2016-02-09 23:12:00 +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).
|
2015-10-14 08:14:53 +00:00
|
|
|
//
|
|
|
|
// 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.
|
2015-10-23 14:36:22 +00:00
|
|
|
#pragma once
|
2016-09-13 19:08:22 +00:00
|
|
|
#include <errno.h>
|
2019-12-08 04:54:27 +00:00
|
|
|
#if defined(ROCKSDB_IOURING_PRESENT)
|
|
|
|
#include <liburing.h>
|
|
|
|
#include <sys/uio.h>
|
|
|
|
#endif
|
2015-10-14 08:14:53 +00:00
|
|
|
#include <unistd.h>
|
2022-03-11 02:28:31 +00:00
|
|
|
|
2016-04-21 17:37:27 +00:00
|
|
|
#include <atomic>
|
2020-03-12 01:36:43 +00:00
|
|
|
#include <functional>
|
|
|
|
#include <map>
|
2016-09-13 19:08:22 +00:00
|
|
|
#include <string>
|
2022-03-11 02:28:31 +00:00
|
|
|
|
2020-03-12 01:36:43 +00:00
|
|
|
#include "port/port.h"
|
2015-10-14 08:14:53 +00:00
|
|
|
#include "rocksdb/env.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 "rocksdb/file_system.h"
|
|
|
|
#include "rocksdb/io_status.h"
|
2022-03-11 02:28:31 +00:00
|
|
|
#include "test_util/sync_point.h"
|
2020-03-12 01:36:43 +00:00
|
|
|
#include "util/mutexlock.h"
|
|
|
|
#include "util/thread_local.h"
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
// For non linux platform, the following macros are used only as place
|
|
|
|
// holder.
|
2017-04-22 03:41:37 +00:00
|
|
|
#if !(defined OS_LINUX) && !(defined CYGWIN) && !(defined OS_AIX)
|
2015-10-14 08:14:53 +00:00
|
|
|
#define POSIX_FADV_NORMAL 0 /* [MC1] no further special treatment */
|
|
|
|
#define POSIX_FADV_RANDOM 1 /* [MC1] expect random page refs */
|
|
|
|
#define POSIX_FADV_SEQUENTIAL 2 /* [MC1] expect sequential page refs */
|
|
|
|
#define POSIX_FADV_WILLNEED 3 /* [MC1] will need these pages */
|
2020-02-28 22:10:51 +00:00
|
|
|
#define POSIX_FADV_DONTNEED 4 /* [MC1] don't need these pages */
|
2015-10-14 08:14:53 +00:00
|
|
|
#endif
|
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
2020-03-12 01:36:43 +00:00
|
|
|
std::string IOErrorMsg(const std::string& context,
|
|
|
|
const std::string& file_name);
|
2017-06-26 19:42:21 +00:00
|
|
|
// file_name can be left empty if it is not unkown.
|
2020-03-12 01:36:43 +00:00
|
|
|
IOStatus IOError(const std::string& context, const std::string& file_name,
|
|
|
|
int err_number);
|
2015-10-14 08:14:53 +00:00
|
|
|
|
2016-04-21 17:37:27 +00:00
|
|
|
class PosixHelper {
|
|
|
|
public:
|
|
|
|
static size_t GetUniqueIdFromFile(int fd, char* id, size_t max_size);
|
2020-03-12 01:36:43 +00:00
|
|
|
static size_t GetLogicalBlockSizeOfFd(int fd);
|
|
|
|
static Status GetLogicalBlockSizeOfDirectory(const std::string& directory,
|
|
|
|
size_t* size);
|
|
|
|
};
|
|
|
|
|
2022-07-06 18:42:59 +00:00
|
|
|
/*
|
|
|
|
* DirectIOHelper
|
|
|
|
*/
|
|
|
|
inline bool IsSectorAligned(const size_t off, size_t sector_size) {
|
|
|
|
assert((sector_size & (sector_size - 1)) == 0);
|
|
|
|
return (off & (sector_size - 1)) == 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
|
|
inline bool IsSectorAligned(const void* ptr, size_t sector_size) {
|
|
|
|
return uintptr_t(ptr) % sector_size == 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2022-03-11 02:28:31 +00:00
|
|
|
#if defined(ROCKSDB_IOURING_PRESENT)
|
|
|
|
struct Posix_IOHandle {
|
2022-07-06 18:42:59 +00:00
|
|
|
Posix_IOHandle(struct io_uring* _iu,
|
|
|
|
std::function<void(const FSReadRequest&, void*)> _cb,
|
|
|
|
void* _cb_arg, uint64_t _offset, size_t _len, char* _scratch,
|
|
|
|
bool _use_direct_io, size_t _alignment)
|
|
|
|
: iu(_iu),
|
|
|
|
cb(_cb),
|
|
|
|
cb_arg(_cb_arg),
|
|
|
|
offset(_offset),
|
|
|
|
len(_len),
|
|
|
|
scratch(_scratch),
|
|
|
|
use_direct_io(_use_direct_io),
|
|
|
|
alignment(_alignment),
|
|
|
|
is_finished(false),
|
|
|
|
req_count(0) {}
|
|
|
|
|
2022-03-11 02:28:31 +00:00
|
|
|
struct iovec iov;
|
|
|
|
struct io_uring* iu;
|
|
|
|
std::function<void(const FSReadRequest&, void*)> cb;
|
|
|
|
void* cb_arg;
|
|
|
|
uint64_t offset;
|
|
|
|
size_t len;
|
|
|
|
char* scratch;
|
2022-07-06 18:42:59 +00:00
|
|
|
bool use_direct_io;
|
|
|
|
size_t alignment;
|
|
|
|
bool is_finished;
|
2022-06-14 01:07:24 +00:00
|
|
|
// req_count is used by AbortIO API to keep track of number of requests.
