2019-05-31 18:52:59 +00:00
|
|
|
|
2017-05-10 21:54:35 +00:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-15 23:03:42 +00:00
|
|
|
// 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).
|
2017-05-10 21:54:35 +00:00
|
|
|
#ifndef ROCKSDB_LITE
|
2017-10-13 21:36:36 +00:00
|
|
|
#include "utilities/blob_db/blob_file.h"
|
2017-05-10 21:54:35 +00:00
|
|
|
|
2017-05-16 18:50:43 +00:00
|
|
|
#include <stdio.h>
|
2019-09-20 19:00:55 +00:00
|
|
|
#include <cinttypes>
|
2017-10-17 19:11:52 +00:00
|
|
|
|
|
|
|
#include <algorithm>
|
2017-05-10 21:54:35 +00:00
|
|
|
#include <memory>
|
|
|
|
|
2017-10-27 20:14:34 +00:00
|
|
|
#include "db/column_family.h"
|
2019-05-31 18:52:59 +00:00
|
|
|
#include "db/db_impl/db_impl.h"
|
Blob DB: Inline small values in base DB
Summary:
Adding the `min_blob_size` option to allow storing small values in base db (in LSM tree) together with the key. The goal is to improve performance for small values, while taking advantage of blob db's low write amplification for large values.
Also adding expiration timestamp to blob index. It will be useful to evict stale blob indexes in base db by adding a compaction filter. I'll work on the compaction filter in future patches.
See blob_index.h for the new blob index format. There are 4 cases when writing a new key:
* small value w/o TTL: put in base db as normal value (i.e. ValueType::kTypeValue)
* small value w/ TTL: put (type, expiration, value) to base db.
* large value w/o TTL: write value to blob log and put (type, file, offset, size, compression) to base db.
* large value w/TTL: write value to blob log and put (type, expiration, file, offset, size, compression) to base db.
Closes https://github.com/facebook/rocksdb/pull/3066
Differential Revision: D6142115
Pulled By: yiwu-arbug
fbshipit-source-id: 9526e76e19f0839310a3f5f2a43772a4ad182cd0
2017-10-26 19:19:43 +00:00
|
|
|
#include "db/dbformat.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 22:47:08 +00:00
|
|
|
#include "env/composite_env_wrapper.h"
|
2019-05-30 03:44:08 +00:00
|
|
|
#include "file/filename.h"
|
2019-09-16 17:31:27 +00:00
|
|
|
#include "file/readahead_raf.h"
|
2019-06-01 00:19:43 +00:00
|
|
|
#include "logging/logging.h"
|
2017-10-13 21:36:36 +00:00
|
|
|
#include "utilities/blob_db/blob_db_impl.h"
|
2017-05-10 21:54:35 +00:00
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
2017-05-10 21:54:35 +00:00
|
|
|
|
|
|
|
namespace blob_db {
|
|
|
|
|
2017-12-11 20:01:22 +00:00
|
|
|
BlobFile::BlobFile(const BlobDBImpl* p, const std::string& bdir, uint64_t fn,
|
|
|
|
Logger* info_log)
|
2019-11-26 21:16:39 +00:00
|
|
|
: parent_(p), path_to_dir_(bdir), file_number_(fn), info_log_(info_log) {}
|
|
|
|
|
|
|
|
BlobFile::BlobFile(const BlobDBImpl* p, const std::string& bdir, uint64_t fn,
|
|
|
|
Logger* info_log, uint32_t column_family_id,
|
|
|
|
CompressionType compression, bool has_ttl,
|
|
|
|
const ExpirationRange& expiration_range)
|
2017-05-10 21:54:35 +00:00
|
|
|
: parent_(p),
|
|
|
|
path_to_dir_(bdir),
|
|
|
|
file_number_(fn),
|
