andreacavalli/rocksdb
0
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

direct io write support

Browse Source 
Summary:
rocksdb direct io support

```
[gzh@dev11575.prn2 ~/rocksdb] ./db_bench -benchmarks=fillseq --num=1000000
Initializing RocksDB Options from the specified file
Initializing RocksDB Options from command-line flags
RocksDB:    version 5.0
Date:       Wed Nov 23 13:17:43 2016
CPU:        40 * Intel(R) Xeon(R) CPU E5-2660 v2 @ 2.20GHz
CPUCache:   25600 KB
Keys:       16 bytes each
Values:     100 bytes each (50 bytes after compression)
Entries:    1000000
Prefix:    0 bytes
Keys per prefix:    0
RawSize:    110.6 MB (estimated)
FileSize:   62.9 MB (estimated)
Write rate: 0 bytes/second
Compression: Snappy
Memtablerep: skip_list
Perf Level: 1
WARNING: Assertions are enabled; benchmarks unnecessarily slow
------------------------------------------------
Initializing RocksDB Options from the specified file
Initializing RocksDB Options from command-line flags
DB path: [/tmp/rocksdbtest-112628/dbbench]
fillseq      :       4.393 micros/op 227639 ops/sec;   25.2 MB/s

[gzh@dev11575.prn2 ~/roc
Closes https://github.com/facebook/rocksdb/pull/1564

Differential Revision: D4241093

Pulled By: lightmark

fbshipit-source-id: 98c29e3
This commit is contained in:
Aaron Gao 2016-12-22 12:51:29 -08:00 committed by Facebook Github Bot
parent 989e644ed8
commit 972f96b3fb
49 changed files with 720 additions and 671 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
db/c.cc
View File

@ -1719,9 +1719,14 @@ void rocksdb_options_set_manifest_preallocation_size(
void rocksdb_options_set_purge_redundant_kvs_while_flush(rocksdb_options_t* opt,
unsigned char v) {}
void rocksdb_options_set_allow_os_buffer(rocksdb_options_t* opt,
void rocksdb_options_set_use_direct_reads(rocksdb_options_t* opt,
unsigned char v) {
opt->rep.allow_os_buffer = v;
opt->rep.use_direct_reads = v;
}
void rocksdb_options_set_use_direct_writes(rocksdb_options_t* opt,
unsigned char v) {
opt->rep.use_direct_writes = v;
}
void rocksdb_options_set_allow_mmap_reads(

10
db/db_impl.cc
View File

@ -255,11 +255,17 @@ static Status ValidateOptions(
"More than four DB paths are not supported yet. ");
}
if (db_options.allow_mmap_reads && !db_options.allow_os_buffer) {
if (db_options.allow_mmap_reads && db_options.use_direct_reads) {
// Protect against assert in PosixMMapReadableFile constructor
return Status::NotSupported(
"If memory mapped reads (allow_mmap_reads) are enabled "
"then os caching (allow_os_buffer) must also be enabled. ");
"then direct I/O reads (use_direct_reads) must be disabled. ");
}
if (db_options.allow_mmap_writes && db_options.use_direct_writes) {
return Status::NotSupported(
"If memory mapped writes (allow_mmap_writes) are enabled "
"then direct I/O writes (use_direct_writes) must be disabled. ");
}
return Status::OK();

10
db/db_test.cc
View File

@ -3396,19 +3396,21 @@ TEST_F(DBTest, TableOptionsSanitizeTest) {
TEST_F(DBTest, MmapAndBufferOptions) {
Options options = CurrentOptions();
options.allow_os_buffer = false;
options.use_direct_reads = true;
options.allow_mmap_reads = true;
ASSERT_NOK(TryReopen(options));
// All other combinations are acceptable
options.allow_os_buffer = true;
options.use_direct_reads = false;
ASSERT_OK(TryReopen(options));
options.allow_os_buffer = false;
if (IsDirectIOSupported()) {
options.use_direct_reads = true;
options.allow_mmap_reads = false;
ASSERT_OK(TryReopen(options));
}
options.allow_os_buffer = true;
options.use_direct_reads = false;
ASSERT_OK(TryReopen(options));
}
#endif

16
db/db_test_util.cc
View File

@ -478,6 +478,22 @@ Status DBTestBase::TryReopen(const Options& options) {
return DB::Open(options, dbname_, &db_);
}
bool DBTestBase::IsDirectIOSupported() {
EnvOptions env_options;
env_options.use_mmap_writes = false;
env_options.use_direct_writes = true;
std::string tmp = TempFileName(dbname_, 999);
Status s;
{
unique_ptr<WritableFile> file;
s = env_->NewWritableFile(tmp, &file, env_options);
}
if (s.ok()) {
s = env_->DeleteFile(tmp);
}
return s.ok();
}
Status DBTestBase::Flush(int cf) {
if (cf == 0) {
return db_->Flush(FlushOptions());

2
db/db_test_util.h
View File

@ -692,6 +692,8 @@ class DBTestBase : public testing::Test {
Status TryReopen(const Options& options);
bool IsDirectIOSupported();
Status Flush(int cf = 0);
Status Put(const Slice& k, const Slice& v, WriteOptions wo = WriteOptions());

2
db/forward_iterator_bench.cc
View File

@ -330,7 +330,7 @@ int main(int argc, char** argv) {
options.compaction_style = rocksdb::CompactionStyle::kCompactionStyleNone;
options.level0_slowdown_writes_trigger = 99999;
options.level0_stop_writes_trigger = 99999;
options.allow_os_buffer = false;
options.use_direct_writes = true;
options.write_buffer_size = FLAGS_memtable_size;
rocksdb::BlockBasedTableOptions table_options;
table_options.block_cache = rocksdb::NewLRUCache(FLAGS_block_cache_size);

5
examples/rocksdb_option_file_example.ini
View File

@ -73,11 +73,12 @@
error_if_exists=false
recycle_log_file_num=0
skip_log_error_on_recovery=false
allow_mmap_reads=false
allow_os_buffer=true
db_log_dir=
new_table_reader_for_compaction_inputs=true
allow_mmap_reads=false
allow_mmap_writes=false
use_direct_reads=false
use_direct_writes=false
[CFOptions "default"]

6
include/rocksdb/c.h
View File

@ -657,12 +657,14 @@ extern ROCKSDB_LIBRARY_API void rocksdb_options_set_manifest_preallocation_size(
extern ROCKSDB_LIBRARY_API void
rocksdb_options_set_purge_redundant_kvs_while_flush(rocksdb_options_t*,
unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_allow_os_buffer(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_allow_mmap_reads(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_allow_mmap_writes(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_use_direct_reads(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_use_direct_writes(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_is_fd_close_on_exec(
rocksdb_options_t*, unsigned char);
extern ROCKSDB_LIBRARY_API void rocksdb_options_set_skip_log_error_on_recovery(

38
include/rocksdb/env.h
View File

@ -61,9 +61,6 @@ struct EnvOptions {
// construct from Options
explicit EnvOptions(const DBOptions& options);
// If true, then allow caching of data in environment buffers
bool use_os_buffer = true;
// If true, then use mmap to read data
bool use_mmap_reads = false;
@ -373,8 +370,8 @@ class Env {
// OptimizeForManifestWrite will create a new EnvOptions object that is a copy
// of the EnvOptions in the parameters, but is optimized for writing manifest
// files. Default implementation returns the copy of the same object.
virtual EnvOptions OptimizeForManifestWrite(const EnvOptions& env_options)
const;
virtual EnvOptions OptimizeForManifestWrite(
const EnvOptions& env_options) const;
// Returns the status of all threads that belong to the current Env.
virtual Status GetThreadList(std::vector<ThreadStatus>* thread_list) {
@ -512,17 +509,15 @@ class WritableFile {
}
virtual ~WritableFile();
// Indicates if the class makes use of unbuffered I/O
// If false you must pass aligned buffer to Write()
virtual bool UseOSBuffer() const {
return true;
}
// Indicates if the class makes use of direct IO
// If true you must pass aligned buffer to Write()
virtual bool UseDirectIO() const { return false; }
const size_t c_DefaultPageSize = 4 * 1024;
// Use the returned alignment value to allocate
// aligned buffer for Write() when UseOSBuffer()
// returns false
// aligned buffer for Write() when UseDirectIO()
// returns true
virtual size_t GetRequiredBufferAlignment() const {
return c_DefaultPageSize;
}
@ -538,7 +533,7 @@ class WritableFile {
// the sector. The implementation thus needs to also rewrite the last
// partial sector.
// Note: PositionAppend does not guarantee moving the file offset after the
// write. A WriteabelFile object must support either Append or
// write. A WritableFile object must support either Append or
// PositionedAppend, so the users cannot mix the two.
//
// PositionedAppend() can only happen on the page/sector boundaries. For that
@ -583,10 +578,6 @@ class WritableFile {
return false;
}
// Indicates the upper layers if the current WritableFile implementation
// uses direct IO.
virtual bool UseDirectIO() const { return false; }
/*
* Change the priority in rate limiter if rate limiting is enabled.
* If rate limiting is not enabled, this call has no effect.
@ -695,17 +686,14 @@ class RandomRWFile {
RandomRWFile() {}
virtual ~RandomRWFile() {}
// Indicates if the class makes use of unbuffered I/O
// Indicates if the class makes use of direct I/O
// If false you must pass aligned buffer to Write()
virtual bool UseOSBuffer() const {
return true;
}
virtual bool UseDirectIO() const { return false; }
const size_t c_DefaultPageSize = 4 * 1024;
// Use the returned alignment value to allocate
// aligned buffer for Write() when UseOSBuffer()
// returns false
// Use the returned alignment value to allocate aligned
// buffer for Write() when UseDirectIO() returns true
virtual size_t GetRequiredBufferAlignment() const {
return c_DefaultPageSize;
}
@ -722,7 +710,7 @@ class RandomRWFile {
virtual void EnableReadAhead() {}
// Write bytes in `data` at offset `offset`, Returns Status::OK() on success.
// Pass aligned buffer when UseOSBuffer() returns false.
// Pass aligned buffer when UseDirectIO() returns true.
virtual Status Write(uint64_t offset, const Slice& data) = 0;
// Read up to `n` bytes starting from offset `offset` and store them in

32
include/rocksdb/options.h
View File

@ -1100,26 +1100,6 @@ struct DBOptions {
// large amounts of data (such as xfs's allocsize option).
size_t manifest_preallocation_size;
// Hint the OS that it should not buffer disk I/O. Enabling this
// parameter may improve performance but increases pressure on the
// system cache.
//
// The exact behavior of this parameter is platform dependent.
//
// On POSIX systems, after RocksDB reads data from disk it will
// mark the pages as "unneeded". The operating system may - or may not
// - evict these pages from memory, reducing pressure on the system
// cache. If the disk block is requested again this can result in
// additional disk I/O.
//
// On WINDOWS system, files will be opened in "unbuffered I/O" mode
// which means that data read from the disk will not be cached or
// bufferized. The hardware buffer of the devices may however still
// be used. Memory mapped files are not impacted by this parameter.
//
// Default: true
bool allow_os_buffer;
// Allow the OS to mmap file for reading sst tables. Default: false
bool allow_mmap_reads;
@ -1128,10 +1108,22 @@ struct DBOptions {
// Default: false
bool allow_mmap_writes;
// Enable direct I/O mode for read/write
// they may or may not improve performance depending on the use case
//
// Files will be opened in "direct I/O" mode
// which means that data r/w from the disk will not be cached or
// bufferized. The hardware buffer of the devices may however still
// be used. Memory mapped files are not impacted by these parameters.
