rocksdb/env/io_posix.cc

1161 lines
32 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#ifdef ROCKSDB_LIB_IO_POSIX
#include "env/io_posix.h"
#include <errno.h>
#include <fcntl.h>
#include <algorithm>
#if defined(OS_LINUX)
#include <linux/fs.h>
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/types.h>
#ifdef OS_LINUX
#include <sys/statfs.h>
#include <sys/syscall.h>
#include <sys/sysmacros.h>
#endif
#include "env/posix_logger.h"
#include "monitoring/iostats_context_imp.h"
#include "port/port.h"
#include "rocksdb/slice.h"
#include "util/coding.h"
#include "util/string_util.h"
#include "util/sync_point.h"
#if defined(OS_LINUX) && !defined(F_SET_RW_HINT)
#define F_LINUX_SPECIFIC_BASE 1024
#define F_SET_RW_HINT (F_LINUX_SPECIFIC_BASE + 12)
#endif
namespace rocksdb {
// A wrapper for fadvise, if the platform doesn't support fadvise,
// it will simply return 0.
int Fadvise(int fd, off_t offset, size_t len, int advice) {
#ifdef OS_LINUX
return posix_fadvise(fd, offset, len, advice);
#else
(void)fd;
(void)offset;
(void)len;
(void)advice;
return 0; // simply do nothing.
#endif
}
namespace {
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
// On MacOS (and probably *BSD), the posix write and pwrite calls do not support
// buffers larger than 2^31-1 bytes. These two wrappers fix this issue by
// cutting the buffer in 1GB chunks. We use this chunk size to be sure to keep
// the writes aligned.
bool PosixWrite(int fd, const char* buf, size_t nbyte) {
const size_t kLimit1Gb = 1UL << 30;
const char* src = buf;
size_t left = nbyte;
while (left != 0) {
size_t bytes_to_write = std::min(left, kLimit1Gb);
ssize_t done = write(fd, src, bytes_to_write);
if (done < 0) {
if (errno == EINTR) {
continue;
}
return false;
}
left -= done;
src += done;
}
return true;
}
bool PosixPositionedWrite(int fd, const char* buf, size_t nbyte, off_t offset) {
const size_t kLimit1Gb = 1UL << 30;
const char* src = buf;
size_t left = nbyte;
while (left != 0) {
size_t bytes_to_write = std::min(left, kLimit1Gb);
ssize_t done = pwrite(fd, src, bytes_to_write, offset);
if (done < 0) {
if (errno == EINTR) {
continue;
}
return false;
}
left -= done;
offset += done;
src += done;
}
return true;
}
size_t GetLogicalBufferSize(int __attribute__((__unused__)) fd) {
#ifdef OS_LINUX
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return kDefaultPageSize;
}
if (major(buf.st_dev) == 0) {
// Unnamed devices (e.g. non-device mounts), reserved as null device number.
// These don't have an entry in /sys/dev/block/. Return a sensible default.
return kDefaultPageSize;
}
// Reading queue/logical_block_size does not require special permissions.
const int kBufferSize = 100;
char path[kBufferSize];
char real_path[PATH_MAX + 1];
snprintf(path, kBufferSize, "/sys/dev/block/%u:%u", major(buf.st_dev),
minor(buf.st_dev));
if (realpath(path, real_path) == nullptr) {
return kDefaultPageSize;
}
std::string device_dir(real_path);
if (!device_dir.empty() && device_dir.back() == '/') {
device_dir.pop_back();
}
// NOTE: sda3 and nvme0n1p1 do not have a `queue/` subdir, only the parent sda
// and nvme0n1 have it.
