mirror of
https://github.com/facebook/rocksdb.git
synced 2024-11-26 07:30:54 +00:00
151242ce46
Summary: **Context:** The existing stat rocksdb.sst.read.micros does not reflect each of compaction and flush cases but aggregate them, which is not so helpful for us to understand IO read behavior of each of them. **Summary** - Update `StopWatch` and `RandomAccessFileReader` to record `rocksdb.sst.read.micros` and `rocksdb.file.{flush/compaction}.read.micros` - Fixed the default histogram in `RandomAccessFileReader` - New field `ReadOptions/IOOptions::io_activity`; Pass `ReadOptions` through paths under db open, flush and compaction to where we can prepare `IOOptions` and pass it to `RandomAccessFileReader` - Use `thread_status_util` for assertion in `DbStressFSWrapper` for continuous testing on we are passing correct `io_activity` under db open, flush and compaction Pull Request resolved: https://github.com/facebook/rocksdb/pull/11288 Test Plan: - **Stress test** - **Db bench 1: rocksdb.sst.read.micros COUNT ≈ sum of rocksdb.file.read.flush.micros's and rocksdb.file.read.compaction.micros's.** (without blob) - May not be exactly the same due to `HistogramStat::Add` only guarantees atomic not accuracy across threads. ``` ./db_bench -db=/dev/shm/testdb/ -statistics=true -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -target_file_size_base=655 -disable_auto_compactions=false -compression_type=none -bloom_bits=3 (-use_plain_table=1 -prefix_size=10) ``` ``` // BlockBasedTable rocksdb.sst.read.micros P50 : 2.009374 P95 : 4.968548 P99 : 8.110362 P100 : 43.000000 COUNT : 40456 SUM : 114805 rocksdb.file.read.flush.micros P50 : 1.871841 P95 : 3.872407 P99 : 5.540541 P100 : 43.000000 COUNT : 2250 SUM : 6116 rocksdb.file.read.compaction.micros P50 : 2.023109 P95 : 5.029149 P99 : 8.196910 P100 : 26.000000 COUNT : 38206 SUM : 108689 // PlainTable Does not apply ``` - **Db bench 2: performance** **Read** SETUP: db with 900 files ``` ./db_bench -db=/dev/shm/testdb/ -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=true -target_file_size_base=655 -compression_type=none ```run till convergence ``` ./db_bench -seed=1678564177044286 -use_existing_db=true -db=/dev/shm/testdb -benchmarks=readrandom[-X60] -statistics=true -num=1000000 -disable_auto_compactions=true -compression_type=none -bloom_bits=3 ``` Pre-change `readrandom [AVG 60 runs] : 21568 (± 248) ops/sec` Post-change (no regression, -0.3%) `readrandom [AVG 60 runs] : 21486 (± 236) ops/sec` **Compaction/Flush**run till convergence ``` ./db_bench -db=/dev/shm/testdb2/ -seed=1678564177044286 -benchmarks="fillseq[-X60]" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=false -target_file_size_base=655 -compression_type=none rocksdb.sst.read.micros COUNT : 33820 rocksdb.sst.read.flush.micros COUNT : 1800 rocksdb.sst.read.compaction.micros COUNT : 32020 ``` Pre-change `fillseq [AVG 46 runs] : 1391 (± 214) ops/sec; 0.7 (± 0.1) MB/sec` Post-change (no regression, ~-0.4%) `fillseq [AVG 46 runs] : 1385 (± 216) ops/sec; 0.7 (± 0.1) MB/sec` Reviewed By: ajkr Differential Revision: D44007011 Pulled By: hx235 fbshipit-source-id: a54c89e4846dfc9a135389edf3f3eedfea257132
329 lines
11 KiB
C++
329 lines
11 KiB
C++
// 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).