2017-12-11 20:01:22 +00:00
|
|
|
info_log_(info_log),
|
2019-11-26 21:16:39 +00:00
|
|
|
column_family_id_(column_family_id),
|
|
|
|
compression_(compression),
|
|
|
|
has_ttl_(has_ttl),
|
|
|
|
expiration_range_(expiration_range),
|
|
|
|
header_(column_family_id, compression, has_ttl, expiration_range),
|
|
|
|
header_valid_(true) {}
|
2017-05-10 21:54:35 +00:00
|
|
|
|
|
|
|
BlobFile::~BlobFile() {
|
2017-11-02 22:50:30 +00:00
|
|
|
if (obsolete_) {
|
2017-05-10 21:54:35 +00:00
|
|
|
std::string pn(PathName());
|
|
|
|
Status s = Env::Default()->DeleteFile(PathName());
|
|
|
|
if (!s.ok()) {
|
2017-07-13 17:03:39 +00:00
|
|
|
// ROCKS_LOG_INFO(db_options_.info_log,
|
2017-05-10 21:54:35 +00:00
|
|
|
// "File could not be deleted %s", pn.c_str());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-11-27 21:07:28 +00:00
|
|
|
uint32_t BlobFile::GetColumnFamilyId() const { return column_family_id_; }
|
2017-10-27 20:14:34 +00:00
|
|
|
|
2017-05-10 21:54:35 +00:00
|
|
|
std::string BlobFile::PathName() const {
|
|
|
|
return BlobFileName(path_to_dir_, file_number_);
|
|
|
|
}
|
|
|
|
|
2020-07-07 00:10:41 +00:00
|
|
|
std::shared_ptr<BlobLogReader> BlobFile::OpenRandomAccessReader(
|
2017-05-10 21:54:35 +00:00
|
|
|
Env* env, const DBOptions& db_options,
|
|
|
|
const EnvOptions& env_options) const {
|
2018-06-24 06:08:33 +00:00
|
|
|
constexpr size_t kReadaheadSize = 2 * 1024 * 1024;
|
|
|
|
std::unique_ptr<RandomAccessFile> sfile;
|
2018-06-21 15:34:24 +00:00
|
|
|
std::string path_name(PathName());
|
2018-06-24 06:08:33 +00:00
|
|
|
Status s = env->NewRandomAccessFile(path_name, &sfile, env_options);
|
2017-05-10 21:54:35 +00:00
|
|
|
if (!s.ok()) {
|
|
|
|
// report something here.
|
|
|
|
return nullptr;
|
|
|
|
}
|
2018-06-24 06:08:33 +00:00
|
|
|
sfile = NewReadaheadRandomAccessFile(std::move(sfile), kReadaheadSize);
|
2017-05-10 21:54:35 +00:00
|
|
|
|
2018-06-24 06:08:33 +00:00
|
|
|
std::unique_ptr<RandomAccessFileReader> sfile_reader;
|
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 22:47:08 +00:00
|
|
|
sfile_reader.reset(new RandomAccessFileReader(
|
|
|
|
NewLegacyRandomAccessFileWrapper(sfile), path_name));
|
2017-05-10 21:54:35 +00:00
|
|
|
|
2020-07-07 00:10:41 +00:00
|
|
|
std::shared_ptr<BlobLogReader> log_reader = std::make_shared<BlobLogReader>(
|
2017-11-28 19:42:28 +00:00
|
|
|
std::move(sfile_reader), db_options.env, db_options.statistics.get());
|
2017-05-10 21:54:35 +00:00
|
|
|
|
|
|
|
return log_reader;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string BlobFile::DumpState() const {
|
|
|
|
char str[1000];
|
2018-03-02 20:54:24 +00:00
|
|
|
snprintf(
|
|
|
|
str, sizeof(str),
|
|
|
|
"path: %s fn: %" PRIu64 " blob_count: %" PRIu64 " file_size: %" PRIu64
|
|
|
|
" closed: %d obsolete: %d expiration_range: (%" PRIu64 ", %" PRIu64
|
|
|
|
"), writer: %d reader: %d",
|
|
|
|
path_to_dir_.c_str(), file_number_, blob_count_.load(), file_size_.load(),
|
|
|
|
closed_.load(), obsolete_.load(), expiration_range_.first,
|
|
|
|
