You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/util/arena.cc

227 lines
6.9 KiB

// 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.
// This source code is also licensed under the GPLv2 license found in the
// COPYING file in the root directory of this source tree.
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "util/arena.h"
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
#ifdef OS_FREEBSD
#include <malloc_np.h>
#else
#include <malloc.h>
#endif
#endif
#ifndef OS_WIN
#include <sys/mman.h>
#endif
#include <algorithm>
#include "port/port.h"
#include "rocksdb/env.h"
#include "util/logging.h"
#include "util/sync_point.h"
namespace rocksdb {
// MSVC complains that it is already defined since it is static in the header.
#ifndef _MSC_VER
const size_t Arena::kInlineSize;
#endif
const size_t Arena::kMinBlockSize = 4096;
const size_t Arena::kMaxBlockSize = 2u << 30;
static const int kAlignUnit = sizeof(void*);
size_t OptimizeBlockSize(size_t block_size) {
// Make sure block_size is in optimal range
block_size = std::max(Arena::kMinBlockSize, block_size);
block_size = std::min(Arena::kMaxBlockSize, block_size);
// make sure block_size is the multiple of kAlignUnit
if (block_size % kAlignUnit != 0) {
block_size = (1 + block_size / kAlignUnit) * kAlignUnit;
}
return block_size;
}
Arena::Arena(size_t block_size, AllocTracker* tracker, size_t huge_page_size)
: kBlockSize(OptimizeBlockSize(block_size)), tracker_(tracker) {
assert(kBlockSize >= kMinBlockSize && kBlockSize <= kMaxBlockSize &&
kBlockSize % kAlignUnit == 0);
TEST_SYNC_POINT_CALLBACK("Arena::Arena:0", const_cast<size_t*>(&kBlockSize));
alloc_bytes_remaining_ = sizeof(inline_block_);
blocks_memory_ += alloc_bytes_remaining_;
aligned_alloc_ptr_ = inline_block_;
unaligned_alloc_ptr_ = inline_block_ + alloc_bytes_remaining_;
#ifdef MAP_HUGETLB
hugetlb_size_ = huge_page_size;
if (hugetlb_size_ && kBlockSize > hugetlb_size_) {
hugetlb_size_ = ((kBlockSize - 1U) / hugetlb_size_ + 1U) * hugetlb_size_;
}
#endif
if (tracker_ != nullptr) {
tracker_->Allocate(kInlineSize);
}
}
Arena::~Arena() {
if (tracker_ != nullptr) {
assert(tracker_->is_freed());
tracker_->FreeMem();
}
for (const auto& block : blocks_) {
delete[] block;
}
#ifdef MAP_HUGETLB
for (const auto& mmap_info : huge_blocks_) {
auto ret = munmap(mmap_info.addr_, mmap_info.length_);
if (ret != 0) {
// TODO(sdong): Better handling
}
}
#endif
}
char* Arena::AllocateFallback(size_t bytes, bool aligned) {
if (bytes > kBlockSize / 4) {
++irregular_block_num;
// Object is more than a quarter of our block size. Allocate it separately
// to avoid wasting too much space in leftover bytes.
return AllocateNewBlock(bytes);
}
// We waste the remaining space in the current block.
size_t size = 0;
char* block_head = nullptr;
#ifdef MAP_HUGETLB
if (hugetlb_size_) {
size = hugetlb_size_;
block_head = AllocateFromHugePage(size);
}
#endif
if (!block_head) {
size = kBlockSize;
block_head = AllocateNewBlock(size);
}
alloc_bytes_remaining_ = size - bytes;
if (aligned) {
aligned_alloc_ptr_ = block_head + bytes;
unaligned_alloc_ptr_ = block_head + size;
return block_head;
} else {
aligned_alloc_ptr_ = block_head;
unaligned_alloc_ptr_ = block_head + size - bytes;
return unaligned_alloc_ptr_;
}
}
char* Arena::AllocateFromHugePage(size_t bytes) {
#ifdef MAP_HUGETLB
if (hugetlb_size_ == 0) {
return nullptr;
}
// already reserve space in huge_blocks_ before calling mmap().
// this way the insertion into the vector below will not throw and we
// won't leak the mapping in that case. if reserve() throws, we
// won't leak either
huge_blocks_.reserve(huge_blocks_.size() + 1);
void* addr = mmap(nullptr, bytes, (PROT_READ | PROT_WRITE),
(MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB), -1, 0);
if (addr == MAP_FAILED) {
return nullptr;
}
// the following shouldn't throw because of the above reserve()
huge_blocks_.emplace_back(MmapInfo(addr, bytes));
blocks_memory_ += bytes;
if (tracker_ != nullptr) {
tracker_->Allocate(bytes);
}
return reinterpret_cast<char*>(addr);
#else
return nullptr;
#endif
}
char* Arena::AllocateAligned(size_t bytes, size_t huge_page_size,
Logger* logger) {
assert((kAlignUnit & (kAlignUnit - 1)) ==
0); // Pointer size should be a power of 2
11 years ago
#ifdef MAP_HUGETLB
if (huge_page_size > 0 && bytes > 0) {
// Allocate from a huge page TBL table.
assert(logger != nullptr); // logger need to be passed in.