// Use O_DIRECT for reading file
// Default: false
bool use_direct_reads;
// Use O_DIRECT for writing file
// Default: false
bool use_direct_writes;
// If false, fallocate() calls are bypassed
bool allow_fallocate;

6
include/rocksdb/table.h
View File

@ -157,9 +157,9 @@ struct BlockBasedTableOptions {
// a SstTable. Instead, buffer in WritableFileWriter will take
// care of the flushing when it is full.
//
// On Windows, this option helps a lot when unbuffered I/O
// (allow_os_buffer = false) is used, since it avoids small
// unbuffered disk write.
// This option helps a lot when direct I/O writes
// (use_direct_writes = true) is used, since it avoids small
// direct disk write.
//
// User may also adjust writable_file_max_buffer_size to optimize disk I/O
// size.

86
include/rocksdb/utilities/env_librados.h
View File

@ -1,5 +1,7 @@
// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
#ifndef ROCKSDB_UTILITIES_ENV_LIBRADOS_H
#define ROCKSDB_UTILITIES_ENV_LIBRADOS_H
@ -22,10 +24,9 @@ public:
// not exist, returns a non-OK status.
//
// The returned file will only be accessed by one thread at a time.
Status NewSequentialFile(
const std::string& fname,
Status NewSequentialFile(const std::string& fname,
std::unique_ptr<SequentialFile>* result,
const EnvOptions& options);
const EnvOptions& options) override;
// Create a brand new random access read-only file with the
// specified name. On success, stores a pointer to the new file in
@ -34,10 +35,9 @@ public:
// status.
//
// The returned file may be concurrently accessed by multiple threads.
Status NewRandomAccessFile(
const std::string& fname,
Status NewRandomAccessFile(const std::string& fname,
std::unique_ptr<RandomAccessFile>* result,
const EnvOptions& options);
const EnvOptions& options) override;
// Create an object that writes to a new file with the specified
// name. Deletes any existing file with the same name and creates a
@ -46,17 +46,15 @@ public:
// returns non-OK.
//
// The returned file will only be accessed by one thread at a time.
Status NewWritableFile(
const std::string& fname,
Status NewWritableFile(const std::string& fname,
std::unique_ptr<WritableFile>* result,
const EnvOptions& options);
const EnvOptions& options) override;
// Reuse an existing file by renaming it and opening it as writable.
Status ReuseWritableFile(
const std::string& fname,
Status ReuseWritableFile(const std::string& fname,
const std::string& old_fname,
std::unique_ptr<WritableFile>* result,
const EnvOptions& options);
const EnvOptions& options) override;
// Create an object that represents a directory. Will fail if directory
// doesn't exist. If the directory exists, it will open the directory
@ -65,47 +63,44 @@ public:
// On success, stores a pointer to the new Directory in
// *result and returns OK. On failure stores nullptr in *result and
// returns non-OK.
Status NewDirectory(
const std::string& name,
std::unique_ptr<Directory>* result);
Status NewDirectory(const std::string& name,
std::unique_ptr<Directory>* result) override;
// Returns OK if the named file exists.
// NotFound if the named file does not exist,
// the calling process does not have permission to determine
// whether this file exists, or if the path is invalid.
// IOError if an IO Error was encountered
Status FileExists(const std::string& fname);
Status FileExists(const std::string& fname) overrdie;
// Store in *result the names of the children of the specified directory.
// The names are relative to "dir".
// Original contents of *results are dropped.
Status GetChildren(const std::string& dir,
std::vector<std::string>* result);
Status GetChildren(const std::string& dir, std::vector<std::string>* result);
// Delete the named file.
Status DeleteFile(const std::string& fname);
Status DeleteFile(const std::string& fname) override;
// Create the specified directory. Returns error if directory exists.
Status CreateDir(const std::string& dirname);
Status CreateDir(const std::string& dirname) override;
// Creates directory if missing. Return Ok if it exists, or successful in
// Creating.
Status CreateDirIfMissing(const std::string& dirname);
Status CreateDirIfMissing(const std::string& dirname) override;
// Delete the specified directory.
Status DeleteDir(const std::string& dirname);
Status DeleteDir(const std::string& dirname) override;
// Store the size of fname in *file_size.
Status GetFileSize(const std::string& fname, uint64_t* file_size);
Status GetFileSize(const std::string& fname, uint64_t* file_size) override;
// Store the last modification time of fname in *file_mtime.
Status GetFileModificationTime(const std::string& fname,
uint64_t* file_mtime);
uint64_t* file_mtime) override;
// Rename file src to target.
Status RenameFile(const std::string& src,
const std::string& target);
Status RenameFile(const std::string& src, const std::string& target) override;
// Hard Link file src to target.
Status LinkFile(const std::string& src, const std::string& target);
Status LinkFile(const std::string& src, const std::string& target) override;
// Lock the specified file. Used to prevent concurrent access to
// the same db by multiple processes. On failure, stores nullptr in
@ -129,8 +124,7 @@ public:
Status UnlockFile(FileLock* lock);
// Get full directory name for this db.
Status GetAbsolutePath(const std::string& db_path,
std::string* output_path);
Status GetAbsolutePath(const std::string& db_path, std::string* output_path);
// Generate unique id
std::string GenerateUniqueId();
@ -142,23 +136,21 @@ public:
const std::string& config_path,
const std::string& db_pool);
explicit EnvLibrados(const std::string& client_name, // first 3 parameters are for RADOS client init
const std::string& cluster_name,
const uint64_t flags,
const std::string& db_name,
const std::string& config_path,
const std::string& db_pool,
const std::string& wal_dir,
const std::string& wal_pool,
const uint64_t write_buffer_size);
~EnvLibrados() {
_rados.shutdown();
}
explicit EnvLibrados(
const std::string& client_name, // first 3 parameters are
// for RADOS client init
const std::string& cluster_name, const uint64_t flags,
const std::string& db_name, const std::string& config_path,
const std::string& db_pool, const std::string& wal_dir,
const std::string& wal_pool, const uint64_t write_buffer_size);
~EnvLibrados() { _rados.shutdown(); }
private:
std::string _client_name;
std::string _cluster_name;
uint64_t _flags;
std::string _db_name; // get from user, readable string; Also used as db_id for db metadata
std::string _db_name; // get from user, readable string; Also used as db_id
// for db metadata
std::string _config_path;
librados::Rados _rados; // RADOS client
std::string _db_pool_name;
@ -175,10 +167,8 @@ private:
Status _RenameFid(const std::string& old_fname, const std::string& new_fname);
Status _AddFid(const std::string& fname, const std::string& fid);
Status _DelFid(const std::string& fname);
Status _GetSubFnames(
const std::string& dirname,
std::vector<std::string> * result
);
Status _GetSubFnames(const std::string& dirname,
std::vector<std::string>* result);
librados::IoCtx* _GetIoctx(const std::string& prefix);
friend class LibradosWritableFile;
};

7
include/rocksdb/utilities/env_mirror.h
View File

@ -15,10 +15,9 @@
// semantics and behavior are correct (in that they match that of an
// existing, stable Env, like the default POSIX one).
#ifndef ROCKSDB_LITE
#pragma once
#ifndef STORAGE_ROCKSDB_INCLUDE_UTILITIES_ENVMIRROR_H_
#define STORAGE_ROCKSDB_INCLUDE_UTLIITIES_ENVMIRROR_H_
#ifndef ROCKSDB_LITE
#include <iostream>
#include <algorithm>
@ -174,6 +173,4 @@ class EnvMirror : public EnvWrapper {
} // namespace rocksdb
#endif // STORAGE_ROCKSDB_INCLUDE_UTILITIES_ENVMIRROR_H_
#endif // ROCKSDB_LITE

10
java/benchmark/src/main/java/org/rocksdb/benchmark/DbBenchmark.java
View File

@ -1417,8 +1417,14 @@ public class DbBenchmark {
}
},
/* TODO(yhchiang): enable the following
bufferedio(rocksdb::EnvOptions().use_os_buffer,
"Allow buffered io using OS buffers.") {
direct_reads(rocksdb::EnvOptions().use_direct_reads,
"Allow direct I/O reads.") {
@Override public Object parseValue(String value) {
return parseBoolean(value);
}
},
direct_writes(rocksdb::EnvOptions().use_direct_reads,
"Allow direct I/O reads.") {
@Override public Object parseValue(String value) {
return parseBoolean(value);
}

75
java/rocksjni/env_options.cc
View File

@ -49,23 +49,43 @@ void Java_org_rocksdb_EnvOptions_disposeInternal(JNIEnv *env, jobject jobj,
/*
* Class: org_rocksdb_EnvOptions
* Method: setUseOsBuffer
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_EnvOptions_setUseOsBuffer(JNIEnv *env, jobject jobj,
void Java_org_rocksdb_EnvOptions_setUseDirectReads(JNIEnv *env, jobject jobj,
jlong jhandle,
jboolean use_os_buffer) {
ENV_OPTIONS_SET_BOOL(jhandle, use_os_buffer);
jboolean use_direct_reads) {
ENV_OPTIONS_SET_BOOL(jhandle, use_direct_reads);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: useOsBuffer
* Method: useDirectReads
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_EnvOptions_useOsBuffer(JNIEnv *env, jobject jobj,
jboolean Java_org_rocksdb_EnvOptions_useDirectReads(JNIEnv *env, jobject jobj,
jlong jhandle) {
return ENV_OPTIONS_GET(jhandle, use_os_buffer);
return ENV_OPTIONS_GET(jhandle, use_direct_reads);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: setUseDirectWrites
* Signature: (JZ)V
*/
void Java_org_rocksdb_EnvOptions_setUseDirectWrites(
JNIEnv *env, jobject jobj, jlong jhandle, jboolean use_direct_writes) {
ENV_OPTIONS_SET_BOOL(jhandle, use_direct_writes);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: useDirectWrites
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_EnvOptions_useDirectWrites(JNIEnv *env, jobject jobj,
jlong jhandle) {
return ENV_OPTIONS_GET(jhandle, use_direct_writes);
}
/*
@ -110,47 +130,6 @@ jboolean Java_org_rocksdb_EnvOptions_useMmapWrites(JNIEnv *env, jobject jobj,
return ENV_OPTIONS_GET(jhandle, use_mmap_writes);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_EnvOptions_setUseDirectReads(JNIEnv *env, jobject jobj,
jlong jhandle,
jboolean use_direct_reads) {
ENV_OPTIONS_SET_BOOL(jhandle, use_direct_reads);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: useDirectReads
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_EnvOptions_useDirectReads(JNIEnv *env, jobject jobj,
jlong jhandle) {
return ENV_OPTIONS_GET(jhandle, use_direct_reads);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: setUseDirectWrites
* Signature: (JZ)V
*/
void Java_org_rocksdb_EnvOptions_setUseDirectWrites(
JNIEnv *env, jobject jobj, jlong jhandle, jboolean use_direct_writes) {
ENV_OPTIONS_SET_BOOL(jhandle, use_direct_writes);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: useDirectWrites
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_EnvOptions_useDirectWrites(JNIEnv *env, jobject jobj,
jlong jhandle) {
return ENV_OPTIONS_GET(jhandle, use_direct_writes);
}
/*
* Class: org_rocksdb_EnvOptions
* Method: setAllowFallocate

108
java/rocksjni/options.cc
View File

@ -870,27 +870,6 @@ void Java_org_rocksdb_Options_setManifestPreallocationSize(
}
}
/*
* Class: org_rocksdb_Options
* Method: allowOsBuffer
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_Options_allowOsBuffer(
JNIEnv* env, jobject jobj, jlong jhandle) {
return reinterpret_cast<rocksdb::Options*>(jhandle)->allow_os_buffer;
}
/*
* Class: org_rocksdb_Options
* Method: setAllowOsBuffer
* Signature: (JZ)V
*/
void Java_org_rocksdb_Options_setAllowOsBuffer(
JNIEnv* env, jobject jobj, jlong jhandle, jboolean allow_os_buffer) {
reinterpret_cast<rocksdb::Options*>(jhandle)->allow_os_buffer =