// $ ls -al '/sys/dev/block/8:3'
// lrwxrwxrwx. 1 root root 0 Jun 26 01:38 /sys/dev/block/8:3 ->
// ../../block/sda/sda3
// $ ls -al '/sys/dev/block/259:4'
// lrwxrwxrwx 1 root root 0 Jan 31 16:04 /sys/dev/block/259:4 ->
// ../../devices/pci0000:17/0000:17:00.0/0000:18:00.0/nvme/nvme0/nvme0n1/nvme0n1p1
size_t parent_end = device_dir.rfind('/', device_dir.length() - 1);
if (parent_end == std::string::npos) {
return kDefaultPageSize;
}
size_t parent_begin = device_dir.rfind('/', parent_end - 1);
if (parent_begin == std::string::npos) {
return kDefaultPageSize;
}
std::string parent =
device_dir.substr(parent_begin + 1, parent_end - parent_begin - 1);
std::string child = device_dir.substr(parent_end + 1, std::string::npos);
if (parent != "block" &&
(child.compare(0, 4, "nvme") || child.find('p') != std::string::npos)) {
device_dir = device_dir.substr(0, parent_end);
}
std::string fname = device_dir + "/queue/logical_block_size";
FILE* fp;
size_t size = 0;
fp = fopen(fname.c_str(), "r");
if (fp != nullptr) {
char* line = nullptr;
size_t len = 0;
if (getline(&line, &len, fp) != -1) {
sscanf(line, "%zu", &size);
}
free(line);
fclose(fp);
}
if (size != 0 && (size & (size - 1)) == 0) {
return size;
}
#endif
return kDefaultPageSize;
}
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
#if !defined(ZFS_SUPER_MAGIC)
// The magic number for ZFS was not exposed until recently. It should be fixed
// forever so we can just copy the magic number here.
#define ZFS_SUPER_MAGIC 0x2fc12fc1
#endif
bool IsSyncFileRangeSupported(int __attribute__((__unused__)) fd) {
// `fstatfs` is only available on Linux, but so is `sync_file_range`, so
// `defined(ROCKSDB_RANGESYNC_PRESENT)` should imply `defined(OS_LINUX)`.
struct statfs buf;
int ret = fstatfs(fd, &buf);
assert(ret == 0);
if (ret != 0) {
// We don't know whether the filesystem properly supports `sync_file_range`.
// Even if it doesn't, we don't know of any safety issue with trying to call
// it anyways. So, to preserve the same behavior as before this `fstatfs`
// check was introduced, we assume `sync_file_range` is usable.
return true;
}
if (buf.f_type == ZFS_SUPER_MAGIC) {
// Testing on ZFS showed the writeback did not happen asynchronously when
// `sync_file_range` was called, even though it returned success. Avoid it
// and use `fdatasync` instead to preserve the contract of `bytes_per_sync`,
// even though this'll incur extra I/O for metadata.
return false;
}
// No known problems with other filesystems' implementations of
// `sync_file_range`, so allow them to use it.
return true;
}
#undef ZFS_SUPER_MAGIC
#endif // ROCKSDB_RANGESYNC_PRESENT
} // anonymous namespace
/*
* DirectIOHelper
*/
#ifndef NDEBUG
namespace {
bool IsSectorAligned(const size_t off, size_t sector_size) {
return off % sector_size == 0;
}
bool IsSectorAligned(const void* ptr, size_t sector_size) {
return uintptr_t(ptr) % sector_size == 0;
}
} // namespace
#endif
/*
* PosixSequentialFile
*/
PosixSequentialFile::PosixSequentialFile(const std::string& fname, FILE* file,
int fd, const EnvOptions& options)
: filename_(fname),
file_(file),
fd_(fd),
use_direct_io_(options.use_direct_reads),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
assert(!options.use_direct_reads || !options.use_mmap_reads);
}
PosixSequentialFile::~PosixSequentialFile() {
if (!use_direct_io()) {
assert(file_);
fclose(file_);
} else {
assert(fd_);
close(fd_);
}
}
Status PosixSequentialFile::Read(size_t n, Slice* result, char* scratch) {
assert(result != nullptr && !use_direct_io());
Status s;
size_t r = 0;
do {
r = fread_unlocked(scratch, 1, n, file_);
} while (r == 0 && ferror(file_) && errno == EINTR);
*result = Slice(scratch, r);
if (r < n) {
if (feof(file_)) {
// We leave status as ok if we hit the end of the file
// We also clear the error so that the reads can continue
// if a new data is written to the file
clearerr(file_);
} else {
// A partial read with an error: return a non-ok status
s = IOError("While reading file sequentially", filename_, errno);
}
}
return s;
}
Status PosixSequentialFile::PositionedRead(uint64_t offset, size_t n,
Slice* result, char* scratch) {
assert(use_direct_io());
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(n, GetRequiredBufferAlignment()));
assert(IsSectorAligned(scratch, GetRequiredBufferAlignment()));
Status s;
ssize_t r = -1;
size_t left = n;
char* ptr = scratch;
while (left > 0) {
r = pread(fd_, ptr, left, static_cast<off_t>(offset));
if (r <= 0) {
if (r == -1 && errno == EINTR) {
continue;
}
break;
}
ptr += r;
offset += r;
left -= r;
if (r % static_cast<ssize_t>(GetRequiredBufferAlignment()) != 0) {
// Bytes reads don't fill sectors. Should only happen at the end
// of the file.