|
2022-07-06 18:42:59 +00:00
|
|
|
uint32_t req_count;
|
2022-03-11 02:28:31 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
inline void UpdateResult(struct io_uring_cqe* cqe, const std::string& file_name,
|
|
|
|
size_t len, size_t iov_len, bool async_read,
|
2022-07-07 18:35:15 +00:00
|
|
|
bool use_direct_io, size_t alignment,
|
2022-04-04 22:35:43 +00:00
|
|
|
size_t& finished_len, FSReadRequest* req,
|
2022-07-06 18:42:59 +00:00
|
|
|
size_t& bytes_read, bool& read_again) {
|
|
|
|
read_again = false;
|
2022-03-11 02:28:31 +00:00
|
|
|
if (cqe->res < 0) {
|
|
|
|
req->result = Slice(req->scratch, 0);
|
|
|
|
req->status = IOError("Req failed", file_name, cqe->res);
|
|
|
|
} else {
|
2022-04-04 22:35:43 +00:00
|
|
|
bytes_read = static_cast<size_t>(cqe->res);
|
2022-03-11 02:28:31 +00:00
|
|
|
TEST_SYNC_POINT_CALLBACK("UpdateResults::io_uring_result", &bytes_read);
|
|
|
|
if (bytes_read == iov_len) {
|
|
|
|
req->result = Slice(req->scratch, req->len);
|
|
|
|
req->status = IOStatus::OK();
|
|
|
|
} else if (bytes_read == 0) {
|
2022-07-06 18:42:59 +00:00
|
|
|
/// cqe->res == 0 can means EOF, or can mean partial results. See
|
|
|
|
// comment
|
|
|
|
// https://github.com/facebook/rocksdb/pull/6441#issuecomment-589843435
|
|
|
|
// Fall back to pread in this case.
|
|
|
|
if (use_direct_io && !IsSectorAligned(finished_len, alignment)) {
|
|
|
|
// Bytes reads don't fill sectors. Should only happen at the end
|
|
|
|
// of the file.
|
|
|
|
req->result = Slice(req->scratch, finished_len);
|
2022-03-11 02:28:31 +00:00
|
|
|
req->status = IOStatus::OK();
|
2022-07-06 18:42:59 +00:00
|
|
|
} else {
|
|
|
|
if (async_read) {
|
|
|
|
// No bytes read. It can means EOF. In case of partial results, it's
|
|
|
|
// caller responsibility to call read/readasync again.
|
|
|
|
req->result = Slice(req->scratch, 0);
|
|
|
|
req->status = IOStatus::OK();
|
|
|
|
} else {
|
|
|
|
read_again = true;
|
|
|
|
}
|
2022-03-11 02:28:31 +00:00
|
|
|
}
|
|
|
|
} else if (bytes_read < iov_len) {
|
|
|
|
assert(bytes_read > 0);
|
|
|
|
if (async_read) {
|
|
|
|
req->result = Slice(req->scratch, bytes_read);
|
|
|
|
req->status = IOStatus::OK();
|
|
|
|
} else {
|
|
|
|
assert(bytes_read + finished_len < len);
|
|
|
|
finished_len += bytes_read;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
req->result = Slice(req->scratch, 0);
|
|
|
|
req->status = IOError("Req returned more bytes than requested", file_name,
|
|
|
|
cqe->res);
|
|
|
|
}
|
|
|
|
}
|
2022-03-13 06:12:18 +00:00
|
|
|
#ifdef NDEBUG
|
|
|
|
(void)len;
|
|
|
|
#endif
|
2022-03-11 02:28:31 +00:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2020-03-12 01:36:43 +00:00
|
|
|
#ifdef OS_LINUX
|
|
|
|
// Files under a specific directory have the same logical block size.