expiration_range_.second, (!!log_writer_), (!!ra_file_reader_));
|
2017-05-10 21:54:35 +00:00
|
|
|
return str;
|
|
|
|
}
|
|
|
|
|
2017-11-02 22:50:30 +00:00
|
|
|
void BlobFile::MarkObsolete(SequenceNumber sequence) {
|
2017-12-11 20:01:22 +00:00
|
|
|
assert(Immutable());
|
2017-11-02 22:50:30 +00:00
|
|
|
obsolete_sequence_ = sequence;
|
|
|
|
obsolete_.store(true);
|
|
|
|
}
|
|
|
|
|
2017-05-10 21:54:35 +00:00
|
|
|
bool BlobFile::NeedsFsync(bool hard, uint64_t bytes_per_sync) const {
|
|
|
|
assert(last_fsync_ <= file_size_);
|
|
|
|
return (hard) ? file_size_ > last_fsync_
|
|
|
|
: (file_size_ - last_fsync_) >= bytes_per_sync;
|
|
|
|
}
|
|
|
|
|
2019-11-19 00:28:04 +00:00
|
|
|
Status BlobFile::WriteFooterAndCloseLocked(SequenceNumber sequence) {
|
2017-05-10 21:54:35 +00:00
|
|
|
BlobLogFooter footer;
|
2017-10-27 20:14:34 +00:00
|
|
|
footer.blob_count = blob_count_;
|
|
|
|
if (HasTTL()) {
|
|
|
|
footer.expiration_range = expiration_range_;
|
|
|
|
}
|
2017-05-10 21:54:35 +00:00
|
|
|
|
|
|
|
// this will close the file and reset the Writable File Pointer.
|
|
|
|
Status s = log_writer_->AppendFooter(footer);
|
|
|
|
if (s.ok()) {
|
|
|
|
closed_ = true;
|
2019-11-19 00:28:04 +00:00
|
|
|
immutable_sequence_ = sequence;
|
2017-10-27 20:14:34 +00:00
|
|
|
file_size_ += BlobLogFooter::kSize;
|
2017-05-10 21:54:35 +00:00
|
|
|
}
|
|
|
|
// delete the sequential writer
|
|
|
|
log_writer_.reset();
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status BlobFile::ReadFooter(BlobLogFooter* bf) {
|
2017-10-27 20:14:34 +00:00
|
|
|
if (file_size_ < (BlobLogHeader::kSize + BlobLogFooter::kSize)) {
|
2017-05-10 21:54:35 +00:00
|
|
|
return Status::IOError("File does not have footer", PathName());
|
|
|
|
}
|
|
|
|
|
2017-10-27 20:14:34 +00:00
|
|
|
uint64_t footer_offset = file_size_ - BlobLogFooter::kSize;
|
2017-05-10 21:54:35 +00:00
|
|
|
// assume that ra_file_reader_ is valid before we enter this
|
|
|
|
assert(ra_file_reader_);
|
|
|
|
|
|
|
|
Slice result;
|
2020-03-06 22:02:09 +00:00
|
|
|
std::string buf;
|
Support direct IO in RandomAccessFileReader::MultiRead (#6446)
Summary:
By supporting direct IO in RandomAccessFileReader::MultiRead, the benefits of parallel IO (IO uring) and direct IO can be combined.
In direct IO mode, read requests are aligned and merged together before being issued to RandomAccessFile::MultiRead, so blocks in the original requests might share the same underlying buffer, the shared buffers are returned in `aligned_bufs`, which is a new parameter of the `MultiRead` API.
For example, suppose alignment requirement for direct IO is 4KB, one request is (offset: 1KB, len: 1KB), another request is (offset: 3KB, len: 1KB), then since they all belong to page (offset: 0, len: 4KB), `MultiRead` only reads the page with direct IO into a buffer on heap, and returns 2 Slices referencing regions in that same buffer. See `random_access_file_reader_test.cc` for more examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6446
Test Plan: Added a new test `random_access_file_reader_test.cc`.