break;
}
}
if (r < 0) {
// An error: return a non-ok status
s = IOError(
"While pread " + ToString(n) + " bytes from offset " + ToString(offset),
filename_, errno);
}
*result = Slice(scratch, (r < 0) ? 0 : n - left);
return s;
}
Status PosixSequentialFile::Skip(uint64_t n) {
if (fseek(file_, static_cast<long int>(n), SEEK_CUR)) {
return IOError("While fseek to skip " + ToString(n) + " bytes", filename_,
errno);
}
return Status::OK();
}
Status PosixSequentialFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
if (!use_direct_io()) {
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret != 0) {
return IOError("While fadvise NotNeeded offset " + ToString(offset) +
" len " + ToString(length),
filename_, errno);
}
}
return Status::OK();
#endif
}
/*
* PosixRandomAccessFile
*/
#if defined(OS_LINUX)
size_t PosixHelper::GetUniqueIdFromFile(int fd, char* id, size_t max_size) {
if (max_size < kMaxVarint64Length * 3) {
return 0;
}
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return 0;
}
long version = 0;
result = ioctl(fd, FS_IOC_GETVERSION, &version);
TEST_SYNC_POINT_CALLBACK("GetUniqueIdFromFile:FS_IOC_GETVERSION", &result);
if (result == -1) {
return 0;
}
uint64_t uversion = (uint64_t)version;
char* rid = id;
rid = EncodeVarint64(rid, buf.st_dev);
rid = EncodeVarint64(rid, buf.st_ino);
rid = EncodeVarint64(rid, uversion);
assert(rid >= id);
return static_cast<size_t>(rid - id);
}
#endif
#if defined(OS_MACOSX) || defined(OS_AIX)
size_t PosixHelper::GetUniqueIdFromFile(int fd, char* id, size_t max_size) {
if (max_size < kMaxVarint64Length * 3) {
return 0;
}
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return 0;
}
char* rid = id;
rid = EncodeVarint64(rid, buf.st_dev);
rid = EncodeVarint64(rid, buf.st_ino);
rid = EncodeVarint64(rid, buf.st_gen);
assert(rid >= id);
return static_cast<size_t>(rid - id);
}
#endif
/*
* PosixRandomAccessFile
*
* pread() based random-access
*/
PosixRandomAccessFile::PosixRandomAccessFile(const std::string& fname, int fd,
const EnvOptions& options)
: filename_(fname),
fd_(fd),
use_direct_io_(options.use_direct_reads),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
assert(!options.use_direct_reads || !options.use_mmap_reads);
assert(!options.use_mmap_reads || sizeof(void*) < 8);
}
PosixRandomAccessFile::~PosixRandomAccessFile() { close(fd_); }
Status PosixRandomAccessFile::Read(uint64_t offset, size_t n, Slice* result,
char* scratch) const {
if (use_direct_io()) {
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(n, GetRequiredBufferAlignment()));
assert(IsSectorAligned(scratch, GetRequiredBufferAlignment()));
}
Status s;
ssize_t r = -1;
size_t left = n;
char* ptr = scratch;
while (left > 0) {
r = pread(fd_, ptr, left, static_cast<off_t>(offset));
if (r <= 0) {
if (r == -1 && errno == EINTR) {
continue;
}
break;
}
ptr += r;
offset += r;
left -= r;
if (use_direct_io() &&
r % static_cast<ssize_t>(GetRequiredBufferAlignment()) != 0) {
// Bytes reads don't fill sectors. Should only happen at the end
// of the file.
