|
|
|
|
ForceError(3);
|
|
|
|
UnmarkEOF();
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
ASSERT_EQ(3U, DroppedBytes());
|
|
|
|
ASSERT_EQ("OK", MatchError("read error"));
|
|
|
|
}
|
|
|
|
|
2015-12-11 20:12:03 +01:00
|
|
|
TEST_P(LogTest, Recycle) {
|
2019-03-27 00:41:31 +01:00
|
|
|
bool recyclable_log = (std::get<0>(GetParam()) != 0);
|
|
|
|
if (!recyclable_log) {
|
2015-12-11 20:12:03 +01:00
|
|
|
return; // test is only valid for recycled logs
|
|
|
|
}
|
|
|
|
Write("foo");
|
|
|
|
Write("bar");
|
|
|
|
Write("baz");
|
|
|
|
Write("bif");
|
|
|
|
Write("blitz");
|
|
|
|
while (get_reader_contents()->size() < log::kBlockSize * 2) {
|
|
|
|
Write("xxxxxxxxxxxxxxxx");
|
|
|
|
}
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<WritableFileWriter> dest_holder(test::GetWritableFileWriter(
|
2018-08-23 19:04:10 +02:00
|
|
|
new test::OverwritingStringSink(get_reader_contents()),
|
|
|
|
"" /* don't care */));
|
2015-12-11 20:12:03 +01:00
|
|
|
Writer recycle_writer(std::move(dest_holder), 123, true);
|
|
|
|
recycle_writer.AddRecord(Slice("foooo"));
|
|
|
|
recycle_writer.AddRecord(Slice("bar"));
|
|
|
|
ASSERT_GE(get_reader_contents()->size(), log::kBlockSize * 2);
|
|
|
|
ASSERT_EQ("foooo", Read());
|
|
|
|
ASSERT_EQ("bar", Read());
|
|
|
|
ASSERT_EQ("EOF", Read());
|
|
|
|
}
|
|
|
|
|
2020-06-04 00:53:09 +02:00
|
|
|
INSTANTIATE_TEST_CASE_P(bool, LogTest,
|
|
|
|
::testing::Values(std::make_tuple(0, false),
|
|
|
|
std::make_tuple(0, true),
|
|
|
|
std::make_tuple(1, false),
|
|
|
|
std::make_tuple(1, true)));
|
2015-10-08 19:07:15 +02:00
|
|
|
|
2018-10-19 20:51:13 +02:00
|
|
|
class RetriableLogTest : public ::testing::TestWithParam<int> {
|
|
|
|
private:
|
|
|
|
class ReportCollector : public Reader::Reporter {
|
|
|
|
public:
|
|
|
|
size_t dropped_bytes_;
|
|
|
|
std::string message_;
|
|
|
|
|
|
|
|
ReportCollector() : dropped_bytes_(0) {}
|
2019-02-14 22:52:47 +01:00
|
|
|
void Corruption(size_t bytes, const Status& status) override {
|
2018-10-19 20:51:13 +02:00
|
|
|
dropped_bytes_ += bytes;
|
|
|
|
message_.append(status.ToString());
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
Slice contents_;
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<WritableFileWriter> dest_holder_;
|
|
|
|
std::unique_ptr<Writer> log_writer_;
|
2018-10-19 20:51:13 +02:00
|
|
|
Env* env_;
|
|
|
|
EnvOptions env_options_;
|
|
|
|
const std::string test_dir_;
|
|
|
|
const std::string log_file_;
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<WritableFileWriter> writer_;
|
|
|
|
std::unique_ptr<SequentialFileReader> reader_;
|
2018-10-19 20:51:13 +02:00
|
|
|
ReportCollector report_;
|
2019-03-27 00:41:31 +01:00
|
|
|
std::unique_ptr<FragmentBufferedReader> log_reader_;
|
2018-10-19 20:51:13 +02:00
|
|
|
|
|
|
|
public:
|
|
|
|
RetriableLogTest()
|
|
|
|
: contents_(),
|
|
|
|
dest_holder_(nullptr),
|
|
|
|
log_writer_(nullptr),
|
|
|
|
env_(Env::Default()),
|
|
|
|
test_dir_(test::PerThreadDBPath("retriable_log_test")),
|
|
|
|
log_file_(test_dir_ + "/log"),
|
|
|
|
writer_(nullptr),
|
|
|
|
reader_(nullptr),
|
|
|
|
log_reader_(nullptr) {}
|
|
|
|
|
|
|
|
Status SetupTestEnv() {
|
|
|
|
dest_holder_.reset(test::GetWritableFileWriter(
|
|
|
|
new test::StringSink(&contents_), "" /* file name */));
|
|
|
|
assert(dest_holder_ != nullptr);
|
|
|
|
log_writer_.reset(new Writer(std::move(dest_holder_), 123, GetParam()));
|
|
|
|
assert(log_writer_ != nullptr);
|
|
|
|
|
|
|
|
Status s;
|
|
|
|
s = env_->CreateDirIfMissing(test_dir_);
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<WritableFile> writable_file;
|
2018-10-19 20:51:13 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
s = env_->NewWritableFile(log_file_, &writable_file, env_options_);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
writer_.reset(new WritableFileWriter(
|
|
|
|
NewLegacyWritableFileWrapper(std::move(writable_file)), log_file_,
|
|
|
|
env_options_));
|
2018-10-19 20:51:13 +02:00
|
|
|
assert(writer_ != nullptr);
|
|
|
|
}
|
2018-11-09 20:17:34 +01:00
|
|
|
std::unique_ptr<SequentialFile> seq_file;
|
2018-10-19 20:51:13 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
s = env_->NewSequentialFile(log_file_, &seq_file, env_options_);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
reader_.reset(new SequentialFileReader(
|
|
|
|
NewLegacySequentialFileWrapper(seq_file), log_file_));
|
2018-10-19 20:51:13 +02:00
|
|
|
assert(reader_ != nullptr);
|
2019-03-27 00:41:31 +01:00
|
|
|
log_reader_.reset(new FragmentBufferedReader(
|
|
|
|
nullptr, std::move(reader_), &report_, true /* checksum */,
|
|
|
|
123 /* log_number */));
|
2018-10-19 20:51:13 +02:00
|
|
|
assert(log_reader_ != nullptr);
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string contents() {
|
Introduce a new storage specific Env API (#5761)
Summary:
The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc.