Reviewed By: anand1976
Differential Revision: D20097518
Pulled By: cheng-chang
fbshipit-source-id: ca48a8faf9c3af146465c102ef6b266a363e78d1
2020-03-20 23:15:40 +00:00
|
|
|
AlignedBuf aligned_buf;
|
2020-03-06 22:02:09 +00:00
|
|
|
Status s;
|
|
|
|
if (ra_file_reader_->use_direct_io()) {
|
2020-04-30 21:48:51 +00:00
|
|
|
s = ra_file_reader_->Read(IOOptions(), footer_offset, BlobLogFooter::kSize,
|
|
|
|
&result, nullptr, &aligned_buf);
|
2020-03-06 22:02:09 +00:00
|
|
|
} else {
|
|
|
|
buf.reserve(BlobLogFooter::kSize + 10);
|
2020-04-30 21:48:51 +00:00
|
|
|
s = ra_file_reader_->Read(IOOptions(), footer_offset, BlobLogFooter::kSize,
|
|
|
|
&result, &buf[0], nullptr);
|
2020-03-06 22:02:09 +00:00
|
|
|
}
|
2017-05-10 21:54:35 +00:00
|
|
|
if (!s.ok()) return s;
|
2017-10-27 20:14:34 +00:00
|
|
|
if (result.size() != BlobLogFooter::kSize) {
|
2017-05-10 21:54:35 +00:00
|
|
|
// should not happen
|
|
|
|
return Status::IOError("EOF reached before footer");
|
|
|
|
}
|
|
|
|
|
2017-05-23 17:30:04 +00:00
|
|
|
s = bf->DecodeFrom(result);
|
2017-05-10 21:54:35 +00:00
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status BlobFile::SetFromFooterLocked(const BlobLogFooter& footer) {
|
|
|
|
// assume that file has been fully fsync'd
|
|
|
|
last_fsync_.store(file_size_);
|
2017-10-27 20:14:34 +00:00
|
|
|
blob_count_ = footer.blob_count;
|
|
|
|
expiration_range_ = footer.expiration_range;
|
2017-05-10 21:54:35 +00:00
|
|
|
closed_ = true;
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2017-12-11 20:01:22 +00:00
|
|
|
Status BlobFile::Fsync() {
|
|
|
|
Status s;
|
2017-05-10 21:54:35 +00:00
|
|
|
if (log_writer_.get()) {
|
2017-12-11 20:01:22 +00:00
|
|
|
s = log_writer_->Sync();
|
2017-05-10 21:54:35 +00:00
|
|
|
last_fsync_.store(file_size_.load());
|
|
|
|
}
|
2017-12-11 20:01:22 +00:00
|
|
|
return s;
|
2017-05-10 21:54:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void BlobFile::CloseRandomAccessLocked() {
|
|
|
|
ra_file_reader_.reset();
|
|
|
|
last_access_ = -1;
|
|
|
|
}
|
|
|
|
|
2018-09-20 23:50:07 +00:00
|
|
|
Status BlobFile::GetReader(Env* env, const EnvOptions& env_options,
|
|
|
|
std::shared_ptr<RandomAccessFileReader>* reader,
|
|
|
|
bool* fresh_open) {
|
|
|
|
assert(reader != nullptr);
|
|
|
|
assert(fresh_open != nullptr);
|
2017-05-10 21:54:35 +00:00
|
|
|
*fresh_open = false;
|
2017-10-13 21:36:36 +00:00
|
|
|
int64_t current_time = 0;
|
|
|
|
env->GetCurrentTime(¤t_time);
|
|
|
|
last_access_.store(current_time);
|
2018-09-20 23:50:07 +00:00
|
|
|
Status s;
|
2017-10-13 21:36:36 +00:00
|
|
|
|
2017-05-10 21:54:35 +00:00
|
|
|
{
|
|
|
|
ReadLock lockbfile_r(&mutex_);
|
2018-09-20 23:50:07 +00:00
|
|
|
if (ra_file_reader_) {
|
|
|
|
*reader = ra_file_reader_;
|
|
|
|
return s;
|
|
|
|
}
|
2017-05-10 21:54:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
WriteLock lockbfile_w(&mutex_);
|
2018-09-20 23:50:07 +00:00
|
|
|
// Double check.