break;
}
}
if (r < 0) {
// An error: return a non-ok status
s = IOError(
"While pread offset " + ToString(offset) + " len " + ToString(n),
filename_, errno);
}
*result = Slice(scratch, (r < 0) ? 0 : n - left);
return s;
}
Status PosixRandomAccessFile::Prefetch(uint64_t offset, size_t n) {
Status s;
if (!use_direct_io()) {
ssize_t r = 0;
#ifdef OS_LINUX
r = readahead(fd_, offset, n);
#endif
#ifdef OS_MACOSX
radvisory advice;
advice.ra_offset = static_cast<off_t>(offset);
advice.ra_count = static_cast<int>(n);
r = fcntl(fd_, F_RDADVISE, &advice);
#endif
if (r == -1) {
s = IOError("While prefetching offset " + ToString(offset) + " len " +
ToString(n),
filename_, errno);
}
}
return s;
}
#if defined(OS_LINUX) || defined(OS_MACOSX) || defined(OS_AIX)
size_t PosixRandomAccessFile::GetUniqueId(char* id, size_t max_size) const {
return PosixHelper::GetUniqueIdFromFile(fd_, id, max_size);
}
#endif
void PosixRandomAccessFile::Hint(AccessPattern pattern) {
if (use_direct_io()) {
return;
}
switch (pattern) {
case NORMAL:
Fadvise(fd_, 0, 0, POSIX_FADV_NORMAL);
break;
case RANDOM:
Fadvise(fd_, 0, 0, POSIX_FADV_RANDOM);
break;
case SEQUENTIAL:
Fadvise(fd_, 0, 0, POSIX_FADV_SEQUENTIAL);
break;
case WILLNEED:
Fadvise(fd_, 0, 0, POSIX_FADV_WILLNEED);
break;
case DONTNEED:
Fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED);
break;
default:
assert(false);
break;
}
}
Status PosixRandomAccessFile::InvalidateCache(size_t offset, size_t length) {
if (use_direct_io()) {
return Status::OK();
}
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded offset " + ToString(offset) +
" len " + ToString(length),
filename_, errno);
#endif
}
/*
* PosixMmapReadableFile
*
* mmap() based random-access
*/
// base[0,length-1] contains the mmapped contents of the file.
PosixMmapReadableFile::PosixMmapReadableFile(const int fd,
const std::string& fname,
void* base, size_t length,
const EnvOptions& options)
: fd_(fd), filename_(fname), mmapped_region_(base), length_(length) {
#ifdef NDEBUG
(void)options;
#endif
fd_ = fd_ + 0; // suppress the warning for used variables
assert(options.use_mmap_reads);
assert(!options.use_direct_reads);
}
PosixMmapReadableFile::~PosixMmapReadableFile() {
int ret = munmap(mmapped_region_, length_);
if (ret != 0) {
fprintf(stdout, "failed to munmap %p length %" ROCKSDB_PRIszt " \n",
mmapped_region_, length_);
}
close(fd_);
}
Status PosixMmapReadableFile::Read(uint64_t offset, size_t n, Slice* result,
char* /*scratch*/) const {
Status s;
if (offset > length_) {
*result = Slice();
return IOError("While mmap read offset " + ToString(offset) +
" larger than file length " + ToString(length_),
filename_, EINVAL);
} else if (offset + n > length_) {
n = static_cast<size_t>(length_ - offset);
}
*result = Slice(reinterpret_cast<char*>(mmapped_region_) + offset, n);
return s;
}
Status PosixMmapReadableFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise not needed. Offset " + ToString(offset) +
" len" + ToString(length),
filename_, errno);
#endif
}
/*
* PosixMmapFile
*
* We preallocate up to an extra megabyte and use memcpy to append new
* data to the file. This is safe since we either properly close the
* file before reading from it, or for log files, the reading code
* knows enough to skip zero suffixes.