|
|
|
|
// This class caches the logical block size for the specified directories to
|
|
|
|
// save the CPU cost of computing the size.
|
|
|
|
// Safe for concurrent access from multiple threads without any external
|
|
|
|
// synchronization.
|
|
|
|
class LogicalBlockSizeCache {
|
|
|
|
public:
|
|
|
|
LogicalBlockSizeCache(
|
|
|
|
std::function<size_t(int)> get_logical_block_size_of_fd =
|
|
|
|
PosixHelper::GetLogicalBlockSizeOfFd,
|
|
|
|
std::function<Status(const std::string&, size_t*)>
|
|
|
|
get_logical_block_size_of_directory =
|
|
|
|
PosixHelper::GetLogicalBlockSizeOfDirectory)
|
|
|
|
: get_logical_block_size_of_fd_(get_logical_block_size_of_fd),
|
|
|
|
get_logical_block_size_of_directory_(
|
|
|
|
get_logical_block_size_of_directory) {}
|
|
|
|
|
|
|
|
// Takes the following actions:
|
|
|
|
// 1. Increases reference count of the directories;
|
|
|
|
// 2. If the directory's logical block size is not cached,
|
|
|
|
// compute the buffer size and cache the result.
|
|
|
|
Status RefAndCacheLogicalBlockSize(
|
|
|
|
const std::vector<std::string>& directories);
|
|
|
|
|
|
|
|
// Takes the following actions:
|
|
|
|
// 1. Decreases reference count of the directories;
|
|
|
|
// 2. If the reference count of a directory reaches 0, remove the directory
|
|
|
|
// from the cache.
|
|
|
|
void UnrefAndTryRemoveCachedLogicalBlockSize(
|
|
|
|
const std::vector<std::string>& directories);
|
|
|
|
|
|
|
|
// Returns the logical block size for the file.
|
|
|
|
//
|
|
|
|
// If the file is under a cached directory, return the cached size.
|
|
|
|
// Otherwise, the size is computed.
|
|
|
|
size_t GetLogicalBlockSize(const std::string& fname, int fd);
|
|
|
|
|
|
|
|
int GetRefCount(const std::string& dir) {
|
|
|
|
ReadLock lock(&cache_mutex_);
|
|
|
|
auto it = cache_.find(dir);
|
|
|
|
if (it == cache_.end()) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return it->second.ref;
|
|
|
|
}
|
|
|
|
|
|
|
|
size_t Size() const { return cache_.size(); }
|
|
|
|
|
|
|
|
bool Contains(const std::string& dir) {
|
|
|
|
ReadLock lock(&cache_mutex_);
|
|
|
|
return cache_.find(dir) != cache_.end();
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
struct CacheValue {
|
|
|
|
CacheValue() : size(0), ref(0) {}
|
|
|
|
|
|
|
|
// Logical block size of the directory.
|
|
|
|
size_t size;
|
|
|
|
// Reference count of the directory.
