Add bitmap control for perf context (#237)

* `PerfFlag` implementation aside the `PerfLevel`
design spec in on: https://docs.google.com/document/d/1JYmWMIZwYV0AZW6rNv_oFVZLBCAkxLLYslt39eEZstM/edit?usp=sharing

Signed-off-by: lemonhx <lemonhx@lemonhx.tech>
Co-authored-by: Xinye Tao <xy.tao@outlook.com>
Signed-off-by: tabokie <xy.tao@outlook.com>

CMakeLists.txt

@@ -759,6 +759,7 @@ set(SOURCES
  monitoring/iostats_context.cc
  monitoring/perf_context.cc
  monitoring/perf_level.cc
+ monitoring/perf_flag.cc
  monitoring/persistent_stats_history.cc
  monitoring/statistics.cc
  monitoring/thread_status_impl.cc

TARGETS

@@ -269,6 +269,7 @@ cpp_library(
  "monitoring/instrumented_mutex.cc",
  "monitoring/iostats_context.cc",
  "monitoring/perf_context.cc",
+ "monitoring/perf_flag.cc",
  "monitoring/perf_level.cc",
  "monitoring/persistent_stats_history.cc",
  "monitoring/statistics.cc",
@@ -600,6 +601,7 @@ cpp_library(
  "monitoring/instrumented_mutex.cc",
  "monitoring/iostats_context.cc",
  "monitoring/perf_context.cc",
+ "monitoring/perf_flag.cc",
  "monitoring/perf_level.cc",
  "monitoring/persistent_stats_history.cc",
  "monitoring/statistics.cc",

db/c.cc

@@ -29,6 +29,7 @@
 #include "rocksdb/merge_operator.h"
 #include "rocksdb/options.h"
 #include "rocksdb/perf_context.h"
+#include "rocksdb/perf_flag.h"
 #include "rocksdb/rate_limiter.h"
 #include "rocksdb/slice_transform.h"
 #include "rocksdb/statistics.h"
@@ -57,6 +58,7 @@ using ROCKSDB_NAMESPACE::BlockBasedTableOptions;
 using ROCKSDB_NAMESPACE::BottommostLevelCompaction;
 using ROCKSDB_NAMESPACE::BytewiseComparator;
 using ROCKSDB_NAMESPACE::Cache;
+using ROCKSDB_NAMESPACE::CheckPerfFlag;
 using ROCKSDB_NAMESPACE::Checkpoint;
 using ROCKSDB_NAMESPACE::ColumnFamilyDescriptor;
 using ROCKSDB_NAMESPACE::ColumnFamilyHandle;
@@ -71,6 +73,8 @@ using ROCKSDB_NAMESPACE::CuckooTableOptions;
 using ROCKSDB_NAMESPACE::DB;
 using ROCKSDB_NAMESPACE::DBOptions;
 using ROCKSDB_NAMESPACE::DbPath;
+using ROCKSDB_NAMESPACE::DisablePerfFlag;
+using ROCKSDB_NAMESPACE::EnablePerfFlag;
 using ROCKSDB_NAMESPACE::Env;
 using ROCKSDB_NAMESPACE::EnvOptions;
 using ROCKSDB_NAMESPACE::FileLock;
@@ -3586,6 +3590,14 @@ void rocksdb_set_perf_level(int v) {
  SetPerfLevel(level);
 }
 
+void rocksdb_enable_perf_flag(uint64_t flag) { EnablePerfFlag(flag); }
+
+void rocksdb_disable_perf_flag(uint64_t flag) { DisablePerfFlag(flag); }
+
+int rocksdb_check_perf_flag(uint64_t flag) {
+ return static_cast<int>(CheckPerfFlag(flag));
+}
+
 rocksdb_perfcontext_t* rocksdb_perfcontext_create() {
  rocksdb_perfcontext_t* context = new rocksdb_perfcontext_t;
  context->rep = ROCKSDB_NAMESPACE::get_perf_context();

db/perf_context_test.cc

@@ -956,6 +956,26 @@ TEST_F(PerfContextTest, CPUTimer) {
  ASSERT_EQ(count, get_perf_context()->iter_seek_cpu_nanos);
  }
 }
+
+TEST_F(PerfContextTest, BitMapControl) {
+ DestroyDB(kDbName, Options());
+ auto db = OpenDb();
+ WriteOptions write_options;
+ SetPerfLevel(PerfLevel::kDisable);
+ EnablePerfFlag(flag_user_key_comparison_count);
+ EnablePerfFlag(flag_write_wal_time);
+
+ for (int i = 0; i < FLAGS_total_keys; ++i) {
+ std::string i_str = ToString(i);
+ std::string key = "k" + i_str;
+ std::string value = "v" + i_str;
+
+ db->Put(write_options, key, value);
+ }
+ ASSERT_GT(get_perf_context()->user_key_comparison_count, 0);
+ ASSERT_GT(get_perf_context()->write_wal_time, 0);
+}
+
 } // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