*/
Status PosixMmapFile::UnmapCurrentRegion() {
TEST_KILL_RANDOM("PosixMmapFile::UnmapCurrentRegion:0", rocksdb_kill_odds);
if (base_ != nullptr) {
int munmap_status = munmap(base_, limit_ - base_);
if (munmap_status != 0) {
return IOError("While munmap", filename_, munmap_status);
}
file_offset_ += limit_ - base_;
base_ = nullptr;
limit_ = nullptr;
last_sync_ = nullptr;
dst_ = nullptr;
// Increase the amount we map the next time, but capped at 1MB
if (map_size_ < (1 << 20)) {
map_size_ *= 2;
}
}
return Status::OK();
}
Status PosixMmapFile::MapNewRegion() {
#ifdef ROCKSDB_FALLOCATE_PRESENT
assert(base_ == nullptr);
TEST_KILL_RANDOM("PosixMmapFile::UnmapCurrentRegion:0", rocksdb_kill_odds);
// we can't fallocate with FALLOC_FL_KEEP_SIZE here
if (allow_fallocate_) {
IOSTATS_TIMER_GUARD(allocate_nanos);
int alloc_status = fallocate(fd_, 0, file_offset_, map_size_);
if (alloc_status != 0) {
// fallback to posix_fallocate
alloc_status = posix_fallocate(fd_, file_offset_, map_size_);
}
if (alloc_status != 0) {
return Status::IOError("Error allocating space to file : " + filename_ +
"Error : " + strerror(alloc_status));
}
}
TEST_KILL_RANDOM("PosixMmapFile::Append:1", rocksdb_kill_odds);
void* ptr = mmap(nullptr, map_size_, PROT_READ | PROT_WRITE, MAP_SHARED, fd_,
file_offset_);
if (ptr == MAP_FAILED) {
return Status::IOError("MMap failed on " + filename_);
}
TEST_KILL_RANDOM("PosixMmapFile::Append:2", rocksdb_kill_odds);
base_ = reinterpret_cast<char*>(ptr);
limit_ = base_ + map_size_;
dst_ = base_;
last_sync_ = base_;
return Status::OK();
#else
return Status::NotSupported("This platform doesn't support fallocate()");
#endif
}
Status PosixMmapFile::Msync() {
if (dst_ == last_sync_) {
return Status::OK();
}
// Find the beginnings of the pages that contain the first and last
// bytes to be synced.
size_t p1 = TruncateToPageBoundary(last_sync_ - base_);
size_t p2 = TruncateToPageBoundary(dst_ - base_ - 1);
last_sync_ = dst_;
TEST_KILL_RANDOM("PosixMmapFile::Msync:0", rocksdb_kill_odds);
if (msync(base_ + p1, p2 - p1 + page_size_, MS_SYNC) < 0) {
return IOError("While msync", filename_, errno);
}
return Status::OK();
}
PosixMmapFile::PosixMmapFile(const std::string& fname, int fd, size_t page_size,
const EnvOptions& options)
: filename_(fname),
fd_(fd),
page_size_(page_size),
map_size_(Roundup(65536, page_size)),
base_(nullptr),
limit_(nullptr),
dst_(nullptr),
last_sync_(nullptr),
file_offset_(0) {
#ifdef ROCKSDB_FALLOCATE_PRESENT
allow_fallocate_ = options.allow_fallocate;
fallocate_with_keep_size_ = options.fallocate_with_keep_size;
#else
(void)options;
#endif
assert((page_size & (page_size - 1)) == 0);
assert(options.use_mmap_writes);
assert(!options.use_direct_writes);
}
PosixMmapFile::~PosixMmapFile() {
if (fd_ >= 0) {
PosixMmapFile::Close();
}
}
Status PosixMmapFile::Append(const Slice& data) {
const char* src = data.data();
size_t left = data.size();
while (left > 0) {
assert(base_ <= dst_);
assert(dst_ <= limit_);
size_t avail = limit_ - dst_;
if (avail == 0) {
Status s = UnmapCurrentRegion();
if (!s.ok()) {
return s;
}
s = MapNewRegion();
if (!s.ok()) {
return s;
}
TEST_KILL_RANDOM("PosixMmapFile::Append:0", rocksdb_kill_odds);
}
size_t n = (left <= avail) ? left : avail;
assert(dst_);
memcpy(dst_, src, n);
dst_ += n;
src += n;
left -= n;
}
return Status::OK();
}
Status PosixMmapFile::Close() {
Status s;
size_t unused = limit_ - dst_;
s = UnmapCurrentRegion();
if (!s.ok()) {
s = IOError("While closing mmapped file", filename_, errno);
} else if (unused > 0) {
// Trim the extra space at the end of the file
if (ftruncate(fd_, file_offset_ - unused) < 0) {
s = IOError("While ftruncating mmaped file", filename_, errno);
}
}
if (close(fd_) < 0) {
if (s.ok()) {
s = IOError("While closing mmapped file", filename_, errno);
}
}
fd_ = -1;
base_ = nullptr;
limit_ = nullptr;
return s;
}
Status PosixMmapFile::Flush() { return Status::OK(); }
Status PosixMmapFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync mmapped file", filename_, errno);
}
return Msync();
}
/**
* Flush data as well as metadata to stable storage.