|
|
|
|
#include "monitoring/thread_status_updater.h"
|
|
|
|
#include <memory>
|
|
|
|
#include "port/likely.h"
|
|
#include "rocksdb/env.h"
|
|
#include "rocksdb/system_clock.h"
|
|
#include "util/mutexlock.h"
|
|
|
|
namespace ROCKSDB_NAMESPACE {
|
|
|
|
#ifdef ROCKSDB_USING_THREAD_STATUS
|
|
|
|
thread_local ThreadStatusData* ThreadStatusUpdater::thread_status_data_ =
|
|
nullptr;
|
|
|
|
void ThreadStatusUpdater::RegisterThread(ThreadStatus::ThreadType ttype,
|
|
uint64_t thread_id) {
|
|
if (UNLIKELY(thread_status_data_ == nullptr)) {
|
|
thread_status_data_ = new ThreadStatusData();
|
|
thread_status_data_->thread_type = ttype;
|
|
thread_status_data_->thread_id = thread_id;
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
thread_data_set_.insert(thread_status_data_);
|
|
}
|
|
|
|
ClearThreadOperationProperties();
|
|
}
|
|
|
|
void ThreadStatusUpdater::UnregisterThread() {
|
|
if (thread_status_data_ != nullptr) {
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
thread_data_set_.erase(thread_status_data_);
|
|
delete thread_status_data_;
|
|
thread_status_data_ = nullptr;
|
|
}
|
|
}
|
|
|
|
void ThreadStatusUpdater::ResetThreadStatus() {
|
|
ClearThreadState();
|
|
ClearThreadOperation();
|
|
SetColumnFamilyInfoKey(nullptr);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetEnableTracking(bool enable_tracking) {
|
|
auto* data = Get();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->enable_tracking.store(enable_tracking, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetColumnFamilyInfoKey(const void* cf_key) {
|
|
auto* data = Get();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->cf_key.store(const_cast<void*>(cf_key), std::memory_order_relaxed);
|
|
}
|
|
|
|
const void* ThreadStatusUpdater::GetColumnFamilyInfoKey() {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return nullptr;
|
|
}
|
|
return data->cf_key.load(std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetThreadOperation(
|
|
const ThreadStatus::OperationType type) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
// NOTE: Our practice here is to set all the thread operation properties
|
|
// and stage before we set thread operation, and thread operation
|
|
// will be set in std::memory_order_release. This is to ensure
|
|
// whenever a thread operation is not OP_UNKNOWN, we will always
|
|
// have a consistent information on its properties.
|
|
data->operation_type.store(type, std::memory_order_release);
|
|
if (type == ThreadStatus::OP_UNKNOWN) {
|
|
data->operation_stage.store(ThreadStatus::STAGE_UNKNOWN,
|
|
std::memory_order_relaxed);
|
|
ClearThreadOperationProperties();
|
|
}
|
|
}
|
|
|
|
ThreadStatus::OperationType ThreadStatusUpdater::GetThreadOperation() {
|
|
ThreadStatusData* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return ThreadStatus::OperationType::OP_UNKNOWN;
|
|
}
|
|
return data->operation_type.load(std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetThreadOperationProperty(int i, uint64_t value) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->op_properties[i].store(value, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::IncreaseThreadOperationProperty(int i,
|
|
uint64_t delta) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->op_properties[i].fetch_add(delta, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetOperationStartTime(const uint64_t start_time) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->op_start_time.store(start_time, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::ClearThreadOperation() {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->operation_stage.store(ThreadStatus::STAGE_UNKNOWN,
|
|
std::memory_order_relaxed);
|
|
data->operation_type.store(ThreadStatus::OP_UNKNOWN,
|
|
std::memory_order_relaxed);
|
|
ClearThreadOperationProperties();
|
|
}
|
|
|
|
void ThreadStatusUpdater::ClearThreadOperationProperties() {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
for (int i = 0; i < ThreadStatus::kNumOperationProperties; ++i) {
|
|
data->op_properties[i].store(0, std::memory_order_relaxed);
|
|
}
|
|
}
|
|
|
|
ThreadStatus::OperationStage ThreadStatusUpdater::SetThreadOperationStage(
|
|
ThreadStatus::OperationStage stage) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return ThreadStatus::STAGE_UNKNOWN;
|
|
}
|
|
return data->operation_stage.exchange(stage, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::SetThreadState(const ThreadStatus::StateType type) {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->state_type.store(type, std::memory_order_relaxed);
|
|
}
|
|
|
|
void ThreadStatusUpdater::ClearThreadState() {
|
|
auto* data = GetLocalThreadStatus();
|
|
if (data == nullptr) {
|
|
return;
|
|
}
|
|
data->state_type.store(ThreadStatus::STATE_UNKNOWN,
|
|
std::memory_order_relaxed);
|
|
}
|
|
|
|
Status ThreadStatusUpdater::GetThreadList(
|
|
std::vector<ThreadStatus>* thread_list) {
|
|
thread_list->clear();
|
|
std::vector<std::shared_ptr<ThreadStatusData>> valid_list;
|
|
uint64_t now_micros = SystemClock::Default()->NowMicros();
|
|
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
for (auto* thread_data : thread_data_set_) {
|
|
assert(thread_data);
|
|
auto thread_id = thread_data->thread_id.load(std::memory_order_relaxed);
|
|
auto thread_type = thread_data->thread_type.load(std::memory_order_relaxed);
|
|
// Since any change to cf_info_map requires thread_list_mutex,
|
|
// which is currently held by GetThreadList(), here we can safely
|
|
// use "memory_order_relaxed" to load the cf_key.