static_cast<bool>(allow_os_buffer);
}
/*
* Method: setTableFactory
* Signature: (JJ)V
@ -943,6 +922,50 @@ void Java_org_rocksdb_Options_setAllowMmapWrites(
static_cast<bool>(allow_mmap_writes);
}
/*
* Class: org_rocksdb_Options
* Method: useDirectReads
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_Options_useDirectReads(JNIEnv* env, jobject jobj,
jlong jhandle) {
return reinterpret_cast<rocksdb::Options*>(jhandle)->use_direct_reads;
}
/*
* Class: org_rocksdb_Options
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_Options_setUseDirectReads(JNIEnv* env, jobject jobj,
jlong jhandle,
jboolean use_direct_reads) {
reinterpret_cast<rocksdb::Options*>(jhandle)->use_direct_reads =
static_cast<bool>(use_direct_reads);
}
/*
* Class: org_rocksdb_Options
* Method: useDirectWrites
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_Options_useDirectWrites(JNIEnv* env, jobject jobj,
jlong jhandle) {
return reinterpret_cast<rocksdb::Options*>(jhandle)->use_direct_writes;
}
/*
* Class: org_rocksdb_Options
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_Options_setUseDirectWrites(JNIEnv* env, jobject jobj,
jlong jhandle,
jboolean use_direct_writes) {
reinterpret_cast<rocksdb::Options*>(jhandle)->use_direct_writes =
static_cast<bool>(use_direct_writes);
}
/*
* Class: org_rocksdb_Options
* Method: isFdCloseOnExec
@ -4144,23 +4167,46 @@ jlong Java_org_rocksdb_DBOptions_manifestPreallocationSize(
/*
* Class: org_rocksdb_DBOptions
* Method: setAllowOsBuffer
* Signature: (JZ)V
* Method: useDirectReads
* Signature: (J)Z
*/
void Java_org_rocksdb_DBOptions_setAllowOsBuffer(
JNIEnv* env, jobject jobj, jlong jhandle, jboolean allow_os_buffer) {
reinterpret_cast<rocksdb::DBOptions*>(jhandle)->allow_os_buffer =
static_cast<bool>(allow_os_buffer);
jboolean Java_org_rocksdb_DBOptions_useDirectReads(JNIEnv* env, jobject jobj,
jlong jhandle) {
return reinterpret_cast<rocksdb::DBOptions*>(jhandle)->use_direct_reads;
}
/*
* Class: org_rocksdb_DBOptions
* Method: allowOsBuffer
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_DBOptions_setUseDirectReads(JNIEnv* env, jobject jobj,
jlong jhandle,
jboolean use_direct_reads) {
reinterpret_cast<rocksdb::DBOptions*>(jhandle)->use_direct_reads =
static_cast<bool>(use_direct_reads);
}
/*
* Class: org_rocksdb_DBOptions
* Method: useDirectWrites
* Signature: (J)Z
*/
jboolean Java_org_rocksdb_DBOptions_allowOsBuffer(
JNIEnv* env, jobject jobj, jlong jhandle) {
return reinterpret_cast<rocksdb::DBOptions*>(jhandle)->allow_os_buffer;
jboolean Java_org_rocksdb_DBOptions_useDirectWrites(JNIEnv* env, jobject jobj,
jlong jhandle) {
return reinterpret_cast<rocksdb::DBOptions*>(jhandle)->use_direct_writes;
}
/*
* Class: org_rocksdb_DBOptions
* Method: setUseDirectReads
* Signature: (JZ)V
*/
void Java_org_rocksdb_DBOptions_setUseDirectWrites(JNIEnv* env, jobject jobj,
jlong jhandle,
jboolean use_direct_writes) {
reinterpret_cast<rocksdb::DBOptions*>(jhandle)->use_direct_writes =
static_cast<bool>(use_direct_writes);
}
/*

31
java/src/main/java/org/rocksdb/DBOptions.java
View File

@ -457,17 +457,31 @@ public class DBOptions extends RocksObject implements DBOptionsInterface {
}
@Override
public DBOptions setAllowOsBuffer(
final boolean allowOsBuffer) {
public DBOptions setUseDirectReads(
final boolean useDirectReads) {
assert(isOwningHandle());
setAllowOsBuffer(nativeHandle_, allowOsBuffer);
setUseDirectReads(nativeHandle_, useDirectReads);
return this;
}
@Override
public boolean allowOsBuffer() {
public boolean useDirectReads() {
assert(isOwningHandle());
return allowOsBuffer(nativeHandle_);
return useDirectReads(nativeHandle_);
}
@Override
public DBOptions setUseDirectWrites(
final boolean useDirectWrites) {
assert(isOwningHandle());
setUseDirectWrites(nativeHandle_, useDirectWrites);
return this;
}
@Override
public boolean useDirectWrites() {
assert(isOwningHandle());
return useDirectWrites(nativeHandle_);
}
@Override
@ -710,9 +724,10 @@ public long delayedWriteRate(){
private native void setManifestPreallocationSize(
long handle, long size) throws IllegalArgumentException;
private native long manifestPreallocationSize(long handle);
private native void setAllowOsBuffer(
long handle, boolean allowOsBuffer);
private native boolean allowOsBuffer(long handle);
private native void setUseDirectReads(long handle, boolean useDirectReads);
private native boolean useDirectReads(long handle);
private native void setUseDirectWrites(long handle, boolean useDirectWrites);
private native boolean useDirectWrites(long handle);
private native void setAllowMmapReads(
long handle, boolean allowMmapReads);
private native boolean allowMmapReads(long handle);

33
java/src/main/java/org/rocksdb/DBOptionsInterface.java
View File

@ -673,21 +673,38 @@ public interface DBOptionsInterface {
long manifestPreallocationSize();
/**
* Data being read from file storage may be buffered in the OS
* Default: true
* Enable the OS to use direct I/O for reading sst tables.
* Default: false
*
* @param allowOsBuffer if true, then OS buffering is allowed.
* @param useDirectReads if true, then direct read is enabled
* @return the instance of the current Object.
*/
Object setAllowOsBuffer(boolean allowOsBuffer);
Object setUseDirectReads(boolean useDirectReads);
/**
* Data being read from file storage may be buffered in the OS
* Default: true
* Enable the OS to use direct I/O for reading sst tables.
* Default: false
*
* @return if true, then OS buffering is allowed.
* @return if true, then direct reads are enabled
*/
boolean allowOsBuffer();
boolean useDirectReads();
/**
* Enable the OS to use direct I/O for writing sst tables.
* Default: false
*
* @param useDirectWrites if true, then direct write is enabled
* @return the instance of the current Object.
*/
Object setUseDirectWrites(boolean useDirectWrites);
/**
* Enable the OS to use direct I/O for writing sst tables.
* Default: false
*
* @return if true, then direct writes are enabled
*/
boolean useDirectWrites();
/**
* Allow the OS to mmap file for reading sst tables.

29
java/src/main/java/org/rocksdb/Options.java
View File

@ -530,18 +530,32 @@ public class Options extends RocksObject
}
@Override
public boolean allowOsBuffer() {
public Options setUseDirectReads(final boolean useDirectReads) {
assert(isOwningHandle());
return allowOsBuffer(nativeHandle_);
setUseDirectReads(nativeHandle_, useDirectReads);
return this;
}
@Override
public Options setAllowOsBuffer(final boolean allowOsBuffer) {
public boolean useDirectReads() {
assert(isOwningHandle());
setAllowOsBuffer(nativeHandle_, allowOsBuffer);
return useDirectReads(nativeHandle_);
}
@Override
public Options setUseDirectWrites(final boolean useDirectWrites) {
assert(isOwningHandle());
setUseDirectWrites(nativeHandle_, useDirectWrites);
return this;
}
@Override
public boolean useDirectWrites() {
assert(isOwningHandle());
return useDirectWrites(nativeHandle_);
}
@Override
public boolean allowMmapReads() {
assert(isOwningHandle());
@ -1289,9 +1303,10 @@ public class Options extends RocksObject
private native void setManifestPreallocationSize(
long handle, long size) throws IllegalArgumentException;
private native long manifestPreallocationSize(long handle);
private native void setAllowOsBuffer(
long handle, boolean allowOsBuffer);
private native boolean allowOsBuffer(long handle);
private native void setUseDirectReads(long handle, boolean useDirectReads);
private native boolean useDirectReads(long handle);
private native void setUseDirectWrites(long handle, boolean useDirectWrites);
private native boolean useDirectWrites(long handle);
private native void setAllowMmapReads(
long handle, boolean allowMmapReads);
private native boolean allowMmapReads(long handle);

15
java/src/test/java/org/rocksdb/DBOptionsTest.java
View File

@ -281,11 +281,20 @@ public class DBOptionsTest {
}
@Test
public void allowOsBuffer() {
public void useDirectReads() {
try(final DBOptions opt = new DBOptions()) {
final boolean boolValue = rand.nextBoolean();
opt.setAllowOsBuffer(boolValue);
assertThat(opt.allowOsBuffer()).isEqualTo(boolValue);
opt.setUseDirectReads(boolValue);
assertThat(opt.useDirectReads()).isEqualTo(boolValue);
}
}
@Test
public void useDirectWrites() {
try(final DBOptions opt = new DBOptions()) {
final boolean boolValue = rand.nextBoolean();
opt.setUseDirectWrites(boolValue);
assertThat(opt.useDirectWrites()).isEqualTo(boolValue);
}
}

9
java/src/test/java/org/rocksdb/EnvOptionsTest.java
View File

@ -18,15 +18,6 @@ public class EnvOptionsTest {
public static final Random rand = PlatformRandomHelper.getPlatformSpecificRandomFactory();
@Test
public void useOsBuffer() {
try (final EnvOptions envOptions = new EnvOptions()) {
final boolean boolValue = rand.nextBoolean();
envOptions.setUseOsBuffer(boolValue);
assertThat(envOptions.useOsBuffer()).isEqualTo(boolValue);
}
}
@Test
public void useMmapReads() {
try (final EnvOptions envOptions = new EnvOptions()) {

15
java/src/test/java/org/rocksdb/OptionsTest.java
View File

@ -572,11 +572,20 @@ public class OptionsTest {
}
@Test
public void allowOsBuffer() {
public void useDirectReads() {
try(final Options opt = new Options()) {
final boolean boolValue = rand.nextBoolean();
opt.setAllowOsBuffer(boolValue);
assertThat(opt.allowOsBuffer()).isEqualTo(boolValue);
opt.setUseDirectReads(boolValue);
assertThat(opt.useDirectReads()).isEqualTo(boolValue);
}
}
@Test
public void useDirectWrites() {
try(final Options opt = new Options()) {
final boolean boolValue = rand.nextBoolean();
opt.setUseDirectWrites(boolValue);
assertThat(opt.useDirectWrites()).isEqualTo(boolValue);
}
}

18
port/win/env_win.cc
View File

@ -7,9 +7,10 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "port/win/env_win.h"
#include <algorithm>
#include <thread>
#include <ctime>
#include <thread>
#include <errno.h>
#include <process.h> // _getpid
@ -25,7 +26,6 @@
#include "port/dirent.h"
#include "port/win/win_logger.h"
#include "port/win/io_win.h"
#include "port/win/env_win.h"
#include "util/iostats_context_imp.h"
@ -148,7 +148,7 @@ Status WinEnvIO::NewRandomAccessFile(const std::string& fname,
// Random access is to disable read-ahead as the system reads too much data
DWORD fileFlags = FILE_ATTRIBUTE_READONLY;
if (!options.use_os_buffer && !options.use_mmap_reads) {
if (options.use_direct_reads && !options.use_mmap_reads) {
fileFlags |= FILE_FLAG_NO_BUFFERING;
} else {
fileFlags |= FILE_FLAG_RANDOM_ACCESS;
@ -240,7 +240,7 @@ Status WinEnvIO::NewWritableFile(const std::string& fname,
DWORD fileFlags = FILE_ATTRIBUTE_NORMAL;
if (!local_options.use_os_buffer && !local_options.use_mmap_writes) {
if (local_options.use_direct_writes && !local_options.use_mmap_writes) {
fileFlags = FILE_FLAG_NO_BUFFERING;
}
@ -305,7 +305,7 @@ Status WinEnvIO::NewRandomRWFile(const std::string & fname,
DWORD creation_disposition = OPEN_ALWAYS; // Create if necessary or open existing
DWORD file_flags = FILE_FLAG_RANDOM_ACCESS;
if (!options.use_os_buffer) {
if (options.use_direct_reads && options.use_direct_writes) {
file_flags |= FILE_FLAG_NO_BUFFERING;
}
@ -744,11 +744,11 @@ std::string WinEnvIO::TimeToString(uint64_t secondsSince1970) {
EnvOptions WinEnvIO::OptimizeForLogWrite(const EnvOptions& env_options,
const DBOptions& db_options) const {
EnvOptions optimized = env_options;
optimized.use_mmap_writes = false;
optimized.bytes_per_sync = db_options.wal_bytes_per_sync;
optimized.use_os_buffer =
true; // This is because we flush only whole pages on unbuffered io and
optimized.use_mmap_writes = false;
// This is because we flush only whole pages on unbuffered io and
// the last records are not guaranteed to be flushed.