*/
Status PosixMmapFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync mmaped file", filename_, errno);
}
return Msync();
}
/**
* Get the size of valid data in the file. This will not match the
* size that is returned from the filesystem because we use mmap
* to extend file by map_size every time.
*/
uint64_t PosixMmapFile::GetFileSize() {
size_t used = dst_ - base_;
return file_offset_ + used;
}
Status PosixMmapFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded mmapped file", filename_, errno);
#endif
}
#ifdef ROCKSDB_FALLOCATE_PRESENT
Status PosixMmapFile::Allocate(uint64_t offset, uint64_t len) {
assert(offset <= std::numeric_limits<off_t>::max());
assert(len <= std::numeric_limits<off_t>::max());
TEST_KILL_RANDOM("PosixMmapFile::Allocate:0", rocksdb_kill_odds);
int alloc_status = 0;
if (allow_fallocate_) {
alloc_status =
fallocate(fd_, fallocate_with_keep_size_ ? FALLOC_FL_KEEP_SIZE : 0,
static_cast<off_t>(offset), static_cast<off_t>(len));
}
if (alloc_status == 0) {
return Status::OK();
} else {
return IOError(
"While fallocate offset " + ToString(offset) + " len " + ToString(len),
filename_, errno);
}
}
#endif
/*
* PosixWritableFile
*
* Use posix write to write data to a file.
*/
PosixWritableFile::PosixWritableFile(const std::string& fname, int fd,
const EnvOptions& options)
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
: WritableFile(options),
filename_(fname),
use_direct_io_(options.use_direct_writes),
fd_(fd),
filesize_(0),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
#ifdef ROCKSDB_FALLOCATE_PRESENT
allow_fallocate_ = options.allow_fallocate;
fallocate_with_keep_size_ = options.fallocate_with_keep_size;
#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
sync_file_range_supported_ = IsSyncFileRangeSupported(fd_);
#endif // ROCKSDB_RANGESYNC_PRESENT
assert(!options.use_mmap_writes);
}
PosixWritableFile::~PosixWritableFile() {
if (fd_ >= 0) {
PosixWritableFile::Close();
}
}
Status PosixWritableFile::Append(const Slice& data) {
if (use_direct_io()) {
assert(IsSectorAligned(data.size(), GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.data(), GetRequiredBufferAlignment()));
}
const char* src = data.data();
size_t nbytes = data.size();
if (!PosixWrite(fd_, src, nbytes)) {
return IOError("While appending to file", filename_, errno);
}
filesize_ += nbytes;
return Status::OK();
}
Status PosixWritableFile::PositionedAppend(const Slice& data, uint64_t offset) {
if (use_direct_io()) {
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.size(), GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.data(), GetRequiredBufferAlignment()));
}
assert(offset <= std::numeric_limits<off_t>::max());
const char* src = data.data();
size_t nbytes = data.size();
if (!PosixPositionedWrite(fd_, src, nbytes, static_cast<off_t>(offset))) {
return IOError("While pwrite to file at offset " + ToString(offset),
filename_, errno);
}
filesize_ = offset + nbytes;
return Status::OK();
}
Status PosixWritableFile::Truncate(uint64_t size) {
Status s;
int r = ftruncate(fd_, size);
if (r < 0) {
s = IOError("While ftruncate file to size " + ToString(size), filename_,
errno);
} else {
filesize_ = size;
}
return s;
}
Status PosixWritableFile::Close() {
Status s;
size_t block_size;
size_t last_allocated_block;
GetPreallocationStatus(&block_size, &last_allocated_block);
if (last_allocated_block > 0) {
// trim the extra space preallocated at the end of the file
// NOTE(ljin): we probably don't want to surface failure as an IOError,
// but it will be nice to log these errors.