This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO.
The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before.
This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection.
The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761
Differential Revision: D18868376
Pulled By: anand1976
fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
|
|
|
auto file = test::GetStringSinkFromLegacyWriter(log_writer_->file());
|
2018-10-19 20:51:13 +02:00
|
|
|
assert(file != nullptr);
|
|
|
|
return file->contents_;
|
|
|
|
}
|
|
|
|
|
|
|
|
void Encode(const std::string& msg) { log_writer_->AddRecord(Slice(msg)); }
|
|
|
|
|
|
|
|
void Write(const Slice& data) {
|
|
|
|
writer_->Append(data);
|
|
|
|
writer_->Sync(true);
|
|
|
|
}
|
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
bool TryRead(std::string* result) {
|
|
|
|
assert(result != nullptr);
|
|
|
|
result->clear();
|
2018-10-19 20:51:13 +02:00
|
|
|
std::string scratch;
|
|
|
|
Slice record;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool r = log_reader_->ReadRecord(&record, &scratch);
|
|
|
|
if (r) {
|
|
|
|
result->assign(record.data(), record.size());
|
|
|
|
return true;
|
2018-10-19 20:51:13 +02:00
|
|
|
} else {
|
2019-03-27 00:41:31 +01:00
|
|
|
return false;
|
2018-10-19 20:51:13 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
TEST_P(RetriableLogTest, TailLog_PartialHeader) {
|
|
|
|
ASSERT_OK(SetupTestEnv());
|
|
|
|
std::vector<int> remaining_bytes_in_last_record;
|
|
|
|
size_t header_size = GetParam() ? kRecyclableHeaderSize : kHeaderSize;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool eof = false;
|
2018-10-19 20:51:13 +02:00
|
|
|
SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
SyncPoint::GetInstance()->LoadDependency(
|
|
|
|
{{"RetriableLogTest::TailLog:AfterPart1",
|
|
|
|
"RetriableLogTest::TailLog:BeforeReadRecord"},
|
2019-03-27 00:41:31 +01:00
|
|
|
{"FragmentBufferedLogReader::TryReadMore:FirstEOF",
|
2018-10-19 20:51:13 +02:00
|
|
|
"RetriableLogTest::TailLog:BeforePart2"}});
|
2019-03-27 00:41:31 +01:00
|
|
|
SyncPoint::GetInstance()->ClearAllCallBacks();
|
|
|
|
SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"FragmentBufferedLogReader::TryReadMore:FirstEOF",
|
|
|
|
[&](void* /*arg*/) { eof = true; });
|
2018-10-19 20:51:13 +02:00
|
|
|
SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
size_t delta = header_size - 1;
|
|
|
|
port::Thread log_writer_thread([&]() {
|
|
|
|
size_t old_sz = contents().size();
|
|
|
|
Encode("foo");
|
|
|
|
size_t new_sz = contents().size();
|
|
|
|
std::string part1 = contents().substr(old_sz, delta);
|
|
|
|
std::string part2 =
|
|
|
|
contents().substr(old_sz + delta, new_sz - old_sz - delta);
|
|
|
|
Write(Slice(part1));
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:AfterPart1");
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:BeforePart2");
|
|
|
|
Write(Slice(part2));
|
|
|
|
});
|
|
|
|
|
|
|
|
std::string record;
|
|
|
|
port::Thread log_reader_thread([&]() {
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:BeforeReadRecord");
|
2019-03-27 00:41:31 +01:00
|
|
|
while (!TryRead(&record)) {
|
|
|
|
}
|
2018-10-19 20:51:13 +02:00
|
|
|
});
|
|
|
|
log_reader_thread.join();
|
|
|
|
log_writer_thread.join();
|
|
|
|
ASSERT_EQ("foo", record);
|
2019-03-27 00:41:31 +01:00
|
|
|
ASSERT_TRUE(eof);
|
2018-10-19 20:51:13 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST_P(RetriableLogTest, TailLog_FullHeader) {
|
|
|
|
ASSERT_OK(SetupTestEnv());
|
|
|
|
std::vector<int> remaining_bytes_in_last_record;