|
|
|
|
int ref;
|
|
|
|
};
|
|
|
|
|
|
|
|
std::function<size_t(int)> get_logical_block_size_of_fd_;
|
|
|
|
std::function<Status(const std::string&, size_t*)>
|
|
|
|
get_logical_block_size_of_directory_;
|
|
|
|
|
|
|
|
std::map<std::string, CacheValue> cache_;
|
|
|
|
port::RWMutex cache_mutex_;
|
2016-04-21 17:37:27 +00:00
|
|
|
};
|
2020-03-12 01:36:43 +00:00
|
|
|
#endif
|
2016-04-21 17:37:27 +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
|
|
|
class PosixSequentialFile : public FSSequentialFile {
|
2015-10-14 08:14:53 +00:00
|
|
|
private:
|
|
|
|
std::string filename_;
|
|
|
|
FILE* file_;
|
|
|
|
int fd_;
|
2016-12-22 20:51:29 +00:00
|
|
|
bool use_direct_io_;
|
2017-02-23 19:17:49 +00:00
|
|
|
size_t logical_sector_size_;
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
public:
|
2017-01-12 00:42:07 +00:00
|
|
|
PosixSequentialFile(const std::string& fname, FILE* file, int fd,
|
2022-05-24 17:28:57 +00:00
|
|
|
size_t logical_block_size, const EnvOptions& options);
|
2015-10-14 08:14:53 +00:00
|
|
|
virtual ~PosixSequentialFile();
|
|
|
|
|
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
|
|
|
virtual IOStatus Read(size_t n, const IOOptions& opts, Slice* result,
|
|
|
|
char* scratch, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus PositionedRead(uint64_t offset, size_t n,
|
|
|
|
const IOOptions& opts, Slice* result,
|
|
|
|
char* scratch, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Skip(uint64_t n) override;
|
|
|
|
virtual IOStatus InvalidateCache(size_t offset, size_t length) override;
|
2017-01-13 20:01:08 +00:00
|
|
|
virtual bool use_direct_io() const override { return use_direct_io_; }
|
2017-02-23 19:17:49 +00:00
|
|
|
virtual size_t GetRequiredBufferAlignment() const override {
|
|
|
|
return logical_sector_size_;
|
|
|
|
}
|
2016-04-21 17:37:27 +00:00
|
|
|
};
|
|
|
|
|
2019-12-08 04:54:27 +00:00
|
|
|
#if defined(ROCKSDB_IOURING_PRESENT)
|
|
|
|
// io_uring instance queue depth
|
|
|
|
const unsigned int kIoUringDepth = 256;
|
|
|
|
|
|
|
|
inline void DeleteIOUring(void* p) {
|
|
|
|
struct io_uring* iu = static_cast<struct io_uring*>(p);
|
|
|
|
delete iu;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline struct io_uring* CreateIOUring() {
|
|
|
|
struct io_uring* new_io_uring = new struct io_uring;
|
|
|
|
int ret = io_uring_queue_init(kIoUringDepth, new_io_uring, 0);
|
|
|
|
if (ret) {
|
|
|
|
delete new_io_uring;
|
|
|
|
new_io_uring = nullptr;
|
|
|
|
}
|
|
|
|
return new_io_uring;
|
|
|
|
}
|
2019-12-09 07:49:32 +00:00
|
|
|
#endif // defined(ROCKSDB_IOURING_PRESENT)
|
2019-12-08 04:54:27 +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
|
|
|
class PosixRandomAccessFile : public FSRandomAccessFile {
|
2016-04-21 17:37:27 +00:00
|
|
|
protected:
|
2015-10-14 08:14:53 +00:00
|
|
|
std::string filename_;
|
|
|
|
int fd_;
|
2016-12-22 20:51:29 +00:00
|
|
|
bool use_direct_io_;
|
2017-02-23 19:17:49 +00:00
|
|
|
size_t logical_sector_size_;
|
2019-12-08 04:54:27 +00:00
|
|
|
#if defined(ROCKSDB_IOURING_PRESENT)
|
|
|
|
ThreadLocalPtr* thread_local_io_urings_;
|
|
|
|
#endif
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
public:
|
|
|
|
PosixRandomAccessFile(const std::string& fname, int fd,
|
2022-10-25 00:54:14 +00:00
|
|
|
size_t logical_block_size, const EnvOptions& options
|
2019-12-08 04:54:27 +00:00
|
|
|
#if defined(ROCKSDB_IOURING_PRESENT)
|
|
|
|
,
|
|
|
|
ThreadLocalPtr* thread_local_io_urings
|
|
|
|
#endif
|
|
|
|
);
|
2015-10-14 08:14:53 +00:00
|
|
|
virtual ~PosixRandomAccessFile();
|
|
|
|
|
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
|
|
|
virtual IOStatus Read(uint64_t offset, size_t n, const IOOptions& opts,
|
|
|
|
Slice* result, char* scratch,
|
|
|
|
IODebugContext* dbg) const override;
|
2017-04-26 21:21:04 +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
|
|
|
virtual IOStatus MultiRead(FSReadRequest* reqs, size_t num_reqs,
|
|
|