int dummy __attribute__((__unused__));
dummy = ftruncate(fd_, filesize_);
#if defined(ROCKSDB_FALLOCATE_PRESENT) && defined(FALLOC_FL_PUNCH_HOLE) && \
!defined(TRAVIS)
// in some file systems, ftruncate only trims trailing space if the
// new file size is smaller than the current size. Calling fallocate
// with FALLOC_FL_PUNCH_HOLE flag to explicitly release these unused
// blocks. FALLOC_FL_PUNCH_HOLE is supported on at least the following
// filesystems:
// XFS (since Linux 2.6.38)
// ext4 (since Linux 3.0)
// Btrfs (since Linux 3.7)
// tmpfs (since Linux 3.5)
// We ignore error since failure of this operation does not affect
// correctness.
// TRAVIS - this code does not work on TRAVIS filesystems.
// the FALLOC_FL_KEEP_SIZE option is expected to not change the size
// of the file, but it does. Simple strace report will show that.
// While we work with Travis-CI team to figure out if this is a
// quirk of Docker/AUFS, we will comment this out.
struct stat file_stats;
int result = fstat(fd_, &file_stats);
// After ftruncate, we check whether ftruncate has the correct behavior.
// If not, we should hack it with FALLOC_FL_PUNCH_HOLE
if (result == 0 &&
(file_stats.st_size + file_stats.st_blksize - 1) /
file_stats.st_blksize !=
file_stats.st_blocks / (file_stats.st_blksize / 512)) {
IOSTATS_TIMER_GUARD(allocate_nanos);
if (allow_fallocate_) {
fallocate(fd_, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, filesize_,
block_size * last_allocated_block - filesize_);
}
}
#endif
}
if (close(fd_) < 0) {
s = IOError("While closing file after writing", filename_, errno);
}
fd_ = -1;
return s;
}
// write out the cached data to the OS cache
Status PosixWritableFile::Flush() { return Status::OK(); }
Status PosixWritableFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync", filename_, errno);
}
return Status::OK();
}
Status PosixWritableFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync", filename_, errno);
}
return Status::OK();
}
bool PosixWritableFile::IsSyncThreadSafe() const { return true; }
uint64_t PosixWritableFile::GetFileSize() { return filesize_; }
void PosixWritableFile::SetWriteLifeTimeHint(Env::WriteLifeTimeHint hint) {
#ifdef OS_LINUX
// Suppress Valgrind "Unimplemented functionality" error.