|
|
|
|
if (ra_file_reader_) {
|
|
|
|
*reader = ra_file_reader_;
|
|
|
|
return s;
|
|
|
|
}
|
2017-05-10 21:54:35 +00:00
|
|
|
|
|
|
|
std::unique_ptr<RandomAccessFile> rfile;
|
2018-09-20 23:50:07 +00:00
|
|
|
s = env->NewRandomAccessFile(PathName(), &rfile, env_options);
|
2017-05-10 21:54:35 +00:00
|
|
|
if (!s.ok()) {
|
2017-12-11 20:01:22 +00:00
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2017-07-13 17:03:39 +00:00
|
|
|
"Failed to open blob file for random-read: %s status: '%s'"
|
|
|
|
" exists: '%s'",
|
|
|
|
PathName().c_str(), s.ToString().c_str(),
|
|
|
|
env->FileExists(PathName()).ToString().c_str());
|
2018-09-20 23:50:07 +00:00
|
|
|
return s;
|
2017-05-10 21:54:35 +00:00
|
|
|
}
|
|
|
|
|
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 22:47:08 +00:00
|
|
|
ra_file_reader_ = std::make_shared<RandomAccessFileReader>(
|
|
|
|
NewLegacyRandomAccessFileWrapper(rfile), PathName());
|
2018-09-20 23:50:07 +00:00
|
|
|
*reader = ra_file_reader_;
|
2017-05-10 21:54:35 +00:00
|
|
|
*fresh_open = true;
|
2018-09-20 23:50:07 +00:00
|
|
|
return s;
|
2017-05-10 21:54:35 +00:00
|
|
|
}
|
|
|
|
|
2017-12-11 20:01:22 +00:00
|
|
|
Status BlobFile::ReadMetadata(Env* env, const EnvOptions& env_options) {
|
|
|
|
assert(Immutable());
|
|
|
|
// Get file size.
|
|
|
|
uint64_t file_size = 0;
|
|
|
|
Status s = env->GetFileSize(PathName(), &file_size);
|
|
|
|
if (s.ok()) {
|
|
|
|
file_size_ = file_size;
|
|
|
|
} else {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Failed to get size of blob file %" PRIu64
|
2017-12-11 20:01:22 +00:00
|
|
|
", status: %s",
|
|
|
|
file_number_, s.ToString().c_str());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
if (file_size < BlobLogHeader::kSize) {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Incomplete blob file blob file %" PRIu64
|
2017-12-11 20:01:22 +00:00
|
|
|
", size: %" PRIu64,
|
|
|
|
file_number_, file_size);
|
|
|
|
return Status::Corruption("Incomplete blob file header.");
|
|
|
|
}
|
|
|
|
|
|
|
|
// Create file reader.
|
|
|
|
std::unique_ptr<RandomAccessFile> file;
|
|
|
|
s = env->NewRandomAccessFile(PathName(), &file, env_options);
|
|
|
|
if (!s.ok()) {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Failed to open blob file %" PRIu64 ", status: %s",
|
2017-12-11 20:01:22 +00:00
|
|
|
file_number_, s.ToString().c_str());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
std::unique_ptr<RandomAccessFileReader> file_reader(
|
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 22:47:08 +00:00
|
|
|
new RandomAccessFileReader(NewLegacyRandomAccessFileWrapper(file),
|
|
|
|
PathName()));
|
2017-12-11 20:01:22 +00:00
|
|
|
|
|
|
|
// Read file header.
|
2020-03-06 22:02:09 +00:00
|
|
|
std::string header_buf;
|
Support direct IO in RandomAccessFileReader::MultiRead (#6446)
Summary:
By supporting direct IO in RandomAccessFileReader::MultiRead, the benefits of parallel IO (IO uring) and direct IO can be combined.