size_t reserved_size =
((bytes - 1U) / huge_page_size + 1U) * huge_page_size;
assert(reserved_size >= bytes);
char* addr = AllocateFromHugePage(reserved_size);
if (addr == nullptr) {
ROCKS_LOG_WARN(logger,
"AllocateAligned fail to allocate huge TLB pages: %s",
strerror(errno));
// fail back to malloc
} else {
return addr;
}
}
#endif
size_t current_mod =
reinterpret_cast<uintptr_t>(aligned_alloc_ptr_) & (kAlignUnit - 1);
size_t slop = (current_mod == 0 ? 0 : kAlignUnit - current_mod);
size_t needed = bytes + slop;
char* result;
if (needed <= alloc_bytes_remaining_) {
result = aligned_alloc_ptr_ + slop;
aligned_alloc_ptr_ += needed;
alloc_bytes_remaining_ -= needed;
} else {
support for concurrent adds to memtable Summary: This diff adds support for concurrent adds to the skiplist memtable implementations. Memory allocation is made thread-safe by the addition of a spinlock, with small per-core buffers to avoid contention. Concurrent memtable writes are made via an additional method and don't impose a performance overhead on the non-concurrent case, so parallelism can be selected on a per-batch basis. Write thread synchronization is an increasing bottleneck for higher levels of concurrency, so this diff adds --enable_write_thread_adaptive_yield (default off). This feature causes threads joining a write batch group to spin for a short time (default 100 usec) using sched_yield, rather than going to sleep on a mutex. If the timing of the yield calls indicates that another thread has actually run during the yield then spinning is avoided. This option improves performance for concurrent situations even without parallel adds, although it has the potential to increase CPU usage (and the heuristic adaptation is not yet mature). Parallel writes are not currently compatible with inplace updates, update callbacks, or delete filtering. Enable it with --allow_concurrent_memtable_write (and --enable_write_thread_adaptive_yield). Parallel memtable writes are performance neutral when there is no actual parallelism, and in my experiments (SSD server-class Linux and varying contention and key sizes for fillrandom) they are always a performance win when there is more than one thread. Statistics are updated earlier in the write path, dropping the number of DB mutex acquisitions from 2 to 1 for almost all cases. This diff was motivated and inspired by Yahoo's cLSM work. It is more conservative than cLSM: RocksDB's write batch group leader role is preserved (along with all of the existing flush and write throttling logic) and concurrent writers are blocked until all memtable insertions have completed and the sequence number has been advanced, to preserve linearizability. My test config is "db_bench -benchmarks=fillrandom -threads=$T -batch_size=1 -memtablerep=skip_list -value_size=100 --num=1000000/$T -level0_slowdown_writes_trigger=9999 -level0_stop_writes_trigger=9999 -disable_auto_compactions --max_write_buffer_number=8 -max_background_flushes=8 --disable_wal --write_buffer_size=160000000 --block_size=16384 --allow_concurrent_memtable_write" on a two-socket Xeon E5-2660 @ 2.2Ghz with lots of memory and an SSD hard drive. With 1 thread I get ~440Kops/sec. Peak performance for 1 socket (numactl -N1) is slightly more than 1Mops/sec, at 16 threads. Peak performance across both sockets happens at 30 threads, and is ~900Kops/sec, although with fewer threads there is less performance loss when the system has background work. Test Plan: 1. concurrent stress tests for InlineSkipList and DynamicBloom 2. make clean; make check 3. make clean; DISABLE_JEMALLOC=1 make valgrind_check; valgrind db_bench 4. make clean; COMPILE_WITH_TSAN=1 make all check; db_bench 5. make clean; COMPILE_WITH_ASAN=1 make all check; db_bench 6. make clean; OPT=-DROCKSDB_LITE make check 7. verify no perf regressions when disabled Reviewers: igor, sdong Reviewed By: sdong Subscribers: MarkCallaghan, IslamAbdelRahman, anthony, yhchiang, rven, sdong, guyg8, kradhakrishnan, dhruba Differential Revision: https://reviews.facebook.net/D50589
9 years ago
// AllocateFallback always returns aligned memory
result = AllocateFallback(bytes, true /* aligned */);
}
assert((reinterpret_cast<uintptr_t>(result) & (kAlignUnit - 1)) == 0);
return result;
}
char* Arena::AllocateNewBlock(size_t block_bytes) {
// already reserve space in blocks_ before allocating memory via new.
// this way the insertion into the vector below will not throw and we
// won't leak the allocated memory in that case. if reserve() throws,
// we won't leak either
blocks_.reserve(blocks_.size() + 1);
char* block = new char[block_bytes];
size_t allocated_size;
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
allocated_size = malloc_usable_size(block);
#ifndef NDEBUG
// It's hard to predict what malloc_usable_size() returns.
// A callback can allow users to change the costed size.
std::pair<size_t*, size_t*> pair(&allocated_size, &block_bytes);
TEST_SYNC_POINT_CALLBACK("Arena::AllocateNewBlock:0", &pair);
#endif // NDEBUG
#else
allocated_size = block_bytes;
#endif // ROCKSDB_MALLOC_USABLE_SIZE
blocks_memory_ += allocated_size;
if (tracker_ != nullptr) {
tracker_->Allocate(allocated_size);
}
// the following shouldn't throw because of the above reserve()
blocks_.push_back(block);
return block;
}
} // namespace rocksdb