include/rocksdb/c.h

@@ -1543,6 +1543,9 @@ enum {
 };
 
 extern ROCKSDB_LIBRARY_API void rocksdb_set_perf_level(int);
+extern ROCKSDB_LIBRARY_API void rocksdb_enable_perf_flag(uint64_t);
+extern ROCKSDB_LIBRARY_API void rocksdb_disable_perf_flag(uint64_t);
+extern ROCKSDB_LIBRARY_API int rocksdb_check_perf_flag(uint64_t);
 extern ROCKSDB_LIBRARY_API rocksdb_perfcontext_t* rocksdb_perfcontext_create(
  void);
 extern ROCKSDB_LIBRARY_API void rocksdb_perfcontext_reset(

include/rocksdb/perf_flag.h

@@ -0,0 +1,127 @@
+// complements to perf_level
+#pragma once
+
+#include <cstdint>
+
+#include "rocksdb/rocksdb_namespace.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+enum {
+ flag_user_key_comparison_count = 0,
+ flag_block_cache_hit_count,
+ flag_block_read_count,
+ flag_block_read_byte,
+ flag_block_read_time,
+ flag_block_cache_index_hit_count,
+ flag_index_block_read_count,
+ flag_block_cache_filter_hit_count,
+ flag_filter_block_read_count,
+ flag_compression_dict_block_read_count,
+ flag_secondary_cache_hit_count,
+ flag_block_checksum_time,
+ flag_block_decompress_time,
+ flag_get_read_bytes,
+ flag_multiget_read_bytes,
+ flag_iter_read_bytes,
+ flag_internal_key_skipped_count,
+ flag_internal_delete_skipped_count,
+ flag_internal_recent_skipped_count,
+ flag_internal_merge_count,
+ flag_get_snapshot_time,
+ flag_get_from_memtable_time,
+ flag_get_from_memtable_count,
+ flag_get_post_process_time,
+ flag_get_from_output_files_time,
+ flag_seek_on_memtable_time,
+ flag_seek_on_memtable_count,
+ flag_next_on_memtable_count,
+ flag_prev_on_memtable_count,
+ flag_seek_child_seek_time,
+ flag_seek_child_seek_count,
+ flag_seek_min_heap_time,
+ flag_seek_max_heap_time,
+ flag_seek_internal_seek_time,
+ flag_find_next_user_entry_time,
+ flag_write_wal_time,
+ flag_write_memtable_time,
+ flag_write_delay_time,
+ flag_write_scheduling_flushes_compactions_time,
+ flag_write_pre_and_post_process_time,
+ flag_write_thread_wait_nanos,
+ flag_db_mutex_lock_nanos,
+ flag_db_condition_wait_nanos,
+ flag_merge_operator_time_nanos,
+ flag_read_index_block_nanos,
+ flag_read_filter_block_nanos,
+ flag_new_table_block_iter_nanos,
+ flag_new_table_iterator_nanos,
+ flag_block_seek_nanos,
+ flag_find_table_nanos,
+ flag_bloom_memtable_hit_count,
+ flag_bloom_memtable_miss_count,
+ flag_bloom_sst_hit_count,
+ flag_bloom_sst_miss_count,
+ flag_key_lock_wait_time,
+ flag_key_lock_wait_count,
+ flag_env_new_sequential_file_nanos,
+ flag_env_new_random_access_file_nanos,
+ flag_env_new_writable_file_nanos,
+ flag_env_reuse_writable_file_nanos,
+ flag_env_new_random_rw_file_nanos,
+ flag_env_new_directory_nanos,
+ flag_env_file_exists_nanos,
+ flag_env_get_children_nanos,
+ flag_env_get_children_file_attributes_nanos,
+ flag_env_delete_file_nanos,
+ flag_env_create_dir_nanos,
+ flag_env_create_dir_if_missing_nanos,
+ flag_env_delete_dir_nanos,
+ flag_env_get_file_size_nanos,
+ flag_env_get_file_modification_time_nanos,
+ flag_env_rename_file_nanos,
+ flag_env_link_file_nanos,
+ flag_env_lock_file_nanos,
+ flag_env_unlock_file_nanos,
+ flag_env_new_logger_nanos,
+ flag_get_cpu_nanos,
+ flag_iter_next_cpu_nanos,
+ flag_iter_prev_cpu_nanos,
+ flag_iter_seek_cpu_nanos,
+ flag_encrypt_data_nanos,
+ flag_decrypt_data_nanos,
+
+ flag_get_from_table_nanos,
+ flag_user_key_return_count,
+ flag_block_cache_miss_count,
+ flag_bloom_filter_full_positive,
+ flag_bloom_filter_useful,
+ flag_bloom_filter_full_true_positive,
+
+ flag_bytes_read,
+ flag_bytes_written,
+ flag_open_nanos,
+ flag_allocate_nanos,
+ flag_write_nanos,
+ flag_read_nanos,
+ flag_range_sync_nanos,
+ flag_prepare_write_nanos,
+ flag_fsync_nanos,
+ flag_logger_nanos,
+ flag_cpu_read_nanos,
+ flag_cpu_write_nanos,
+ // Should always be the last
+ FLAG_END
+};
+
+// FLAGS_LEN = ceiling(FLAG_END / bits(uint8_t))
+#define FLAGS_LEN \
+ (((uint64_t)FLAG_END & (uint64_t)0b111) == 0 \
+ ? ((uint64_t)FLAG_END >> 3) \
+ : ((uint64_t)FLAG_END >> 3) + 1)
+
+void EnablePerfFlag(uint64_t flag);
+void DisablePerfFlag(uint64_t flag);
+bool CheckPerfFlag(uint64_t flag);
+
+} // namespace ROCKSDB_NAMESPACE