optimized.use_direct_writes = false;
// TODO(icanadi) it's faster if fallocate_with_keep_size is false, but it
// breaks TransactionLogIteratorStallAtLastRecord unit test. Fix the unit
// test and make this false
@ -760,7 +760,7 @@ EnvOptions WinEnvIO::OptimizeForManifestWrite(
const EnvOptions& env_options) const {
EnvOptions optimized = env_options;
optimized.use_mmap_writes = false;
optimized.use_os_buffer = true;
optimized.use_direct_writes = false;
optimized.fallocate_with_keep_size = true;
return optimized;
}

76
port/win/io_win.cc
View File

@ -12,7 +12,6 @@
#include "util/sync_point.h"
#include "util/coding.h"
#include "util/iostats_context_imp.h"
#include "util/sync_point.h"
#include "util/aligned_buffer.h"
@ -158,9 +157,11 @@ size_t GetUniqueIdFromFile(HANDLE hFile, char* id, size_t max_size) {
////////////////////////////////////////////////////////////////////////////////////////////////////
// WinMmapReadableFile
WinMmapReadableFile::WinMmapReadableFile(const std::string& fileName, HANDLE hFile, HANDLE hMap,
const void* mapped_region, size_t length)
: WinFileData(fileName, hFile, false),
WinMmapReadableFile::WinMmapReadableFile(const std::string& fileName,
HANDLE hFile, HANDLE hMap,
const void* mapped_region,
size_t length)
: WinFileData(fileName, hFile, false /* use_direct_io */),
hMap_(hMap),
mapped_region_(mapped_region),
length_(length) {}
@ -522,8 +523,7 @@ size_t WinMmapFile::GetUniqueId(char* id, size_t max_size) const {
WinSequentialFile::WinSequentialFile(const std::string& fname, HANDLE f,
const EnvOptions& options)
: WinFileData(fname, f, options.use_os_buffer)
{}
: WinFileData(fname, f, options.use_direct_reads) {}
WinSequentialFile::~WinSequentialFile() {
assert(hFile_ != INVALID_HANDLE_VALUE);
@ -661,8 +661,8 @@ WinRandomAccessImpl::WinRandomAccessImpl(WinFileData* file_base,
assert(!options.use_mmap_reads);
// Unbuffered access, use internal buffer for reads
if (!file_base_->UseOSBuffer()) {
// Direct access, use internal buffer for reads
if (file_base_->UseDirectIO()) {
// Do not allocate the buffer either until the first request or
// until there is a call to allocate a read-ahead buffer
buffer_.Alignment(alignment);
@ -683,11 +683,10 @@ Status WinRandomAccessImpl::ReadImpl(uint64_t offset, size_t n, Slice* result,
return s;
}
// When in unbuffered mode we need to do the following changes:
// When in direct I/O mode we need to do the following changes:
// - use our own aligned buffer
// - always read at the offset of that is a multiple of alignment
if (!file_base_->UseOSBuffer()) {
if (file_base_->UseDirectIO()) {
uint64_t first_page_start = 0;
size_t actual_bytes_toread = 0;
size_t bytes_requested = left;
@ -778,9 +777,7 @@ Status WinRandomAccessImpl::ReadImpl(uint64_t offset, size_t n, Slice* result,
inline
void WinRandomAccessImpl::HintImpl(RandomAccessFile::AccessPattern pattern) {
if (pattern == RandomAccessFile::SEQUENTIAL &&
!file_base_->UseOSBuffer() &&
if (pattern == RandomAccessFile::SEQUENTIAL && file_base_->UseDirectIO() &&
compaction_readahead_size_ > 0) {
std::lock_guard<std::mutex> lg(buffer_mut_);
if (!read_ahead_) {
@ -798,11 +795,11 @@ void WinRandomAccessImpl::HintImpl(RandomAccessFile::AccessPattern pattern) {
///////////////////////////////////////////////////////////////////////////////////////////////////
/// WinRandomAccessFile
WinRandomAccessFile::WinRandomAccessFile(const std::string& fname, HANDLE hFile, size_t alignment,
const EnvOptions& options) :
WinFileData(fname, hFile, options.use_os_buffer),
WinRandomAccessImpl(this, alignment, options) {
}
WinRandomAccessFile::WinRandomAccessFile(const std::string& fname, HANDLE hFile,
size_t alignment,
const EnvOptions& options)
: WinFileData(fname, hFile, options.use_direct_reads),
WinRandomAccessImpl(this, alignment, options) {}
WinRandomAccessFile::~WinRandomAccessFile() {
}
@ -851,7 +848,7 @@ WinWritableImpl::WinWritableImpl(WinFileData* file_data, size_t alignment)
Status WinWritableImpl::AppendImpl(const Slice& data) {
// Used for buffered access ONLY
assert(file_data_->UseOSBuffer());
assert(!file_data_->UseDirectIO());
assert(data.size() < std::numeric_limits<DWORD>::max());
Status s;
@ -934,9 +931,8 @@ Status WinWritableImpl::SyncImpl() {
// Calls flush buffers
if (fsync(file_data_->GetFileHandle()) < 0) {
auto lastError = GetLastError();
s = IOErrorFromWindowsError("fsync failed at Sync() for: " +
file_data_->GetName(),
lastError);
s = IOErrorFromWindowsError(
"fsync failed at Sync() for: " + file_data_->GetName(), lastError);
}
return s;
}
@ -967,21 +963,19 @@ Status WinWritableImpl::AllocateImpl(uint64_t offset, uint64_t len) {
////////////////////////////////////////////////////////////////////////////////
/// WinWritableFile
WinWritableFile::WinWritableFile(const std::string& fname, HANDLE hFile, size_t alignment,
size_t /* capacity */, const EnvOptions& options)
: WinFileData(fname, hFile, options.use_os_buffer),
WinWritableFile::WinWritableFile(const std::string& fname, HANDLE hFile,
size_t alignment, size_t /* capacity */,
const EnvOptions& options)
: WinFileData(fname, hFile, options.use_direct_writes),
WinWritableImpl(this, alignment) {
assert(!options.use_mmap_writes);
}
WinWritableFile::~WinWritableFile() {
}
// Indicates if the class makes use of unbuffered I/O
bool WinWritableFile::UseOSBuffer() const {
return WinFileData::UseOSBuffer();
}
// Indicates if the class makes use of direct I/O
bool WinWritableFile::UseDirectIO() const { return WinFileData::UseDirectIO(); }
size_t WinWritableFile::GetRequiredBufferAlignment() const {
return GetAlignement();
@ -1015,9 +1009,7 @@ Status WinWritableFile::Sync() {
return SyncImpl();
}
Status WinWritableFile::Fsync() {
return SyncImpl();
}
Status WinWritableFile::Fsync() { return SyncImpl(); }
uint64_t WinWritableFile::GetFileSize() {
return GetFileSizeImpl();
@ -1034,17 +1026,14 @@ size_t WinWritableFile::GetUniqueId(char* id, size_t max_size) const {
/////////////////////////////////////////////////////////////////////////
/// WinRandomRWFile
WinRandomRWFile::WinRandomRWFile(const std::string& fname, HANDLE hFile, size_t alignment,
const EnvOptions& options) :
WinFileData(fname, hFile, options.use_os_buffer),
WinRandomRWFile::WinRandomRWFile(const std::string& fname, HANDLE hFile,
size_t alignment, const EnvOptions& options)
: WinFileData(fname, hFile,
options.use_direct_reads && options.use_direct_writes),
WinRandomAccessImpl(this, alignment, options),
WinWritableImpl(this, alignment) {
WinWritableImpl(this, alignment) {}
}
bool WinRandomRWFile::UseOSBuffer() const {
return WinFileData::UseOSBuffer();
}
bool WinRandomRWFile::UseDirectIO() const { return WinFileData::UseDirectIO(); }
size_t WinRandomRWFile::GetRequiredBufferAlignment() const {
return GetAlignement();
@ -1094,4 +1083,3 @@ WinFileLock::~WinFileLock() {
}
}

107
port/win/io_win.h
View File

@ -8,17 +8,16 @@
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#pragma once
#include <rocksdb/Status.h>
#include <rocksdb/env.h>
#include "util/aligned_buffer.h"
#include <string>
#include <stdint.h>
#include <mutex>
#include <string>
#include "rocksdb/Status.h"
#include "rocksdb/env.h"
#include "util/aligned_buffer.h"
#include <Windows.h>
#include <mutex>
namespace rocksdb {
namespace port {
@ -26,9 +25,9 @@ namespace port {
std::string GetWindowsErrSz(DWORD err);
inline Status IOErrorFromWindowsError(const std::string& context, DWORD err) {
return ((err == ERROR_HANDLE_DISK_FULL) || (err == ERROR_DISK_FULL)) ?
Status::NoSpace(context, GetWindowsErrSz(err)) :
Status::IOError(context, GetWindowsErrSz(err));
return ((err == ERROR_HANDLE_DISK_FULL) || (err == ERROR_DISK_FULL))
? Status::NoSpace(context, GetWindowsErrSz(err))
: Status::IOError(context, GetWindowsErrSz(err));
}
inline Status IOErrorFromLastWindowsError(const std::string& context) {
@ -36,9 +35,9 @@ inline Status IOErrorFromLastWindowsError(const std::string& context) {
}
inline Status IOError(const std::string& context, int err_number) {
return (err_number == ENOSPC) ?
Status::NoSpace(context, strerror(err_number)) :
Status::IOError(context, strerror(err_number));
return (err_number == ENOSPC)
? Status::NoSpace(context, strerror(err_number))
: Status::IOError(context, strerror(err_number));
}
// Note the below two do not set errno because they are used only here in this
@ -54,49 +53,34 @@ inline int fsync(HANDLE hFile) {
return 0;
}
SSIZE_T pwrite(HANDLE hFile, const char* src, size_t numBytes,
uint64_t offset);
SSIZE_T pwrite(HANDLE hFile, const char* src, size_t numBytes, uint64_t offset);
SSIZE_T pread(HANDLE hFile, char* src, size_t numBytes, uint64_t offset);
Status fallocate(const std::string& filename, HANDLE hFile,
uint64_t to_size);
Status ftruncate(const std::string& filename, HANDLE hFile,
uint64_t toSize);
Status fallocate(const std::string& filename, HANDLE hFile, uint64_t to_size);
Status ftruncate(const std::string& filename, HANDLE hFile, uint64_t toSize);
size_t GetUniqueIdFromFile(HANDLE hFile, char* id, size_t max_size);
class WinFileData {
protected:
const std::string filename_;
HANDLE hFile_;
// There is no equivalent of advising away buffered pages as in posix.