|
const IOOptions& options,
|
|
|
|
IODebugContext* dbg) override;
|
2019-12-08 04:54:27 +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
|
|
|
virtual IOStatus Prefetch(uint64_t offset, size_t n, const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2017-04-26 21:21:04 +00:00
|
|
|
|
2017-04-22 03:41:37 +00:00
|
|
|
#if defined(OS_LINUX) || defined(OS_MACOSX) || defined(OS_AIX)
|
2015-10-14 08:14:53 +00:00
|
|
|
virtual size_t GetUniqueId(char* id, size_t max_size) const override;
|
|
|
|
#endif
|
|
|
|
virtual void Hint(AccessPattern pattern) override;
|
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
|
|
|
virtual IOStatus InvalidateCache(size_t offset, size_t length) override;
|
2017-01-13 20:01:08 +00:00
|
|
|
virtual bool use_direct_io() const override { return use_direct_io_; }
|
2017-02-23 19:17:49 +00:00
|
|
|
virtual size_t GetRequiredBufferAlignment() const override {
|
|
|
|
return logical_sector_size_;
|
|
|
|
}
|
2022-03-11 02:28:31 +00:00
|
|
|
// EXPERIMENTAL
|
|
|
|
virtual IOStatus ReadAsync(
|
|
|
|
FSReadRequest& req, const IOOptions& opts,
|
|
|
|
std::function<void(const FSReadRequest&, void*)> cb, void* cb_arg,
|
|
|
|
void** io_handle, IOHandleDeleter* del_fn, IODebugContext* dbg) override;
|
2016-04-21 17:37:27 +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
|
|
|
class PosixWritableFile : public FSWritableFile {
|
2016-04-21 17:37:27 +00:00
|
|
|
protected:
|
2015-10-14 08:14:53 +00:00
|
|
|
const std::string filename_;
|
2017-01-13 20:01:08 +00:00
|
|
|
const bool use_direct_io_;
|
2015-10-14 08:14:53 +00:00
|
|
|
int fd_;
|
|
|
|
uint64_t filesize_;
|
2017-02-23 19:17:49 +00:00
|
|
|
size_t logical_sector_size_;
|
2015-10-28 18:55:20 +00:00
|
|
|
#ifdef ROCKSDB_FALLOCATE_PRESENT
|
2015-10-14 08:14:53 +00:00
|
|
|
bool allow_fallocate_;
|
|
|
|
bool fallocate_with_keep_size_;
|
2015-10-28 18:55:20 +00:00
|
|
|
#endif
|
Optionally wait on bytes_per_sync to smooth I/O (#5183)
Summary:
The existing implementation does not guarantee bytes reach disk every `bytes_per_sync` when writing SST files, or every `wal_bytes_per_sync` when writing WALs. This can cause confusing behavior for users who enable this feature to avoid large syncs during flush and compaction, but then end up hitting them anyways.
My understanding of the existing behavior is we used `sync_file_range` with `SYNC_FILE_RANGE_WRITE` to submit ranges for async writeback, such that we could continue processing the next range of bytes while that I/O is happening. I believe we can preserve that benefit while also limiting how far the processing can get ahead of the I/O, which prevents huge syncs from happening when the file finishes.
Consider this `sync_file_range` usage: `sync_file_range(fd_, 0, static_cast<off_t>(offset + nbytes), SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE)`. Expanding the range to start at 0 and adding the `SYNC_FILE_RANGE_WAIT_BEFORE` flag causes any pending writeback (like from a previous call to `sync_file_range`) to finish before it proceeds to submit the latest `nbytes` for writeback. The latest `nbytes` are still written back asynchronously, unless processing exceeds I/O speed, in which case the following `sync_file_range` will need to wait on it.
There is a second change in this PR to use `fdatasync` when `sync_file_range` is unavailable (determined statically) or has some known problem with the underlying filesystem (determined dynamically).
The above two changes only apply when the user enables a new option, `strict_bytes_per_sync`.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5183
Differential Revision: D14953553
Pulled By: siying
fbshipit-source-id: 445c3862e019fb7b470f9c7f314fc231b62706e9
2019-04-22 18:48:45 +00:00
|
|
|
#ifdef ROCKSDB_RANGESYNC_PRESENT
|
|
|
|
// Even if the syscall is present, the filesystem may still not properly
|
|
|
|
// support it, so we need to do a dynamic check too.