In direct IO mode, read requests are aligned and merged together before being issued to RandomAccessFile::MultiRead, so blocks in the original requests might share the same underlying buffer, the shared buffers are returned in `aligned_bufs`, which is a new parameter of the `MultiRead` API.
For example, suppose alignment requirement for direct IO is 4KB, one request is (offset: 1KB, len: 1KB), another request is (offset: 3KB, len: 1KB), then since they all belong to page (offset: 0, len: 4KB), `MultiRead` only reads the page with direct IO into a buffer on heap, and returns 2 Slices referencing regions in that same buffer. See `random_access_file_reader_test.cc` for more examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6446
Test Plan: Added a new test `random_access_file_reader_test.cc`.
Reviewed By: anand1976
Differential Revision: D20097518
Pulled By: cheng-chang
fbshipit-source-id: ca48a8faf9c3af146465c102ef6b266a363e78d1
2020-03-20 23:15:40 +00:00
|
|
|
AlignedBuf aligned_buf;
|
2017-12-11 20:01:22 +00:00
|
|
|
Slice header_slice;
|
2020-03-06 22:02:09 +00:00
|
|
|
if (file_reader->use_direct_io()) {
|
2020-04-30 21:48:51 +00:00
|
|
|
s = file_reader->Read(IOOptions(), 0, BlobLogHeader::kSize, &header_slice,
|
|
|
|
nullptr, &aligned_buf);
|
2020-03-06 22:02:09 +00:00
|
|
|
} else {
|
|
|
|
header_buf.reserve(BlobLogHeader::kSize);
|
2020-04-30 21:48:51 +00:00
|
|
|
s = file_reader->Read(IOOptions(), 0, BlobLogHeader::kSize, &header_slice,
|
2020-03-06 22:02:09 +00:00
|
|
|
&header_buf[0], nullptr);
|
|
|
|
}
|
2017-12-11 20:01:22 +00:00
|
|
|
if (!s.ok()) {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Failed to read header of blob file %" PRIu64
|
2017-12-11 20:01:22 +00:00
|
|
|
", status: %s",
|
|
|
|
file_number_, s.ToString().c_str());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
BlobLogHeader header;
|
|
|
|
s = header.DecodeFrom(header_slice);
|
|
|
|
if (!s.ok()) {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Failed to decode header of blob file %" PRIu64
|
2017-12-11 20:01:22 +00:00
|
|
|
", status: %s",
|
|
|
|
file_number_, s.ToString().c_str());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
column_family_id_ = header.column_family_id;
|
|
|
|
compression_ = header.compression;
|
|
|
|
has_ttl_ = header.has_ttl;
|
|
|
|
if (has_ttl_) {
|
|
|
|
expiration_range_ = header.expiration_range;
|
|
|
|
}
|
|
|
|
header_valid_ = true;
|
|
|
|
|
|
|
|
// Read file footer.
|
|
|
|
if (file_size_ < BlobLogHeader::kSize + BlobLogFooter::kSize) {
|
|
|
|
// OK not to have footer.
|
|
|
|
assert(!footer_valid_);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
2020-03-06 22:02:09 +00:00
|
|
|
std::string footer_buf;
|
2017-12-11 20:01:22 +00:00
|
|
|
Slice footer_slice;
|
2020-03-06 22:02:09 +00:00
|
|
|
if (file_reader->use_direct_io()) {
|
2020-04-30 21:48:51 +00:00
|
|
|
s = file_reader->Read(IOOptions(), file_size - BlobLogFooter::kSize,
|
Support direct IO in RandomAccessFileReader::MultiRead (#6446)
Summary:
By supporting direct IO in RandomAccessFileReader::MultiRead, the benefits of parallel IO (IO uring) and direct IO can be combined.