// To implement this flag we would need to do unbuffered reads which
// If ture, the I/O issued would be direct I/O which the buffer
// will need to be aligned (not sure there is a guarantee that the buffer
// passed in is aligned).
// Hence we currently ignore this flag. It is used only in a few cases
// which should not be perf critical.
// If perf evaluation finds this to be a problem, we can look into
// implementing this.
const bool use_os_buffer_;
const bool use_direct_io_;
public:
// We want this class be usable both for inheritance (prive
// or protected) and for containment so __ctor and __dtor public
WinFileData(const std::string& filename, HANDLE hFile, bool use_os_buffer) :
filename_(filename), hFile_(hFile), use_os_buffer_(use_os_buffer)
{}
WinFileData(const std::string& filename, HANDLE hFile, bool use_direct_io)
: filename_(filename), hFile_(hFile), use_direct_io_(use_direct_io) {}
virtual ~WinFileData() {
this->CloseFile();
}
virtual ~WinFileData() { this->CloseFile(); }
bool CloseFile() {
bool result = true;
if (hFile_ != NULL && hFile_ != INVALID_HANDLE_VALUE) {
@ -111,13 +95,12 @@ public:
HANDLE GetFileHandle() const { return hFile_; }
bool UseOSBuffer() const { return use_os_buffer_; }
bool UseDirectIO() const { return use_direct_io_; }
WinFileData(const WinFileData&) = delete;
WinFileData& operator=(const WinFileData&) = delete;
};
// mmap() based random-access
class WinMmapReadableFile : private WinFileData, public RandomAccessFile {
HANDLE hMap_;
@ -185,7 +168,6 @@ private:
virtual Status PreallocateInternal(uint64_t spaceToReserve);
public:
WinMmapFile(const std::string& fname, HANDLE hFile, size_t page_size,
size_t allocation_granularity, const EnvOptions& options);
@ -245,7 +227,6 @@ public:
class WinRandomAccessImpl {
protected:
WinFileData* file_base_;
bool read_ahead_;
const size_t compaction_readahead_size_;
@ -301,11 +282,9 @@ protected:
virtual ~WinRandomAccessImpl() {}
public:
WinRandomAccessImpl(const WinRandomAccessImpl&) = delete;
WinRandomAccessImpl& operator=(const WinRandomAccessImpl&) = delete;
Status ReadImpl(uint64_t offset, size_t n, Slice* result,
char* scratch) const;
@ -313,10 +292,11 @@ public:
};
// pread() based random-access
class WinRandomAccessFile : private WinFileData,
protected WinRandomAccessImpl, // Want to be able to override PositionedReadInternal
class WinRandomAccessFile
: private WinFileData,
protected WinRandomAccessImpl, // Want to be able to override
// PositionedReadInternal
public RandomAccessFile {
public:
WinRandomAccessFile(const std::string& fname, HANDLE hFile, size_t alignment,
const EnvOptions& options);
@ -337,7 +317,6 @@ public:
virtual size_t GetUniqueId(char* id, size_t max_size) const override;
};
// This is a sequential write class. It has been mimicked (as others) after
// the original Posix class. We add support for unbuffered I/O on windows as
// well
@ -352,7 +331,6 @@ public:
// buffered access.
class WinWritableImpl {
protected:
WinFileData* file_data_;
const uint64_t alignment_;
uint64_t filesize_; // How much data is actually written disk
@ -368,7 +346,8 @@ protected:
Status AppendImpl(const Slice& data);
// Requires that the data is aligned as specified by GetRequiredBufferAlignment()
// Requires that the data is aligned as specified by
// GetRequiredBufferAlignment()
Status PositionedAppendImpl(const Slice& data, uint64_t offset);
Status TruncateImpl(uint64_t size);
@ -380,7 +359,8 @@ protected:
uint64_t GetFileSizeImpl() {
// Double accounting now here with WritableFileWriter
// and this size will be wrong when unbuffered access is used
// but tests implement their own writable files and do not use WritableFileWrapper
// but tests implement their own writable files and do not use
// WritableFileWrapper
// so we need to squeeze a square peg through
// a round hole here.
return filesize_;
@ -389,31 +369,29 @@ protected:
Status AllocateImpl(uint64_t offset, uint64_t len);
public:
WinWritableImpl(const WinWritableImpl&) = delete;
WinWritableImpl& operator=(const WinWritableImpl&) = delete;
};
class WinWritableFile : private WinFileData,
protected WinWritableImpl,
public WritableFile {
public:
WinWritableFile(const std::string& fname, HANDLE hFile, size_t alignment,
size_t capacity, const EnvOptions& options);
~WinWritableFile();
// Indicates if the class makes use of unbuffered I/O
// Indicates if the class makes use of direct I/O
// Use PositionedAppend
virtual bool UseOSBuffer() const override;
virtual bool UseDirectIO() const override;
virtual size_t GetRequiredBufferAlignment() const override;
virtual Status Append(const Slice& data) override;
// Requires that the data is aligned as specified by GetRequiredBufferAlignment()
// Requires that the data is aligned as specified by
// GetRequiredBufferAlignment()
virtual Status PositionedAppend(const Slice& data, uint64_t offset) override;
// Need to implement this so the file is truncated correctly
@ -437,30 +415,27 @@ public:
virtual size_t GetUniqueId(char* id, size_t max_size) const override;
};
class WinRandomRWFile : private WinFileData,
protected WinRandomAccessImpl,
protected WinWritableImpl,
public RandomRWFile {
public:
WinRandomRWFile(const std::string& fname, HANDLE hFile, size_t alignment,
const EnvOptions& options);
~WinRandomRWFile() {}
// Indicates if the class makes use of unbuffered I/O
// Indicates if the class makes use of direct I/O
// If false you must pass aligned buffer to Write()
virtual bool UseOSBuffer() const override;
virtual bool UseDirectIO() const override;
// Use the returned alignment value to allocate
// aligned buffer for Write() when UseOSBuffer()
// returns false
// Use the returned alignment value to allocate aligned
// buffer for Write() when UseDirectIO() returns true
virtual size_t GetRequiredBufferAlignment() const override;
// Used by the file_reader_writer to decide if the ReadAhead wrapper
// should simply forward the call and do not enact read_ahead buffering or locking.
// should simply forward the call and do not enact read_ahead buffering or
// locking.
// The implementation below takes care of reading ahead
virtual bool ShouldForwardRawRequest() const override;
@ -469,7 +444,7 @@ public:
virtual void EnableReadAhead() override;
// Write bytes in `data` at offset `offset`, Returns Status::OK() on success.
// Pass aligned buffer when UseOSBuffer() returns false.
// Pass aligned buffer when UseDirectIO() returns true.
virtual Status Write(uint64_t offset, const Slice& data) override;
// Read up to `n` bytes starting from offset `offset` and store them in
@ -487,7 +462,6 @@ public:
virtual Status Close() override;
};
class WinDirectory : public Directory {
public:
WinDirectory() {}
@ -507,6 +481,5 @@ public:
private:
HANDLE hFile_;
};
}
}

10
tools/db_bench_tool.cc
View File

@ -771,9 +771,6 @@ DEFINE_uint64(wal_size_limit_MB, 0, "Set the size limit for the WAL Files"
" in MB.");
DEFINE_uint64(max_total_wal_size, 0, "Set total max WAL size");
DEFINE_bool(bufferedio, rocksdb::EnvOptions().use_os_buffer,
"Allow buffered io using OS buffers");
DEFINE_bool(mmap_read, rocksdb::EnvOptions().use_mmap_reads,
"Allow reads to occur via mmap-ing files");
@ -783,6 +780,9 @@ DEFINE_bool(mmap_write, rocksdb::EnvOptions().use_mmap_writes,
DEFINE_bool(use_direct_reads, rocksdb::EnvOptions().use_direct_reads,
"Use O_DIRECT for reading data");
DEFINE_bool(use_direct_writes, rocksdb::EnvOptions().use_direct_writes,
"Use O_DIRECT for writing data");
DEFINE_bool(advise_random_on_open, rocksdb::Options().advise_random_on_open,
"Advise random access on table file open");
@ -2750,6 +2750,7 @@ class Benchmark {
options.allow_mmap_reads = FLAGS_mmap_read;
options.allow_mmap_writes = FLAGS_mmap_write;
options.use_direct_reads = FLAGS_use_direct_reads;
options.use_direct_writes = FLAGS_use_direct_writes;
if (FLAGS_prefix_size != 0) {
options.prefix_extractor.reset(
NewFixedPrefixTransform(FLAGS_prefix_size));
@ -2951,9 +2952,6 @@ class Benchmark {
options.optimize_filters_for_hits = FLAGS_optimize_filters_for_hits;
// fill storage options
options.allow_os_buffer = FLAGS_bufferedio;
options.allow_mmap_reads = FLAGS_mmap_read;
options.allow_mmap_writes = FLAGS_mmap_write;
options.advise_random_on_open = FLAGS_advise_random_on_open;
options.access_hint_on_compaction_start = FLAGS_compaction_fadvice_e;
options.use_adaptive_mutex = FLAGS_use_adaptive_mutex;

5
tools/db_bench_tool_test.cc
View File

@ -208,13 +208,14 @@ const std::string options_file_content = R"OPTIONS_FILE(
max_background_flushes=1
create_if_missing=true
error_if_exists=false
allow_os_buffer=true
delayed_write_rate=1048576
manifest_preallocation_size=4194304
allow_mmap_reads=false
allow_mmap_writes=false
use_direct_reads=false
use_direct_writes=false
stats_dump_period_sec=600
allow_fallocate=true
allow_mmap_reads=false
max_log_file_size=83886080
random_access_max_buffer_size=1048576
advise_random_on_open=true

2
util/aligned_buffer.h
View File

@ -24,7 +24,7 @@ inline size_t Roundup(size_t x, size_t y) {
}
// This class is to manage an aligned user
// allocated buffer for unbuffered I/O purposes
// allocated buffer for direct I/O purposes
// though can be used for any purpose.