#ifndef ROCKSDB_VALGRIND_RUN
if (hint == write_hint_) {
return;
}
if (fcntl(fd_, F_SET_RW_HINT, &hint) == 0) {
write_hint_ = hint;
}
#else
(void)hint;
#endif // ROCKSDB_VALGRIND_RUN
#else
(void)hint;
#endif // OS_LINUX
}
Status PosixWritableFile::InvalidateCache(size_t offset, size_t length) {
if (use_direct_io()) {
return Status::OK();
}
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded", filename_, errno);
#endif
}
#ifdef ROCKSDB_FALLOCATE_PRESENT
Status PosixWritableFile::Allocate(uint64_t offset, uint64_t len) {
assert(offset <= std::numeric_limits<off_t>::max());
assert(len <= std::numeric_limits<off_t>::max());
TEST_KILL_RANDOM("PosixWritableFile::Allocate:0", rocksdb_kill_odds);
IOSTATS_TIMER_GUARD(allocate_nanos);
int alloc_status = 0;
if (allow_fallocate_) {
alloc_status =
fallocate(fd_, fallocate_with_keep_size_ ? FALLOC_FL_KEEP_SIZE : 0,
static_cast<off_t>(offset), static_cast<off_t>(len));
}
if (alloc_status == 0) {
return Status::OK();
} else {
return IOError(
"While fallocate offset " + ToString(offset) + " len " + ToString(len),
filename_, errno);
}
}
#endif
Status PosixWritableFile::RangeSync(uint64_t offset, uint64_t nbytes) {
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
assert(offset <= std::numeric_limits<off_t>::max());
assert(nbytes <= std::numeric_limits<off_t>::max());
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
if (sync_file_range_supported_) {
int ret;
if (strict_bytes_per_sync_) {
// Specifying `SYNC_FILE_RANGE_WAIT_BEFORE` together with an offset/length
// that spans all bytes written so far tells `sync_file_range` to wait for
// any outstanding writeback requests to finish before issuing a new one.
ret =
sync_file_range(fd_, 0, static_cast<off_t>(offset + nbytes),
SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE);
} else {
ret = sync_file_range(fd_, static_cast<off_t>(offset),
static_cast<off_t>(nbytes), SYNC_FILE_RANGE_WRITE);
}
if (ret != 0) {
return IOError("While sync_file_range returned " + ToString(ret),
filename_, errno);
}
return Status::OK();
}
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
#endif // ROCKSDB_RANGESYNC_PRESENT
return WritableFile::RangeSync(offset, nbytes);
}
#ifdef OS_LINUX
size_t PosixWritableFile::GetUniqueId(char* id, size_t max_size) const {
return PosixHelper::GetUniqueIdFromFile(fd_, id, max_size);
}
#endif
/*
* PosixRandomRWFile
*/
PosixRandomRWFile::PosixRandomRWFile(const std::string& fname, int fd,
const EnvOptions& /*options*/)
: filename_(fname), fd_(fd) {}
PosixRandomRWFile::~PosixRandomRWFile() {
if (fd_ >= 0) {
Close();
}
}
Status PosixRandomRWFile::Write(uint64_t offset, const Slice& data) {
const char* src = data.data();
size_t nbytes = data.size();
if (!PosixPositionedWrite(fd_, src, nbytes, static_cast<off_t>(offset))) {
return IOError(
"While write random read/write file at offset " + ToString(offset),
filename_, errno);
}
return Status::OK();
}
Status PosixRandomRWFile::Read(uint64_t offset, size_t n, Slice* result,
char* scratch) const {
size_t left = n;
char* ptr = scratch;
while (left > 0) {
ssize_t done = pread(fd_, ptr, left, offset);
if (done < 0) {
// error while reading from file
if (errno == EINTR) {
// read was interrupted, try again.
continue;
}
return IOError("While reading random read/write file offset " +
ToString(offset) + " len " + ToString(n),
filename_, errno);
} else if (done == 0) {
// Nothing more to read
break;
}
// Read `done` bytes
ptr += done;
offset += done;
left -= done;
}
*result = Slice(scratch, n - left);
return Status::OK();
}
Status PosixRandomRWFile::Flush() { return Status::OK(); }
Status PosixRandomRWFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync random read/write file", filename_, errno);
}
return Status::OK();
}
Status PosixRandomRWFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync random read/write file", filename_, errno);
}
return Status::OK();
}
Status PosixRandomRWFile::Close() {
if (close(fd_) < 0) {
return IOError("While close random read/write file", filename_, errno);
}
fd_ = -1;
return Status::OK();
}
PosixMemoryMappedFileBuffer::~PosixMemoryMappedFileBuffer() {
// TODO should have error handling though not much we can do...
munmap(this->base_, length_);
}
/*
* PosixDirectory
*/
PosixDirectory::~PosixDirectory() { close(fd_); }
Status PosixDirectory::Fsync() {
#ifndef OS_AIX
if (fsync(fd_) == -1) {
return IOError("While fsync", "a directory", errno);
}
#endif
return Status::OK();
}
} // namespace rocksdb
#endif