|
|
auto cf_key = thread_data->cf_key.load(std::memory_order_relaxed);
|
|
|
|
ThreadStatus::OperationType op_type = ThreadStatus::OP_UNKNOWN;
|
|
ThreadStatus::OperationStage op_stage = ThreadStatus::STAGE_UNKNOWN;
|
|
ThreadStatus::StateType state_type = ThreadStatus::STATE_UNKNOWN;
|
|
uint64_t op_elapsed_micros = 0;
|
|
uint64_t op_props[ThreadStatus::kNumOperationProperties] = {0};
|
|
|
|
auto iter = cf_info_map_.find(cf_key);
|
|
if (iter != cf_info_map_.end()) {
|
|
op_type = thread_data->operation_type.load(std::memory_order_acquire);
|
|
// display lower-level info only when higher-level info is available.
|
|
if (op_type != ThreadStatus::OP_UNKNOWN) {
|
|
op_elapsed_micros = now_micros - thread_data->op_start_time.load(
|
|
std::memory_order_relaxed);
|
|
op_stage = thread_data->operation_stage.load(std::memory_order_relaxed);
|
|
state_type = thread_data->state_type.load(std::memory_order_relaxed);
|
|
for (int i = 0; i < ThreadStatus::kNumOperationProperties; ++i) {
|
|
op_props[i] =
|
|
thread_data->op_properties[i].load(std::memory_order_relaxed);
|
|
}
|
|
}
|
|
}
|
|
|
|
thread_list->emplace_back(
|
|
thread_id, thread_type,
|
|
iter != cf_info_map_.end() ? iter->second.db_name : "",
|
|
iter != cf_info_map_.end() ? iter->second.cf_name : "", op_type,
|
|
op_elapsed_micros, op_stage, op_props, state_type);
|
|
}
|
|
|
|
return Status::OK();
|
|
}
|
|
|
|
ThreadStatusData* ThreadStatusUpdater::GetLocalThreadStatus() {
|
|
if (thread_status_data_ == nullptr) {
|
|
return nullptr;
|
|
}
|
|
if (!thread_status_data_->enable_tracking.load(std::memory_order_relaxed)) {
|
|
return nullptr;
|
|
}
|
|
return thread_status_data_;
|
|
}
|
|
|
|
void ThreadStatusUpdater::NewColumnFamilyInfo(const void* db_key,
|
|
const std::string& db_name,
|
|
const void* cf_key,
|
|
const std::string& cf_name) {
|
|
// Acquiring same lock as GetThreadList() to guarantee
|
|
// a consistent view of global column family table (cf_info_map).
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
|
|
cf_info_map_.emplace(std::piecewise_construct, std::make_tuple(cf_key),
|
|
std::make_tuple(db_key, db_name, cf_name));
|
|
db_key_map_[db_key].insert(cf_key);
|
|
}
|
|
|
|
void ThreadStatusUpdater::EraseColumnFamilyInfo(const void* cf_key) {
|
|
// Acquiring same lock as GetThreadList() to guarantee
|
|
// a consistent view of global column family table (cf_info_map).