class AlignedBuffer {
size_t alignment_;

16
util/db_options.cc
View File

@ -50,10 +50,10 @@ ImmutableDBOptions::ImmutableDBOptions(const DBOptions& options)
wal_ttl_seconds(options.WAL_ttl_seconds),
wal_size_limit_mb(options.WAL_size_limit_MB),
manifest_preallocation_size(options.manifest_preallocation_size),
allow_os_buffer(options.allow_os_buffer),
allow_mmap_reads(options.allow_mmap_reads),
allow_mmap_writes(options.allow_mmap_writes),
use_direct_reads(options.use_direct_reads),
use_direct_writes(options.use_direct_writes),
allow_fallocate(options.allow_fallocate),
is_fd_close_on_exec(options.is_fd_close_on_exec),
stats_dump_period_sec(options.stats_dump_period_sec),
@ -119,16 +119,16 @@ void ImmutableDBOptions::Dump(Logger* log) const {
Header(log,
" Options.recycle_log_file_num: %" ROCKSDB_PRIszt,
recycle_log_file_num);
Header(log, " Options.allow_os_buffer: %d",
allow_os_buffer);
Header(log, " Options.allow_mmap_reads: %d",
allow_mmap_reads);
Header(log, " Options.allow_fallocate: %d",
allow_fallocate);
Header(log, " Options.allow_mmap_reads: %d",
allow_mmap_reads);
Header(log, " Options.allow_mmap_writes: %d",
allow_mmap_writes);
Header(log, " Options.use_direct_reads: %d",
use_direct_reads);
Header(log, " Options.use_direct_writes: %d",
use_direct_writes);
Header(log, " Options.create_missing_column_families: %d",
create_missing_column_families);
Header(log, " Options.db_log_dir: %s",
@ -148,12 +148,6 @@ void ImmutableDBOptions::Dump(Logger* log) const {
Header(log,
" Options.manifest_preallocation_size: %" ROCKSDB_PRIszt,
manifest_preallocation_size);
Header(log, " Options.allow_os_buffer: %d",
allow_os_buffer);
Header(log, " Options.allow_mmap_reads: %d",
allow_mmap_reads);
Header(log, " Options.allow_mmap_writes: %d",
allow_mmap_writes);
Header(log, " Options.is_fd_close_on_exec: %d",
is_fd_close_on_exec);
Header(log, " Options.stats_dump_period_sec: %u",

2
util/db_options.h
View File

@ -46,10 +46,10 @@ struct ImmutableDBOptions {
uint64_t wal_ttl_seconds;
uint64_t wal_size_limit_mb;
size_t manifest_preallocation_size;
bool allow_os_buffer;
bool allow_mmap_reads;
bool allow_mmap_writes;
bool use_direct_reads;
bool use_direct_writes;
bool allow_fallocate;
bool is_fd_close_on_exec;
unsigned int stats_dump_period_sec;

2
util/env.cc
View File

@ -313,10 +313,10 @@ EnvWrapper::~EnvWrapper() {
namespace { // anonymous namespace
void AssignEnvOptions(EnvOptions* env_options, const DBOptions& options) {
env_options->use_os_buffer = options.allow_os_buffer;
env_options->use_mmap_reads = options.allow_mmap_reads;
env_options->use_mmap_writes = options.allow_mmap_writes;
env_options->use_direct_reads = options.use_direct_reads;
env_options->use_direct_writes = options.use_direct_writes;
env_options->set_fd_cloexec = options.is_fd_close_on_exec;
env_options->bytes_per_sync = options.bytes_per_sync;
env_options->compaction_readahead_size = options.compaction_readahead_size;

122
util/env_posix.cc
View File

@ -38,6 +38,7 @@
#endif
#include <deque>
#include <set>
#include <vector>
#include "port/port.h"
#include "rocksdb/slice.h"
#include "util/coding.h"
@ -250,31 +251,9 @@ class PosixEnv : public Env {
result->reset();
Status s;
int fd = -1;
do {
IOSTATS_TIMER_GUARD(open_nanos);
fd = open(fname.c_str(), O_CREAT | O_RDWR | O_TRUNC, 0644);
} while (fd < 0 && errno == EINTR);
if (fd < 0) {
s = IOError(fname, errno);
} else {
SetFD_CLOEXEC(fd, &options);
if (options.use_mmap_writes) {
if (!checkedDiskForMmap_) {
// this will be executed once in the program's lifetime.
// do not use mmapWrite on non ext-3/xfs/tmpfs systems.
if (!SupportsFastAllocate(fname)) {
forceMmapOff = true;
}
checkedDiskForMmap_ = true;
}
}
if (options.use_mmap_writes && !forceMmapOff) {
result->reset(new PosixMmapFile(fname, fd, page_size_, options));
} else if (options.use_direct_writes) {
close(fd);
#ifdef OS_MACOSX
int flags = O_WRONLY | O_APPEND | O_TRUNC | O_CREAT;
#else
int flags = O_CREAT | O_TRUNC;
// Direct IO mode with O_DIRECT flag or F_NOCAHCE (MAC OSX)
if (options.use_direct_writes && !options.use_mmap_writes) {
// Note: we should avoid O_APPEND here due to ta the following bug:
// POSIX requires that opening a file with the O_APPEND flag should
// have no affect on the location at which pwrite() writes data.
@ -282,32 +261,56 @@ class PosixEnv : public Env {
// appends data to the end of the file, regardless of the value of
// offset.
// More info here: https://linux.die.net/man/2/pwrite
int flags = O_WRONLY | O_TRUNC | O_CREAT | O_DIRECT;
flags |= O_WRONLY;
#ifndef OS_MACOSX
flags |= O_DIRECT;
#endif
TEST_SYNC_POINT_CALLBACK("NewWritableFile:O_DIRECT", &flags);
} else if (options.use_mmap_writes) {
// non-direct I/O
flags |= O_RDWR;
} else {
flags |= O_WRONLY;
}
do {
IOSTATS_TIMER_GUARD(open_nanos);
fd = open(fname.c_str(), flags, 0644);
} while (fd < 0 && errno == EINTR);
if (fd < 0) {
s = IOError(fname, errno);
} else {
std::unique_ptr<PosixDirectIOWritableFile> file(
new PosixDirectIOWritableFile(fname, fd));
*result = std::move(file);
s = Status::OK();
return s;
}
SetFD_CLOEXEC(fd, &options);
if (options.use_mmap_writes) {
if (!checkedDiskForMmap_) {
// this will be executed once in the program's lifetime.
// do not use mmapWrite on non ext-3/xfs/tmpfs systems.
if (!SupportsFastAllocate(fname)) {
forceMmapOff_ = true;
}
checkedDiskForMmap_ = true;
}
}
if (options.use_mmap_writes && !forceMmapOff_) {
result->reset(new PosixMmapFile(fname, fd, page_size_, options));
} else if (options.use_direct_writes && !options.use_mmap_writes) {
#ifdef OS_MACOSX
if (fcntl(fd, F_NOCACHE, 1) == -1) {
close(fd);
s = IOError(fname, errno);
return s;
}
#endif
}
result->reset(new PosixWritableFile(fname, fd, options));
} else {
// disable mmap writes
EnvOptions no_mmap_writes_options = options;
no_mmap_writes_options.use_mmap_writes = false;
result->reset(new PosixWritableFile(fname, fd, no_mmap_writes_options));
}
}
return s;
}
@ -318,40 +321,68 @@ class PosixEnv : public Env {
result->reset();
Status s;
int fd = -1;
int flags = 0;
// Direct IO mode with O_DIRECT flag or F_NOCAHCE (MAC OSX)
if (options.use_direct_writes && !options.use_mmap_writes) {
flags |= O_WRONLY;
#ifndef OS_MACOSX
flags |= O_DIRECT;
#endif
TEST_SYNC_POINT_CALLBACK("NewWritableFile:O_DIRECT", &flags);
} else if (options.use_mmap_writes) {
// mmap needs O_RDWR mode
flags |= O_RDWR;
} else {
flags |= O_WRONLY;
}
do {
IOSTATS_TIMER_GUARD(open_nanos);
fd = open(old_fname.c_str(), O_RDWR, 0644);
fd = open(old_fname.c_str(), flags, 0644);
} while (fd < 0 && errno == EINTR);
if (fd < 0) {
s = IOError(fname, errno);
} else {
return s;
}
SetFD_CLOEXEC(fd, &options);
// rename into place
if (rename(old_fname.c_str(), fname.c_str()) != 0) {
Status r = IOError(old_fname, errno);
s = IOError(old_fname, errno);
close(fd);
return r;
return s;
}
if (options.use_mmap_writes) {
if (!checkedDiskForMmap_) {
// this will be executed once in the program's lifetime.
// do not use mmapWrite on non ext-3/xfs/tmpfs systems.
if (!SupportsFastAllocate(fname)) {
forceMmapOff = true;
forceMmapOff_ = true;
}
checkedDiskForMmap_ = true;
}
}
if (options.use_mmap_writes && !forceMmapOff) {
if (options.use_mmap_writes && !forceMmapOff_) {
result->reset(new PosixMmapFile(fname, fd, page_size_, options));
} else if (options.use_direct_writes && !options.use_mmap_writes) {
#ifdef OS_MACOSX
if (fcntl(fd, F_NOCACHE, 1) == -1) {
close(fd);
s = IOError(fname, errno);
return s;
}
#endif
result->reset(new PosixWritableFile(fname, fd, options));
} else {
// disable mmap writes
EnvOptions no_mmap_writes_options = options;
no_mmap_writes_options.use_mmap_writes = false;
result->reset(new PosixWritableFile(fname, fd, no_mmap_writes_options));
}
}
return s;
return s;
}
@ -724,6 +755,7 @@ class PosixEnv : public Env {
const DBOptions& db_options) const override {
EnvOptions optimized = env_options;
optimized.use_mmap_writes = false;
optimized.use_direct_writes = false;
optimized.bytes_per_sync = db_options.wal_bytes_per_sync;
// TODO(icanadi) it's faster if fallocate_with_keep_size is false, but it
// breaks TransactionLogIteratorStallAtLastRecord unit test. Fix the unit
@ -736,14 +768,14 @@ class PosixEnv : public Env {
const EnvOptions& env_options) const override {
EnvOptions optimized = env_options;
optimized.use_mmap_writes = false;
optimized.use_direct_writes = false;
optimized.fallocate_with_keep_size = true;
return optimized;
}
private:
bool checkedDiskForMmap_;
bool forceMmapOff; // do we override Env options?
bool forceMmapOff_; // do we override Env options?
// Returns true iff the named directory exists and is a directory.
virtual bool DirExists(const std::string& dname) {
@ -784,7 +816,7 @@ class PosixEnv : public Env {
PosixEnv::PosixEnv()
: checkedDiskForMmap_(false),
forceMmapOff(false),
forceMmapOff_(false),
page_size_(getpagesize()),
thread_pools_(Priority::TOTAL) {
ThreadPoolImpl::PthreadCall("mutex_init", pthread_mutex_init(&mu_, nullptr));

34
util/file_reader_writer.cc
View File

@ -61,9 +61,8 @@ Status WritableFileWriter::Append(const Slice& data) {
writable_file_->PrepareWrite(static_cast<size_t>(GetFileSize()), left);
}
// Flush only when I/O is buffered
if (use_os_buffer_ &&
(buf_.Capacity() - buf_.CurrentSize()) < left) {
// Flush only when buffered I/O
if (!direct_io_ && (buf_.Capacity() - buf_.CurrentSize()) < left) {
if (buf_.CurrentSize() > 0) {
s = Flush();
if (!s.ok()) {
@ -79,10 +78,10 @@ Status WritableFileWriter::Append(const Slice& data) {
assert(buf_.CurrentSize() == 0);
}
// We never write directly to disk with unbuffered I/O on.
// We never write directly to disk with direct I/O on.
// or we simply use it for its original purpose to accumulate many small
// chunks
if (!use_os_buffer_ || (buf_.Capacity() >= left)) {
if (direct_io_ || (buf_.Capacity() >= left)) {
while (left > 0) {
size_t appended = buf_.Append(src, left);
left -= appended;
@ -96,7 +95,7 @@ Status WritableFileWriter::Append(const Slice& data) {
// We double the buffer here because
// Flush calls do not keep up with the incoming bytes
// This is the only place when buffer is changed with unbuffered I/O
// This is the only place when buffer is changed with direct I/O
if (buf_.Capacity() < max_buffer_size_) {
size_t desiredCapacity = buf_.Capacity() * 2;
desiredCapacity = std::min(desiredCapacity, max_buffer_size_);
@ -132,7 +131,7 @@ Status WritableFileWriter::Close() {
s = Flush(); // flush cache to OS
// In unbuffered mode we write whole pages so
// In direct I/O mode we write whole pages so
// we need to let the file know where data ends.