|
|
|
|
bool sync_file_range_supported_;
|
|
|
|
#endif // ROCKSDB_RANGESYNC_PRESENT
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
public:
|
2016-04-21 17:37:27 +00:00
|
|
|
explicit PosixWritableFile(const std::string& fname, int fd,
|
2020-03-12 01:36:43 +00:00
|
|
|
size_t logical_block_size,
|
2016-04-21 17:37:27 +00:00
|
|
|
const EnvOptions& options);
|
|
|
|
virtual ~PosixWritableFile();
|
2015-10-14 08:14:53 +00:00
|
|
|
|
2017-02-22 18:00:25 +00:00
|
|
|
// Need to implement this so the file is truncated correctly
|
|
|
|
// with direct I/O
|
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
|
|
|
virtual IOStatus Truncate(uint64_t size, const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Close(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Append(const Slice& data, const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2020-09-24 02:00:30 +00:00
|
|
|
virtual IOStatus Append(const Slice& data, const IOOptions& opts,
|
|
|
|
const DataVerificationInfo& /* verification_info */,
|
|
|
|
IODebugContext* dbg) override {
|
|
|
|
return Append(data, opts, dbg);
|
|
|
|
}
|
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
|
|
|
virtual IOStatus PositionedAppend(const Slice& data, uint64_t offset,
|
|
|
|
const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2020-09-24 02:00:30 +00:00
|
|
|
virtual IOStatus PositionedAppend(
|
|
|
|
const Slice& data, uint64_t offset, const IOOptions& opts,
|
|
|
|
const DataVerificationInfo& /* verification_info */,
|
|
|
|
IODebugContext* dbg) override {
|
|
|
|
return PositionedAppend(data, offset, opts, dbg);
|
|
|
|
}
|
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
|
|
|
virtual IOStatus Flush(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Sync(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Fsync(const IOOptions& opts, IODebugContext* dbg) override;
|
2015-10-14 08:14:53 +00:00
|
|
|
virtual bool IsSyncThreadSafe() const override;
|
2017-01-13 20:01:08 +00:00
|
|
|
virtual bool use_direct_io() const override { return use_direct_io_; }
|
2017-11-10 17:25:26 +00:00
|
|
|
virtual void SetWriteLifeTimeHint(Env::WriteLifeTimeHint hint) override;
|
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
|
|
|
virtual uint64_t GetFileSize(const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus InvalidateCache(size_t offset, size_t length) override;
|
2016-12-22 20:51:29 +00:00
|
|
|
virtual size_t GetRequiredBufferAlignment() const override {
|
2017-02-23 19:17:49 +00:00
|
|
|
return logical_sector_size_;
|
2016-12-22 20:51:29 +00:00
|
|
|
}
|
2015-10-28 18:55:20 +00:00
|
|
|
#ifdef ROCKSDB_FALLOCATE_PRESENT
|
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
|
|
|
virtual IOStatus Allocate(uint64_t offset, uint64_t len,
|
|
|
|
const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2017-02-01 18:19:47 +00:00
|
|
|
#endif
|
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
|
|
|
virtual IOStatus RangeSync(uint64_t offset, uint64_t nbytes,
|
|
|
|
const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2017-04-22 03:41:37 +00:00
|
|
|
#ifdef OS_LINUX
|
2015-10-14 08:14:53 +00:00
|
|
|
virtual size_t GetUniqueId(char* id, size_t max_size) const override;
|
2017-02-02 18:32:40 +00:00
|
|
|
#endif
|
2015-10-14 08:14:53 +00:00
|
|
|
};
|
|
|
|
|
2016-12-22 20:51:29 +00:00
|
|
|
// mmap() based random-access
|
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
|
|
|
class PosixMmapReadableFile : public FSRandomAccessFile {
|
2015-10-14 08:14:53 +00:00
|
|
|
private:
|
|
|
|
int fd_;
|
|
|
|
std::string filename_;
|
|
|
|
void* mmapped_region_;
|
|
|
|
size_t length_;
|
|
|
|
|
|
|
|
public:
|
|
|
|
PosixMmapReadableFile(const int fd, const std::string& fname, void* base,
|
|
|
|
size_t length, const EnvOptions& options);
|
|
|
|
virtual ~PosixMmapReadableFile();
|
2022-06-10 23:34:01 +00:00
|
|
|
IOStatus Read(uint64_t offset, size_t n, const IOOptions& opts, Slice* result,
|
|
|
|
char* scratch, IODebugContext* dbg) const override;
|
|
|
|
void Hint(AccessPattern pattern) override;
|
|
|
|
IOStatus InvalidateCache(size_t offset, size_t length) override;
|
2015-10-14 08:14:53 +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
|
|
|
class PosixMmapFile : public FSWritableFile {
|
2015-10-14 08:14:53 +00:00
|
|
|
private:
|
|
|
|
std::string filename_;
|
|
|
|
int fd_;
|
|
|
|
size_t page_size_;
|
|
|
|
size_t map_size_; // How much extra memory to map at a time
|
|
|
|
char* base_; // The mapped region
|
|
|
|
char* limit_; // Limit of the mapped region
|
|
|
|
char* dst_; // Where to write next (in range [base_,limit_])
|
|
|
|
char* last_sync_; // Where have we synced up to
|
|
|
|
uint64_t file_offset_; // Offset of base_ in file
|
2015-10-28 18:55:20 +00:00
|
|
|
#ifdef ROCKSDB_FALLOCATE_PRESENT
|
2015-10-14 08:14:53 +00:00
|
|
|
bool allow_fallocate_; // If false, fallocate calls are bypassed
|
|
|
|