In direct IO mode, read requests are aligned and merged together before being issued to RandomAccessFile::MultiRead, so blocks in the original requests might share the same underlying buffer, the shared buffers are returned in `aligned_bufs`, which is a new parameter of the `MultiRead` API.
For example, suppose alignment requirement for direct IO is 4KB, one request is (offset: 1KB, len: 1KB), another request is (offset: 3KB, len: 1KB), then since they all belong to page (offset: 0, len: 4KB), `MultiRead` only reads the page with direct IO into a buffer on heap, and returns 2 Slices referencing regions in that same buffer. See `random_access_file_reader_test.cc` for more examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6446
Test Plan: Added a new test `random_access_file_reader_test.cc`.
Reviewed By: anand1976
Differential Revision: D20097518
Pulled By: cheng-chang
fbshipit-source-id: ca48a8faf9c3af146465c102ef6b266a363e78d1
2020-03-20 23:15:40 +00:00
|
|
|
BlobLogFooter::kSize, &footer_slice, nullptr,
|
|
|
|
&aligned_buf);
|
2020-03-06 22:02:09 +00:00
|
|
|
} else {
|
|
|
|
footer_buf.reserve(BlobLogFooter::kSize);
|
2020-04-30 21:48:51 +00:00
|
|
|
s = file_reader->Read(IOOptions(), file_size - BlobLogFooter::kSize,
|
Support direct IO in RandomAccessFileReader::MultiRead (#6446)
Summary:
By supporting direct IO in RandomAccessFileReader::MultiRead, the benefits of parallel IO (IO uring) and direct IO can be combined.
In direct IO mode, read requests are aligned and merged together before being issued to RandomAccessFile::MultiRead, so blocks in the original requests might share the same underlying buffer, the shared buffers are returned in `aligned_bufs`, which is a new parameter of the `MultiRead` API.
For example, suppose alignment requirement for direct IO is 4KB, one request is (offset: 1KB, len: 1KB), another request is (offset: 3KB, len: 1KB), then since they all belong to page (offset: 0, len: 4KB), `MultiRead` only reads the page with direct IO into a buffer on heap, and returns 2 Slices referencing regions in that same buffer. See `random_access_file_reader_test.cc` for more examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6446
Test Plan: Added a new test `random_access_file_reader_test.cc`.
Reviewed By: anand1976
Differential Revision: D20097518
Pulled By: cheng-chang
fbshipit-source-id: ca48a8faf9c3af146465c102ef6b266a363e78d1
2020-03-20 23:15:40 +00:00
|
|
|
BlobLogFooter::kSize, &footer_slice, &footer_buf[0],
|
|
|
|
nullptr);
|
2020-03-06 22:02:09 +00:00
|
|
|
}
|
2017-12-11 20:01:22 +00:00
|
|
|
if (!s.ok()) {
|
|
|
|
ROCKS_LOG_ERROR(info_log_,
|
2019-04-04 19:05:42 +00:00
|
|
|
"Failed to read footer of blob file %" PRIu64
|
2017-12-11 20:01:22 +00:00
|
|
|
", status: %s",
|
|
|
|
file_number_, s.ToString().c_str());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
BlobLogFooter footer;
|
|
|
|
s = footer.DecodeFrom(footer_slice);
|
|
|
|
if (!s.ok()) {
|
|
|
|
// OK not to have footer.
|
|
|
|
assert(!footer_valid_);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
blob_count_ = footer.blob_count;
|
|
|
|
if (has_ttl_) {
|
|
|
|
assert(header.expiration_range.first <= footer.expiration_range.first);
|
|
|
|
assert(header.expiration_range.second >= footer.expiration_range.second);
|
|
|
|
expiration_range_ = footer.expiration_range;
|
|
|
|
}
|
|
|
|
footer_valid_ = true;
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
2017-05-10 21:54:35 +00:00
|
|
|
} // namespace blob_db
|
2020-02-20 20:07:53 +00:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|
2017-05-10 21:54:35 +00:00
|
|
|
#endif // ROCKSDB_LITE
|