Status interim = writable_file_->Truncate(filesize_);
if (!interim.ok() && s.ok()) {
@ -151,17 +150,18 @@ Status WritableFileWriter::Close() {
return s;
}
// write out the cached data to the OS cache
// write out the cached data to the OS cache or storage if direct I/O
// enabled
Status WritableFileWriter::Flush() {
Status s;
TEST_KILL_RANDOM("WritableFileWriter::Flush:0",
rocksdb_kill_odds * REDUCE_ODDS2);
if (buf_.CurrentSize() > 0) {
if (use_os_buffer_) {
s = WriteBuffered(buf_.BufferStart(), buf_.CurrentSize());
if (direct_io_) {
s = WriteDirect();
} else {
s = WriteUnbuffered();
s = WriteBuffered(buf_.BufferStart(), buf_.CurrentSize());
}
if (!s.ok()) {
return s;
@ -259,8 +259,8 @@ size_t WritableFileWriter::RequestToken(size_t bytes, bool align) {
if (align) {
// Here we may actually require more than burst and block
// but we can not write less than one page at a time on unbuffered
// thus we may want not to use ratelimiter s
// but we can not write less than one page at a time on direct I/O
// thus we may want not to use ratelimiter
size_t alignment = buf_.Alignment();
bytes = std::max(alignment, TruncateToPageBoundary(alignment, bytes));
}
@ -273,7 +273,7 @@ size_t WritableFileWriter::RequestToken(size_t bytes, bool align) {
// limiter if available
Status WritableFileWriter::WriteBuffered(const char* data, size_t size) {
Status s;
assert(use_os_buffer_);
assert(!direct_io_);
const char* src = data;
size_t left = size;
@ -308,10 +308,10 @@ Status WritableFileWriter::WriteBuffered(const char* data, size_t size) {
// whole number of pages to be written again on the next flush because we can
// only write on aligned
// offsets.
Status WritableFileWriter::WriteUnbuffered() {
Status WritableFileWriter::WriteDirect() {
Status s;
assert(!use_os_buffer_);
assert(direct_io_);
const size_t alignment = buf_.Alignment();
assert((next_write_offset_ % alignment) == 0);
@ -339,7 +339,7 @@ Status WritableFileWriter::WriteUnbuffered() {
{
IOSTATS_TIMER_GUARD(write_nanos);
TEST_SYNC_POINT("WritableFileWriter::Flush:BeforeAppend");
// Unbuffered writes must be positional
// direct writes must be positional
s = writable_file_->PositionedAppend(Slice(src, size), write_offset);
if (!s.ok()) {
buf_.Size(file_advance + leftover_tail);

6
util/file_reader_writer.h
View File

@ -103,7 +103,6 @@ class WritableFileWriter {
uint64_t next_write_offset_;
bool pending_sync_;
const bool direct_io_;
const bool use_os_buffer_;
uint64_t last_sync_size_;
uint64_t bytes_per_sync_;
RateLimiter* rate_limiter_;
@ -118,7 +117,6 @@ class WritableFileWriter {
next_write_offset_(0),
pending_sync_(false),
direct_io_(writable_file_->UseDirectIO()),
use_os_buffer_(writable_file_->UseOSBuffer()),
last_sync_size_(0),
bytes_per_sync_(options.bytes_per_sync),
rate_limiter_(options.rate_limiter) {
@ -156,8 +154,8 @@ class WritableFileWriter {
private:
// Used when os buffering is OFF and we are writing
// DMA such as in Windows unbuffered mode
Status WriteUnbuffered();
// DMA such as in Direct I/O mode
Status WriteDirect();
// Normal write
Status WriteBuffered(const char* data, size_t size);
Status RangeSync(uint64_t offset, uint64_t nbytes);

10
util/file_reader_writer_test.cc
View File

@ -88,9 +88,9 @@ TEST_F(WritableFileWriterTest, RangeSync) {
TEST_F(WritableFileWriterTest, AppendStatusReturn) {
class FakeWF : public WritableFile {
public:
explicit FakeWF() : use_os_buffer_(true), io_error_(false) {}
explicit FakeWF() : use_direct_io_(false), io_error_(false) {}
virtual bool UseOSBuffer() const override { return use_os_buffer_; }
virtual bool UseDirectIO() const override { return use_direct_io_; }
Status Append(const Slice& data) override {
if (io_error_) {
return Status::IOError("Fake IO error");
@ -106,15 +106,15 @@ TEST_F(WritableFileWriterTest, AppendStatusReturn) {
Status Close() override { return Status::OK(); }
Status Flush() override { return Status::OK(); }
Status Sync() override { return Status::OK(); }
void SetUseOSBuffer(bool val) { use_os_buffer_ = val; }
void SetUseDirectIO(bool val) { use_direct_io_ = val; }
void SetIOError(bool val) { io_error_ = val; }
protected:
bool use_os_buffer_;
bool use_direct_io_;
bool io_error_;
};
unique_ptr<FakeWF> wf(new FakeWF());
wf->SetUseOSBuffer(false);
wf->SetUseDirectIO(true);
unique_ptr<WritableFileWriter> writer(
new WritableFileWriter(std::move(wf), EnvOptions()));

48
util/io_posix.cc
View File

@ -165,7 +165,7 @@ PosixSequentialFile::PosixSequentialFile(const std::string& fname, FILE* f,
: filename_(fname),
file_(f),
fd_(fileno(f)),
use_os_buffer_(options.use_os_buffer) {}
use_direct_io_(options.use_direct_reads) {}
PosixSequentialFile::~PosixSequentialFile() { fclose(file_); }
@ -187,7 +187,7 @@ Status PosixSequentialFile::Read(size_t n, Slice* result, char* scratch) {
s = IOError(filename_, errno);
}
}
if (!use_os_buffer_) {
if (use_direct_io_) {
// we need to fadvise away the entire range of pages because
// we do not want readahead pages to be cached.
Fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED); // free OS pages
@ -294,7 +294,7 @@ size_t PosixHelper::GetUniqueIdFromFile(int fd, char* id, size_t max_size) {
*/
PosixRandomAccessFile::PosixRandomAccessFile(const std::string& fname, int fd,
const EnvOptions& options)
: filename_(fname), fd_(fd), use_os_buffer_(options.use_os_buffer) {
: filename_(fname), fd_(fd), use_direct_io_(options.use_direct_reads) {
assert(!options.use_mmap_reads || sizeof(void*) < 8);
}
@ -325,7 +325,8 @@ Status PosixRandomAccessFile::Read(uint64_t offset, size_t n, Slice* result,
// An error: return a non-ok status
s = IOError(filename_, errno);
}
if (!use_os_buffer_) {
if (use_direct_io_) {
// we need to fadvise away the entire range of pages because
// we do not want readahead pages to be cached.
Fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED); // free OS pages
@ -397,7 +398,7 @@ PosixMmapReadableFile::PosixMmapReadableFile(const int fd,
: fd_(fd), filename_(fname), mmapped_region_(base), length_(length) {
fd_ = fd_ + 0; // suppress the warning for used variables
assert(options.use_mmap_reads);
assert(options.use_os_buffer);
assert(!options.use_direct_reads);
}
PosixMmapReadableFile::~PosixMmapReadableFile() {
@ -533,6 +534,7 @@ PosixMmapFile::PosixMmapFile(const std::string& fname, int fd, size_t page_size,
#endif
assert((page_size & (page_size - 1)) == 0);
assert(options.use_mmap_writes);
assert(!options.use_direct_writes);
}
PosixMmapFile::~PosixMmapFile() {
@ -665,7 +667,10 @@ Status PosixMmapFile::Allocate(uint64_t offset, uint64_t len) {
*/
PosixWritableFile::PosixWritableFile(const std::string& fname, int fd,
const EnvOptions& options)
: filename_(fname), fd_(fd), filesize_(0) {
: filename_(fname),
direct_io_(options.use_direct_writes),
fd_(fd),
filesize_(0) {
#ifdef ROCKSDB_FALLOCATE_PRESENT
allow_fallocate_ = options.allow_fallocate;
fallocate_with_keep_size_ = options.fallocate_with_keep_size;
@ -680,6 +685,7 @@ PosixWritableFile::~PosixWritableFile() {
}
Status PosixWritableFile::Append(const Slice& data) {
assert(!direct_io_|| (IsSectorAligned(data.size()) && IsPageAligned(data.data())));
const char* src = data.data();
size_t left = data.size();
while (left != 0) {
@ -698,6 +704,8 @@ Status PosixWritableFile::Append(const Slice& data) {
}
Status PosixWritableFile::PositionedAppend(const Slice& data, uint64_t offset) {
assert(direct_io_ && IsSectorAligned(offset) &&
IsSectorAligned(data.size()) && IsPageAligned(data.data()));
assert(offset <= std::numeric_limits<off_t>::max());
const char* src = data.data();
size_t left = data.size();
@ -713,7 +721,7 @@ Status PosixWritableFile::PositionedAppend(const Slice& data, uint64_t offset) {
offset += done;
src += done;
}
filesize_ = offset + data.size();
filesize_ = offset;
return Status::OK();
}
@ -778,6 +786,9 @@ bool PosixWritableFile::IsSyncThreadSafe() const { return true; }
uint64_t PosixWritableFile::GetFileSize() { return filesize_; }
Status PosixWritableFile::InvalidateCache(size_t offset, size_t length) {
if (direct_io_) {
return Status::OK();
}
#ifndef OS_LINUX
return Status::OK();
#else
@ -825,29 +836,6 @@ size_t PosixWritableFile::GetUniqueId(char* id, size_t max_size) const {
}
#endif
/*
* PosixDirectIOWritableFile
*/
Status PosixDirectIOWritableFile::Append(const Slice& data) {
assert(IsSectorAligned(data.size()) && IsPageAligned(data.data()));
if (!IsSectorAligned(data.size()) || !IsPageAligned(data.data())) {
return Status::IOError("Unaligned buffer for direct IO");
}
return PosixWritableFile::Append(data);
}
Status PosixDirectIOWritableFile::PositionedAppend(const Slice& data,
uint64_t offset) {
assert(IsSectorAligned(offset));
assert(IsSectorAligned(data.size()));
assert(IsPageAligned(data.data()));
if (!IsSectorAligned(offset) || !IsSectorAligned(data.size()) ||
!IsPageAligned(data.data())) {
return Status::IOError("offset or size is not aligned");
}
return PosixWritableFile::PositionedAppend(data, offset);
}
/*
* PosixRandomRWFile
*/

28
util/io_posix.h
View File

@ -41,7 +41,7 @@ class PosixSequentialFile : public SequentialFile {
std::string filename_;
FILE* file_;
int fd_;
bool use_os_buffer_;
bool use_direct_io_;
public:
PosixSequentialFile(const std::string& fname, FILE* f,
@ -74,7 +74,7 @@ class PosixRandomAccessFile : public RandomAccessFile {
protected:
std::string filename_;
int fd_;
bool use_os_buffer_;
bool use_direct_io_;
public:
PosixRandomAccessFile(const std::string& fname, int fd,
@ -108,6 +108,7 @@ class PosixDirectIORandomAccessFile : public PosixRandomAccessFile {
class PosixWritableFile : public WritableFile {
protected:
const std::string filename_;
const bool direct_io_;
int fd_;
uint64_t filesize_;
#ifdef ROCKSDB_FALLOCATE_PRESENT
@ -130,7 +131,13 @@ class PosixWritableFile : public WritableFile {
virtual Status Sync() override;
virtual Status Fsync() override;
virtual bool IsSyncThreadSafe() const override;
virtual bool UseDirectIO() const override { return direct_io_; }
virtual uint64_t GetFileSize() override;
virtual size_t GetRequiredBufferAlignment() const override {