bool fallocate_with_keep_size_;
|
2015-10-28 18:55:20 +00:00
|
|
|
#endif
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
// Roundup x to a multiple of y
|
|
|
|
static size_t Roundup(size_t x, size_t y) { return ((x + y - 1) / y) * y; }
|
|
|
|
|
|
|
|
size_t TruncateToPageBoundary(size_t s) {
|
|
|
|
s -= (s & (page_size_ - 1));
|
|
|
|
assert((s % page_size_) == 0);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
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
|
|
|
IOStatus MapNewRegion();
|
|
|
|
IOStatus UnmapCurrentRegion();
|
|
|
|
IOStatus Msync();
|
2015-10-14 08:14:53 +00:00
|
|
|
|
|
|
|
public:
|
|
|
|
PosixMmapFile(const std::string& fname, int fd, size_t page_size,
|
|
|
|
const EnvOptions& options);
|
|
|
|
~PosixMmapFile();
|
|
|
|
|
|
|
|
// Means Close() will properly take care of truncate
|
|
|
|
// and it does not need any additional information
|
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
|
|
|
virtual IOStatus Truncate(uint64_t /*size*/, const IOOptions& /*opts*/,
|
|
|
|
IODebugContext* /*dbg*/) override {
|
|
|
|
return IOStatus::OK();
|
|
|
|
}
|
|
|
|
virtual IOStatus Close(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Append(const Slice& data, const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2020-09-24 02:00:30 +00:00
|
|
|
virtual IOStatus Append(const Slice& data, const IOOptions& opts,
|
|
|
|
const DataVerificationInfo& /* verification_info */,
|
|
|
|
IODebugContext* dbg) override {
|
|
|
|
return Append(data, opts, dbg);
|
|
|
|
}
|
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
|
|
|
virtual IOStatus Flush(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Sync(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Fsync(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual uint64_t GetFileSize(const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus InvalidateCache(size_t offset, size_t length) override;
|
2015-10-28 18:55:20 +00:00
|
|
|
#ifdef ROCKSDB_FALLOCATE_PRESENT
|
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
|
|
|
virtual IOStatus Allocate(uint64_t offset, uint64_t len,
|
|
|
|
const IOOptions& opts,
|
|
|
|
IODebugContext* dbg) override;
|
2015-10-28 18:55:20 +00:00
|
|
|
#endif
|
2015-10-14 08:14:53 +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
|
|
|
class PosixRandomRWFile : public FSRandomRWFile {
|
2016-09-13 19:08:22 +00:00
|
|
|
public:
|
|
|
|
explicit PosixRandomRWFile(const std::string& fname, int fd,
|
|
|
|
const EnvOptions& options);
|
|
|
|
virtual ~PosixRandomRWFile();
|
|
|
|
|
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
|
|
|
virtual IOStatus Write(uint64_t offset, const Slice& data,
|
|
|
|
const IOOptions& opts, IODebugContext* dbg) override;
|
2016-09-13 19:08:22 +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
|
|
|
virtual IOStatus Read(uint64_t offset, size_t n, const IOOptions& opts,
|
|
|
|
Slice* result, char* scratch,
|
|
|
|
IODebugContext* dbg) const override;
|
2016-09-13 19:08:22 +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
|
|
|
virtual IOStatus Flush(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Sync(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Fsync(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
virtual IOStatus Close(const IOOptions& opts, IODebugContext* dbg) override;
|
2016-09-13 19:08:22 +00:00
|
|
|
|
|
|
|
private:
|
|
|
|
const std::string filename_;
|
|
|
|
int fd_;
|
|
|
|
};
|
|
|
|
|
2018-04-30 19:23:45 +00:00
|
|
|
struct PosixMemoryMappedFileBuffer : public MemoryMappedFileBuffer {
|
|
|
|
PosixMemoryMappedFileBuffer(void* _base, size_t _length)
|
|
|
|
: MemoryMappedFileBuffer(_base, _length) {}
|
|
|
|
virtual ~PosixMemoryMappedFileBuffer();
|
|
|
|
};
|
|
|
|
|
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
|
|
|
class PosixDirectory : public FSDirectory {
|
2015-10-14 08:14:53 +00:00
|
|
|
public:
|
Explicitly closing all directory file descriptors (#10049)
Summary:
Currently, the DB directory file descriptor is left open until the deconstruction process (`DB::Close()` does not close the file descriptor). To verify this, comment out the lines between `db_ = nullptr` and `db_->Close()` (line 512, 513, 514, 515 in ldb_cmd.cc) to leak the ``db_'' object, build `ldb` tool and run
```
strace --trace=open,openat,close ./ldb --db=$TEST_TMPDIR --ignore_unknown_options put K1 V1 --create_if_missing
```
There is one directory file descriptor that is not closed in the strace log.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10049
Test Plan: Add a new unit test DBBasicTest.DBCloseAllDirectoryFDs: Open a database with different WAL directory and three different data directories, and all directory file descriptors should be closed after calling Close(). Explicitly call Close() after a directory file descriptor is not used so that the counter of directory open and close should be equivalent.