monitoring/iostats_context_imp.h

@@ -29,15 +29,16 @@ extern __thread IOStatsContext iostats_context;
 #define IOSTATS(metric) (iostats_context.metric)
 
 // Declare and set start time of the timer
-#define IOSTATS_TIMER_GUARD(metric) \
- PerfStepTimer iostats_step_timer_##metric(&(iostats_context.metric)); \
+#define IOSTATS_TIMER_GUARD(metric) \
+ PerfStepTimer iostats_step_timer_##metric(&(iostats_context.metric), \
+ CheckPerfFlag(flag_##metric)); \
  iostats_step_timer_##metric.Start();
 
 // Declare and set start time of the timer
-#define IOSTATS_CPU_TIMER_GUARD(metric, clock) \
- PerfStepTimer iostats_step_timer_##metric( \
- &(iostats_context.metric), clock, true,  \
- PerfLevel::kEnableTimeAndCPUTimeExceptForMutex); \
+#define IOSTATS_CPU_TIMER_GUARD(metric, clock)  \
+ PerfStepTimer iostats_step_timer_##metric(  \
+ &(iostats_context.metric), CheckPerfFlag(flag_##metric), clock, true, \
+ PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);  \
  iostats_step_timer_##metric.Start();
 
 #else // ROCKSDB_SUPPORT_THREAD_LOCAL && !NIOSTATS_CONTEXT

monitoring/perf_context_imp.h

@@ -41,44 +41,47 @@ extern thread_local PerfContext perf_context;
 #define PERF_TIMER_START(metric) perf_step_timer_##metric.Start();
 
 // Declare and set start time of the timer
-#define PERF_TIMER_GUARD(metric) \
- PerfStepTimer perf_step_timer_##metric(&(perf_context.metric)); \
+#define PERF_TIMER_GUARD(metric) \
+ PerfStepTimer perf_step_timer_##metric(&(perf_context.metric), \
+ CheckPerfFlag(flag_##metric)); \
  perf_step_timer_##metric.Start();
 
 // Declare and set start time of the timer
-#define PERF_TIMER_GUARD_WITH_CLOCK(metric, clock) \
- PerfStepTimer perf_step_timer_##metric(&(perf_context.metric), clock); \
+#define PERF_TIMER_GUARD_WITH_CLOCK(metric, clock) \
+ PerfStepTimer perf_step_timer_##metric(&(perf_context.metric), \
+ CheckPerfFlag(flag_##metric), clock); \
  perf_step_timer_##metric.Start();
 
 // Declare and set start time of the timer
-#define PERF_CPU_TIMER_GUARD(metric, clock) \
- PerfStepTimer perf_step_timer_##metric( \
- &(perf_context.metric), clock, true,  \
- PerfLevel::kEnableTimeAndCPUTimeExceptForMutex); \
+#define PERF_CPU_TIMER_GUARD(metric, clock)  \
+ PerfStepTimer perf_step_timer_##metric(  \
+ &(perf_context.metric), CheckPerfFlag(flag_##metric), clock, true, \
+ PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);  \
  perf_step_timer_##metric.Start();
 