// TODO(gzh): It should be the logical sector size/filesystem block size
// hardcoded as 4k for most cases
return 4 * 1024;
}
virtual Status InvalidateCache(size_t offset, size_t length) override;
#ifdef ROCKSDB_FALLOCATE_PRESENT
virtual Status Allocate(uint64_t offset, uint64_t len) override;
@ -139,22 +146,7 @@ class PosixWritableFile : public WritableFile {
#endif
};
class PosixDirectIOWritableFile : public PosixWritableFile {
public:
explicit PosixDirectIOWritableFile(const std::string& filename, int fd)
: PosixWritableFile(filename, fd, EnvOptions()) {}
virtual ~PosixDirectIOWritableFile() {}
bool UseOSBuffer() const override { return false; }
size_t GetRequiredBufferAlignment() const override { return 4 * 1024; }
Status Append(const Slice& data) override;
Status PositionedAppend(const Slice& data, uint64_t offset) override;
bool UseDirectIO() const override { return true; }
Status InvalidateCache(size_t offset, size_t length) override {
return Status::OK();
}
};
// mmap() based random-access
class PosixMmapReadableFile : public RandomAccessFile {
private:
int fd_;

3
util/log_write_bench.cc
View File

@ -34,8 +34,7 @@ namespace rocksdb {
void RunBenchmark() {
std::string file_name = test::TmpDir() + "/log_write_benchmark.log";
Env* env = Env::Default();
EnvOptions env_options;
env_options.use_mmap_writes = false;
EnvOptions env_options = env->OptimizeForLogWrite(EnvOptions());
env_options.bytes_per_sync = FLAGS_bytes_per_sync;
unique_ptr<WritableFile> file;
env->NewWritableFile(file_name, &file, env_options);

4
util/memenv.cc
View File

@ -221,9 +221,7 @@ class RandomAccessFileImpl : public RandomAccessFile {
class WritableFileImpl : public WritableFile {
public:
WritableFileImpl(FileState* file) : file_(file) {
file_->Ref();
}
explicit WritableFileImpl(FileState* file) : file_(file) { file_->Ref(); }
~WritableFileImpl() {
file_->Unref();

14
util/options.cc
View File

@ -198,10 +198,10 @@ DBOptions::DBOptions()
WAL_ttl_seconds(0),
WAL_size_limit_MB(0),
manifest_preallocation_size(4 * 1024 * 1024),
allow_os_buffer(true),
allow_mmap_reads(false),
allow_mmap_writes(false),
use_direct_reads(false),
use_direct_writes(false),
allow_fallocate(true),
is_fd_close_on_exec(true),
skip_log_error_on_recovery(false),
@ -269,10 +269,10 @@ DBOptions::DBOptions(const Options& options)
WAL_ttl_seconds(options.WAL_ttl_seconds),
WAL_size_limit_MB(options.WAL_size_limit_MB),
manifest_preallocation_size(options.manifest_preallocation_size),
allow_os_buffer(options.allow_os_buffer),
allow_mmap_reads(options.allow_mmap_reads),
allow_mmap_writes(options.allow_mmap_writes),
use_direct_reads(options.use_direct_reads),
use_direct_writes(options.use_direct_writes),
allow_fallocate(options.allow_fallocate),
is_fd_close_on_exec(options.is_fd_close_on_exec),
skip_log_error_on_recovery(options.skip_log_error_on_recovery),
@ -336,11 +336,11 @@ void DBOptions::Dump(Logger* log) const {
keep_log_file_num);
Header(log, " Options.recycle_log_file_num: %" ROCKSDB_PRIszt,
recycle_log_file_num);
Header(log, " Options.allow_os_buffer: %d", allow_os_buffer);
Header(log, " Options.allow_mmap_reads: %d", allow_mmap_reads);
Header(log, " Options.allow_fallocate: %d", allow_fallocate);
Header(log, " Options.allow_mmap_reads: %d", allow_mmap_reads);
Header(log, " Options.allow_mmap_writes: %d", allow_mmap_writes);
Header(log, " Options.use_direct_reads: %d", use_direct_reads);
Header(log, " Options.use_direct_writes: %d", use_direct_writes);
Header(log, " Options.create_missing_column_families: %d",
create_missing_column_families);
Header(log, " Options.db_log_dir: %s",
@ -366,12 +366,6 @@ void DBOptions::Dump(Logger* log) const {
Header(log,
" Options.manifest_preallocation_size: %" ROCKSDB_PRIszt,
manifest_preallocation_size);
Header(log, " Options.allow_os_buffer: %d",
allow_os_buffer);
Header(log, " Options.allow_mmap_reads: %d",
allow_mmap_reads);
Header(log, " Options.allow_mmap_writes: %d",
allow_mmap_writes);
Header(log, " Options.is_fd_close_on_exec: %d",
is_fd_close_on_exec);
Header(log, " Options.stats_dump_period_sec: %u",

2
util/options_helper.cc
View File

@ -71,10 +71,10 @@ DBOptions BuildDBOptions(const ImmutableDBOptions& immutable_db_options,
options.WAL_size_limit_MB = immutable_db_options.wal_size_limit_mb;
options.manifest_preallocation_size =
immutable_db_options.manifest_preallocation_size;
options.allow_os_buffer = immutable_db_options.allow_os_buffer;
options.allow_mmap_reads = immutable_db_options.allow_mmap_reads;
options.allow_mmap_writes = immutable_db_options.allow_mmap_writes;
options.use_direct_reads = immutable_db_options.use_direct_reads;
options.use_direct_writes = immutable_db_options.use_direct_writes;
options.allow_fallocate = immutable_db_options.allow_fallocate;
options.is_fd_close_on_exec = immutable_db_options.is_fd_close_on_exec;
options.stats_dump_period_sec = immutable_db_options.stats_dump_period_sec;

6
util/options_helper.h
View File

@ -186,12 +186,14 @@ static std::unordered_map<std::string, OptionTypeInfo> db_options_type_info = {
{"use_direct_reads",
{offsetof(struct DBOptions, use_direct_reads), OptionType::kBoolean,
OptionVerificationType::kNormal, false, 0}},
{"use_direct_writes",
{offsetof(struct DBOptions, use_direct_writes), OptionType::kBoolean,
OptionVerificationType::kNormal, false, 0}},
{"allow_2pc",
{offsetof(struct DBOptions, allow_2pc), OptionType::kBoolean,
OptionVerificationType::kNormal, false, 0}},
{"allow_os_buffer",
{offsetof(struct DBOptions, allow_os_buffer), OptionType::kBoolean,
OptionVerificationType::kNormal, false, 0}},
{0, OptionType::kBoolean, OptionVerificationType::kDeprecated, true, 0}},
{"create_if_missing",
{offsetof(struct DBOptions, create_if_missing), OptionType::kBoolean,
OptionVerificationType::kNormal, false, 0}},

2
util/options_settable_test.cc
View File

@ -267,7 +267,6 @@ TEST_F(OptionsSettableTest, DBOptionsAllFieldsSettable) {
"max_background_flushes=35;"
"create_if_missing=false;"
"error_if_exists=true;"
"allow_os_buffer=false;"
"delayed_write_rate=4294976214;"
"manifest_preallocation_size=1222;"
"allow_mmap_writes=false;"
@ -275,6 +274,7 @@ TEST_F(OptionsSettableTest, DBOptionsAllFieldsSettable) {
"allow_fallocate=true;"
"allow_mmap_reads=false;"
"use_direct_reads=false;"
"use_direct_writes=false;"
"max_log_file_size=4607;"
"random_access_max_buffer_size=1048576;"
"advise_random_on_open=true;"

6
util/options_test.cc
View File

@ -116,9 +116,10 @@ TEST_F(OptionsTest, GetOptionsFromMapTest) {
{"WAL_ttl_seconds", "43"},
{"WAL_size_limit_MB", "44"},
{"manifest_preallocation_size", "45"},
{"allow_os_buffer", "false"},
{"allow_mmap_reads", "true"},
{"allow_mmap_writes", "false"},
{"use_direct_reads", "false"},
{"use_direct_writes", "false"},
{"is_fd_close_on_exec", "true"},
{"skip_log_error_on_recovery", "false"},
{"stats_dump_period_sec", "46"},
@ -231,9 +232,10 @@ TEST_F(OptionsTest, GetOptionsFromMapTest) {
ASSERT_EQ(new_db_opt.WAL_ttl_seconds, static_cast<uint64_t>(43));
ASSERT_EQ(new_db_opt.WAL_size_limit_MB, static_cast<uint64_t>(44));
ASSERT_EQ(new_db_opt.manifest_preallocation_size, 45U);
ASSERT_EQ(new_db_opt.allow_os_buffer, false);
ASSERT_EQ(new_db_opt.allow_mmap_reads, true);
ASSERT_EQ(new_db_opt.allow_mmap_writes, false);
ASSERT_EQ(new_db_opt.use_direct_reads, false);
ASSERT_EQ(new_db_opt.use_direct_writes, false);
ASSERT_EQ(new_db_opt.is_fd_close_on_exec, true);
ASSERT_EQ(new_db_opt.skip_log_error_on_recovery, false);
ASSERT_EQ(new_db_opt.stats_dump_period_sec, 46U);

3
util/testutil.cc
View File

@ -240,7 +240,8 @@ void RandomInitDBOptions(DBOptions* db_opt, Random* rnd) {
db_opt->advise_random_on_open = rnd->Uniform(2);
db_opt->allow_mmap_reads = rnd->Uniform(2);
db_opt->allow_mmap_writes = rnd->Uniform(2);
db_opt->allow_os_buffer = rnd->Uniform(2);
db_opt->use_direct_reads = rnd->Uniform(2);
db_opt->use_direct_writes = rnd->Uniform(2);
db_opt->create_if_missing = rnd->Uniform(2);
db_opt->create_missing_column_families = rnd->Uniform(2);
db_opt->disableDataSync = rnd->Uniform(2);

5
utilities/backupable/backupable_db.cc
View File

@ -1162,7 +1162,7 @@ Status BackupEngineImpl::CopyOrCreateFile(
unique_ptr<SequentialFile> src_file;
EnvOptions env_options;
env_options.use_mmap_writes = false;
env_options.use_os_buffer = false;
// TODO:(gzh) maybe use direct writes here if possible
if (size != nullptr) {
*size = 0;
}
@ -1365,7 +1365,7 @@ Status BackupEngineImpl::CalculateChecksum(const std::string& src, Env* src_env,
EnvOptions env_options;
env_options.use_mmap_writes = false;
env_options.use_os_buffer = false;
env_options.use_direct_reads = false;
std::unique_ptr<SequentialFile> src_file;
Status s = src_env->NewSequentialFile(src, &src_file, env_options);
@ -1671,6 +1671,7 @@ Status BackupEngineImpl::BackupMeta::StoreToFile(bool sync) {
unique_ptr<WritableFile> backup_meta_file;
EnvOptions env_options;
env_options.use_mmap_writes = false;
env_options.use_direct_writes = false;
s = env_->NewWritableFile(meta_filename_ + ".tmp", &backup_meta_file,
env_options);
if (!s.ok()) {

6
utilities/env_mirror.cc
View File

@ -18,7 +18,7 @@ class SequentialFileMirror : public SequentialFile {
public:
unique_ptr<SequentialFile> a_, b_;
std::string fname;
SequentialFileMirror(std::string f) : fname(f) {}
explicit SequentialFileMirror(std::string f) : fname(f) {}
Status Read(size_t n, Slice* result, char* scratch) {
Slice aslice;
@ -62,7 +62,7 @@ class RandomAccessFileMirror : public RandomAccessFile {
public:
unique_ptr<RandomAccessFile> a_, b_;
std::string fname;
RandomAccessFileMirror(std::string f) : fname(f) {}
explicit RandomAccessFileMirror(std::string f) : fname(f) {}
Status Read(uint64_t offset, size_t n, Slice* result, char* scratch) const {
Status as = a_->Read(offset, n, result, scratch);
@ -101,7 +101,7 @@ class WritableFileMirror : public WritableFile {
public:
unique_ptr<WritableFile> a_, b_;
std::string fname;
WritableFileMirror(std::string f) : fname(f) {}
explicit WritableFileMirror(std::string f) : fname(f) {}
Status Append(const Slice& data) override {
Status as = a_->Append(data);