Reviewed By: ajkr, hx235
Differential Revision: D36722135
Pulled By: littlepig2013
fbshipit-source-id: 07bdc2abc417c6b30997b9bbef1f79aa757b21ff
2022-06-02 01:03:34 +00:00
|
|
|
explicit PosixDirectory(int fd, const std::string& directory_name);
|
2015-10-27 19:15:55 +00:00
|
|
|
~PosixDirectory();
|
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
|
|
|
virtual IOStatus Fsync(const IOOptions& opts, IODebugContext* dbg) override;
|
2015-10-14 08:14:53 +00:00
|
|
|
|
Explicitly closing all directory file descriptors (#10049)
Summary:
Currently, the DB directory file descriptor is left open until the deconstruction process (`DB::Close()` does not close the file descriptor). To verify this, comment out the lines between `db_ = nullptr` and `db_->Close()` (line 512, 513, 514, 515 in ldb_cmd.cc) to leak the ``db_'' object, build `ldb` tool and run
```
strace --trace=open,openat,close ./ldb --db=$TEST_TMPDIR --ignore_unknown_options put K1 V1 --create_if_missing
```
There is one directory file descriptor that is not closed in the strace log.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10049
Test Plan: Add a new unit test DBBasicTest.DBCloseAllDirectoryFDs: Open a database with different WAL directory and three different data directories, and all directory file descriptors should be closed after calling Close(). Explicitly call Close() after a directory file descriptor is not used so that the counter of directory open and close should be equivalent.
Reviewed By: ajkr, hx235
Differential Revision: D36722135
Pulled By: littlepig2013
fbshipit-source-id: 07bdc2abc417c6b30997b9bbef1f79aa757b21ff
2022-06-02 01:03:34 +00:00
|
|
|
virtual IOStatus Close(const IOOptions& opts, IODebugContext* dbg) override;
|
|
|
|
|
2021-11-03 19:20:19 +00:00
|
|
|
virtual IOStatus FsyncWithDirOptions(
|
|
|
|
const IOOptions&, IODebugContext*,
|
|
|
|
const DirFsyncOptions& dir_fsync_options) override;
|
|
|
|
|
2015-10-14 08:14:53 +00:00
|
|
|
private:
|
|
|
|
int fd_;
|
2021-11-03 19:20:19 +00:00
|
|
|
bool is_btrfs_;
|
Explicitly closing all directory file descriptors (#10049)
Summary:
Currently, the DB directory file descriptor is left open until the deconstruction process (`DB::Close()` does not close the file descriptor). To verify this, comment out the lines between `db_ = nullptr` and `db_->Close()` (line 512, 513, 514, 515 in ldb_cmd.cc) to leak the ``db_'' object, build `ldb` tool and run
```
strace --trace=open,openat,close ./ldb --db=$TEST_TMPDIR --ignore_unknown_options put K1 V1 --create_if_missing
```
There is one directory file descriptor that is not closed in the strace log.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10049
Test Plan: Add a new unit test DBBasicTest.DBCloseAllDirectoryFDs: Open a database with different WAL directory and three different data directories, and all directory file descriptors should be closed after calling Close(). Explicitly call Close() after a directory file descriptor is not used so that the counter of directory open and close should be equivalent.
Reviewed By: ajkr, hx235
Differential Revision: D36722135
Pulled By: littlepig2013
fbshipit-source-id: 07bdc2abc417c6b30997b9bbef1f79aa757b21ff
2022-06-02 01:03:34 +00:00
|
|
|
const std::string directory_name_;
|
2015-10-14 08:14:53 +00:00
|
|
|
};
|
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|