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
|
|
auto cf_pair = cf_info_map_.find(cf_key);
|
|
if (cf_pair != cf_info_map_.end()) {
|
|
// Remove its entry from db_key_map_ by the following steps:
|
|
// 1. Obtain the entry in db_key_map_ whose set contains cf_key
|
|
// 2. Remove it from the set.
|
|
ConstantColumnFamilyInfo& cf_info = cf_pair->second;
|
|
auto db_pair = db_key_map_.find(cf_info.db_key);
|
|
assert(db_pair != db_key_map_.end());
|
|
size_t result __attribute__((__unused__));
|
|
result = db_pair->second.erase(cf_key);
|
|
assert(result);
|
|
cf_info_map_.erase(cf_pair);
|
|
}
|
|
}
|
|
|
|
void ThreadStatusUpdater::EraseDatabaseInfo(const void* db_key) {
|
|
// Acquiring same lock as GetThreadList() to guarantee
|
|
// a consistent view of global column family table (cf_info_map).
|
|
std::lock_guard<std::mutex> lck(thread_list_mutex_);
|
|
auto db_pair = db_key_map_.find(db_key);
|
|
if (UNLIKELY(db_pair == db_key_map_.end())) {
|
|
// In some occasional cases such as DB::Open fails, we won't
|
|
// register ColumnFamilyInfo for a db.
|
|
return;
|
|
}
|
|
|
|
for (auto cf_key : db_pair->second) {
|
|
auto cf_pair = cf_info_map_.find(cf_key);
|
|
if (cf_pair != cf_info_map_.end()) {
|
|
cf_info_map_.erase(cf_pair);
|
|
}
|
|
}
|
|
db_key_map_.erase(db_key);
|
|
}
|
|
|
|
#else
|
|
|
|
void ThreadStatusUpdater::RegisterThread(ThreadStatus::ThreadType /*ttype*/,
|
|
uint64_t /*thread_id*/) {}
|
|
|
|
void ThreadStatusUpdater::UnregisterThread() {}
|
|
|
|
void ThreadStatusUpdater::ResetThreadStatus() {}
|
|
|
|
void ThreadStatusUpdater::SetColumnFamilyInfoKey(const void* /*cf_key*/) {}
|
|
|
|
void ThreadStatusUpdater::SetThreadOperation(
|
|
const ThreadStatus::OperationType /*type*/) {}
|
|
|
|
void ThreadStatusUpdater::ClearThreadOperation() {}
|
|
|
|
void ThreadStatusUpdater::SetThreadState(
|
|
const ThreadStatus::StateType /*type*/) {}
|
|
|
|
void ThreadStatusUpdater::ClearThreadState() {}
|
|
|
|
Status ThreadStatusUpdater::GetThreadList(
|
|
std::vector<ThreadStatus>* /*thread_list*/) {
|
|
return Status::NotSupported(
|
|
"GetThreadList is not supported in the current running environment.");
|
|
}
|
|
|
|
void ThreadStatusUpdater::NewColumnFamilyInfo(const void* /*db_key*/,
|
|
const std::string& /*db_name*/,
|
|
const void* /*cf_key*/,
|
|
const std::string& /*cf_name*/) {}
|
|
|
|
void ThreadStatusUpdater::EraseColumnFamilyInfo(const void* /*cf_key*/) {}
|
|
|
|
void ThreadStatusUpdater::EraseDatabaseInfo(const void* /*db_key*/) {}
|
|
|
|
void ThreadStatusUpdater::SetThreadOperationProperty(int /*i*/,
|
|
uint64_t /*value*/) {}
|
|
|
|
void ThreadStatusUpdater::IncreaseThreadOperationProperty(int /*i*/,
|
|
uint64_t /*delta*/) {}
|
|
|
|
#endif // ROCKSDB_USING_THREAD_STATUS
|
|
} // namespace ROCKSDB_NAMESPACE
|