-#define PERF_CONDITIONAL_TIMER_FOR_MUTEX_GUARD(metric, condition, stats,  \
- ticker_type)  \
- PerfStepTimer perf_step_timer_##metric(&(perf_context.metric), nullptr, \
-  false, PerfLevel::kEnableTime, stats, \
-  ticker_type); \
- if (condition) {  \
- perf_step_timer_##metric.Start();  \
+#define PERF_CONDITIONAL_TIMER_FOR_MUTEX_GUARD(metric, condition, stats, \
+ ticker_type) \
+ PerfStepTimer perf_step_timer_##metric(  \
+ &(perf_context.metric), CheckPerfFlag(flag_##metric), nullptr, false, \
+ PerfLevel::kEnableTime, stats, ticker_type);  \
+ if (condition) { \
+ perf_step_timer_##metric.Start(); \
  }
 
 // Update metric with time elapsed since last START. start time is reset
 // to current timestamp.
 #define PERF_TIMER_MEASURE(metric) perf_step_timer_##metric.Measure();
 
 // Increase metric value
-#define PERF_COUNTER_ADD(metric, value) \
- if (perf_level >= PerfLevel::kEnableCount) { \
- perf_context.metric += value; \
+#define PERF_COUNTER_ADD(metric, value)  \
+ if (perf_level >= PerfLevel::kEnableCount || CheckPerfFlag(flag_##metric)) { \
+ perf_context.metric += value;  \
  }
 
 // Increase metric value
 #define PERF_COUNTER_BY_LEVEL_ADD(metric, value, level) \
- if (perf_level >= PerfLevel::kEnableCount && \
+ if ((perf_level >= PerfLevel::kEnableCount || \
+ CheckPerfFlag(flag_##metric)) && \
  perf_context.per_level_perf_context_enabled && \
  perf_context.level_to_perf_context) { \
  if ((*(perf_context.level_to_perf_context)).find(level) != \

monitoring/perf_flag.cc

@@ -0,0 +1,32 @@
+#include "rocksdb/perf_flag.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+#ifdef ROCKSDB_SUPPORT_THREAD_LOCAL
+__thread uint8_t perf_flags[FLAGS_LEN] = {0};
+#else
+uint8_t perf_flags[FLAGS_LEN] = {0};
+#endif
+
+#define GET_FLAG(flag) perf_flags[(uint64_t)(flag) >> 3]
+
+void EnablePerfFlag(uint64_t flag) {
+ if (!CheckPerfFlag(flag)) {
+ // & 0b111 means find the flag location is a alternative way to do mod
+ // operation
+ GET_FLAG(flag) ^= (uint64_t)0b1 << ((uint64_t)flag & (uint64_t)0b111);
+ }
+}
+
+void DisablePerfFlag(uint64_t flag) {
+ if (CheckPerfFlag(flag)) {
+ GET_FLAG(flag) ^= (uint64_t)0b1 << ((uint64_t)flag & (uint64_t)0b111);
+ }
+}
+
+bool CheckPerfFlag(uint64_t flag) {
+ return ((uint64_t)GET_FLAG(flag) & (uint64_t)0b1
+ << (flag & (uint64_t)0b111)) != 0;
+}
+
+} // namespace ROCKSDB_NAMESPACE

monitoring/perf_flag_imp.h

@@ -0,0 +1,18 @@
+// 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).
+//
+#pragma once
+
+#include <cstdint>
+
+#include "rocksdb/perf_flag.h"
+
+namespace ROCKSDB_NAMESPACE {
+#ifdef ROCKSDB_SUPPORT_THREAD_LOCAL
+extern __thread uint8_t perf_flags[FLAGS_LEN];
+#else
+extern uint8_t perf_flags[FLAGS_LEN];
+#endif
+} // namespace ROCKSDB_NAMESPACE

monitoring/perf_step_timer.h

@@ -4,6 +4,7 @@
 // (found in the LICENSE.Apache file in the root directory).
 //
 #pragma once
+#include "monitoring/perf_flag_imp.h"
 #include "monitoring/perf_level_imp.h"
 #include "monitoring/statistics.h"
 #include "rocksdb/system_clock.h"
@@ -13,10 +14,11 @@ namespace ROCKSDB_NAMESPACE {
 class PerfStepTimer {
  public:
  explicit PerfStepTimer(
- uint64_t* metric, SystemClock* clock = nullptr, bool use_cpu_time = false,
+ uint64_t* metric, bool enable_flag = false, SystemClock* clock = nullptr,
+ bool use_cpu_time = false,
  PerfLevel enable_level = PerfLevel::kEnableTimeExceptForMutex,
  Statistics* statistics = nullptr, uint32_t ticker_type = 0)
- : perf_counter_enabled_(perf_level >= enable_level),
+ : perf_counter_enabled_(perf_level >= enable_level || enable_flag),
  use_cpu_time_(use_cpu_time),
  ticker_type_(ticker_type),
  clock_((perf_counter_enabled_ || statistics != nullptr)