|
|
|
|
size_t header_size = GetParam() ? kRecyclableHeaderSize : kHeaderSize;
|
2019-03-27 00:41:31 +01:00
|
|
|
bool eof = false;
|
2018-10-19 20:51:13 +02:00
|
|
|
SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
SyncPoint::GetInstance()->LoadDependency(
|
|
|
|
{{"RetriableLogTest::TailLog:AfterPart1",
|
|
|
|
"RetriableLogTest::TailLog:BeforeReadRecord"},
|
2019-03-27 00:41:31 +01:00
|
|
|
{"FragmentBufferedLogReader::TryReadMore:FirstEOF",
|
2018-10-19 20:51:13 +02:00
|
|
|
"RetriableLogTest::TailLog:BeforePart2"}});
|
2019-03-27 00:41:31 +01:00
|
|
|
SyncPoint::GetInstance()->ClearAllCallBacks();
|
|
|
|
SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"FragmentBufferedLogReader::TryReadMore:FirstEOF",
|
|
|
|
[&](void* /*arg*/) { eof = true; });
|
2018-10-19 20:51:13 +02:00
|
|
|
SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
size_t delta = header_size + 1;
|
|
|
|
port::Thread log_writer_thread([&]() {
|
|
|
|
size_t old_sz = contents().size();
|
|
|
|
Encode("foo");
|
|
|
|
size_t new_sz = contents().size();
|
|
|
|
std::string part1 = contents().substr(old_sz, delta);
|
|
|
|
std::string part2 =
|
|
|
|
contents().substr(old_sz + delta, new_sz - old_sz - delta);
|
|
|
|
Write(Slice(part1));
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:AfterPart1");
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:BeforePart2");
|
|
|
|
Write(Slice(part2));
|
2019-03-27 00:41:31 +01:00
|
|
|
ASSERT_TRUE(eof);
|
2018-10-19 20:51:13 +02:00
|
|
|
});
|
|
|
|
|
|
|
|
std::string record;
|
|
|
|
port::Thread log_reader_thread([&]() {
|
|
|
|
TEST_SYNC_POINT("RetriableLogTest::TailLog:BeforeReadRecord");
|
2019-03-27 00:41:31 +01:00
|
|
|
while (!TryRead(&record)) {
|
|
|
|
}
|
2018-10-19 20:51:13 +02:00
|
|
|
});
|
|
|
|
log_reader_thread.join();
|
|
|
|
log_writer_thread.join();
|
|
|
|
ASSERT_EQ("foo", record);
|
|
|
|
}
|
|
|
|
|
2019-03-27 00:41:31 +01:00
|
|
|
TEST_P(RetriableLogTest, NonBlockingReadFullRecord) {
|
|
|
|
// Clear all sync point callbacks even if this test does not use sync point.
|
|
|
|
// It is necessary, otherwise the execute of this test may hit a sync point
|
|
|
|
// with which a callback is registered. The registered callback may access
|
|
|
|
// some dead variable, causing segfault.
|
|
|
|
SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
SyncPoint::GetInstance()->ClearAllCallBacks();
|
|
|
|
ASSERT_OK(SetupTestEnv());
|
|
|
|
size_t header_size = GetParam() ? kRecyclableHeaderSize : kHeaderSize;
|
|
|
|
size_t delta = header_size - 1;
|
|
|
|
size_t old_sz = contents().size();
|
|
|
|
Encode("foo-bar");
|
|
|
|
size_t new_sz = contents().size();
|
|
|
|
std::string part1 = contents().substr(old_sz, delta);
|
|
|
|
std::string part2 =
|
|
|
|
contents().substr(old_sz + delta, new_sz - old_sz - delta);
|
|
|
|
Write(Slice(part1));
|
|
|
|
std::string record;
|
|
|
|
ASSERT_FALSE(TryRead(&record));
|
|
|
|
ASSERT_TRUE(record.empty());
|
|
|
|
Write(Slice(part2));
|
|
|
|
ASSERT_TRUE(TryRead(&record));
|
|
|
|
ASSERT_EQ("foo-bar", record);
|
|
|
|
}
|
|
|
|
|
2020-06-04 00:53:09 +02:00
|
|
|
INSTANTIATE_TEST_CASE_P(bool, RetriableLogTest, ::testing::Values(0, 2));
|
2018-10-19 20:51:13 +02:00
|
|
|
|
2011-10-31 18:22:06 +01:00
|
|
|
} // namespace log
|
2020-02-20 21:07:53 +01:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
int main(int argc, char** argv) {
|
2015-03-17 22:08:00 +01:00
|
|
|
::testing::InitGoogleTest(&argc, argv);
|
|
|
|
return RUN_ALL_TESTS();
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|