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>

CMakeLists.txt

@@ -585,6 +585,7 @@ 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

db/c.cc

@@ -24,6 +24,7 @@
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/merge_operator.h"
 #include "rocksdb/options.h"
+#include "rocksdb/perf_flag.h"
 #include "rocksdb/perf_context.h"
 #include "rocksdb/rate_limiter.h"
 #include "rocksdb/slice_transform.h"
@@ -114,6 +115,9 @@ using rocksdb::Checkpoint;
 using rocksdb::TransactionLogIterator;
 using rocksdb::BatchResult;
 using rocksdb::PerfLevel;
+using rocksdb::EnablePerfFlag;
+using rocksdb::DisablePerfFlag;
+using rocksdb::CheckPerfFlag;
 using rocksdb::PerfContext;
 using rocksdb::MemoryUtil;
 
@@ -2749,6 +2753,18 @@ 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 (int)CheckPerfFlag(flag);
+}
+
 rocksdb_perfcontext_t* rocksdb_perfcontext_create() {
  rocksdb_perfcontext_t* context = new rocksdb_perfcontext_t;
  context->rep = rocksdb::get_perf_context();

include/rocksdb/c.h

@@ -1119,7 +1119,11 @@ enum {
  rocksdb_total_metric_count = 68
 };
 
+#include "perf_flag_defs.h"
 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();
 extern ROCKSDB_LIBRARY_API void rocksdb_perfcontext_reset(
  rocksdb_perfcontext_t* context);

include/rocksdb/perf_flag.h

@@ -0,0 +1,15 @@
+// complements to perf_level
+#include <cstdint>
+#pragma once
+#define FLAGS_LEN \
+ (((uint64_t)FLAG_END & (uint64_t)0b111) == 0 \
+ ? ((uint64_t)FLAG_END >> (uint64_t)0b11) \
+ : ((uint64_t)FLAG_END >> (uint64_t)0b11) + (uint64_t)1)
+namespace rocksdb{
+ void EnablePerfFlag(uint64_t flag);
+ void DisablePerfFlag(uint64_t flag);
+ bool CheckPerfFlag(uint64_t flag);
+}
+
+#define GET_FLAG(flag) perf_flags[(uint64_t)(flag) >> (uint64_t)0b11]
+#include "perf_flag_defs.h"

include/rocksdb/perf_flag_defs.h

@@ -0,0 +1,107 @@
+#pragma once
+#include <stdint.h>
+
+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_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,
+ FLAG_END
+};

monitoring/iostats_context_imp.h

@@ -34,13 +34,13 @@ extern __thread IOStatsContext iostats_context;
 
 // Declare and set start time of the timer
 #define IOSTATS_TIMER_GUARD(metric) \
- PerfStepTimer iostats_step_timer_##metric(&(iostats_context.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, env) \
  PerfStepTimer iostats_step_timer_##metric( \
- &(iostats_context.metric), env, true, \
+ &(iostats_context.metric), CheckPerfFlag(flag_##metric), env, true, \
  PerfLevel::kEnableTimeAndCPUTimeExceptForMutex); \
  iostats_step_timer_##metric.Start();
 

monitoring/perf_context_imp.h

@@ -38,24 +38,24 @@ extern thread_local PerfContext perf_context;
 
 // Declare and set start time of the timer
 #define PERF_TIMER_GUARD(metric) \
- PerfStepTimer perf_step_timer_##metric(&(perf_context.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_ENV(metric, env) \
- PerfStepTimer perf_step_timer_##metric(&(perf_context.metric), env); \
+ PerfStepTimer perf_step_timer_##metric(&(perf_context.metric),CheckPerfFlag(flag_##metric),env); \
  perf_step_timer_##metric.Start();
 
 // Declare and set start time of the timer
 #define PERF_CPU_TIMER_GUARD(metric, env) \
  PerfStepTimer perf_step_timer_##metric( \
- &(perf_context.metric), env, true, \
+ &(perf_context.metric),CheckPerfFlag(flag_##metric), env, 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, \
+ PerfStepTimer perf_step_timer_##metric(&(perf_context.metric),CheckPerfFlag(flag_##metric), nullptr, \
  false, PerfLevel::kEnableTime, stats, \
  ticker_type); \
  if (condition) { \
@@ -68,26 +68,25 @@ extern thread_local PerfContext perf_context;
 
 // Increase metric value
 #define PERF_COUNTER_ADD(metric, value) \
- if (perf_level >= PerfLevel::kEnableCount) { \
+ 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 && \
- perf_context.per_level_perf_context_enabled && \
- perf_context.level_to_perf_context) { \
- if ((*(perf_context.level_to_perf_context)).find(level) != \
- (*(perf_context.level_to_perf_context)).end()) { \
- (*(perf_context.level_to_perf_context))[level].metric += value; \
- } \
- else { \
- PerfContextByLevel empty_context; \
- (*(perf_context.level_to_perf_context))[level] = empty_context; \
- (*(perf_context.level_to_perf_context))[level].metric += value; \
- } \
- } \
+#define PERF_COUNTER_BY_LEVEL_ADD(metric, value, level) \
+ 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) != \
+ (*(perf_context.level_to_perf_context)).end()) { \
+ (*(perf_context.level_to_perf_context))[level].metric += value; \
+ } else { \
+ PerfContextByLevel empty_context; \
+ (*(perf_context.level_to_perf_context))[level] = empty_context; \
+ (*(perf_context.level_to_perf_context))[level].metric += value; \
+ } \
+ }
 
 #endif
 
-}
+} // namespace rocksdb

monitoring/perf_flag_imp.h

@@ -0,0 +1,10 @@
+#include <cstdint>
+#include "rocksdb/perf_flag.h"
+
+namespace rocksdb {
+#ifdef ROCKSDB_SUPPORT_THREAD_LOCAL
+extern __thread uint8_t perf_flags[FLAGS_LEN];
+#else
+extern uint8_t perf_flags[FLAGS_LEN];
+#endif
+}

monitoring/perf_step_timer.h

@@ -5,6 +5,7 @@
 //
 #pragma once
 #include "monitoring/perf_level_imp.h"
+#include "monitoring/perf_flag_imp.h"
 #include "rocksdb/env.h"
 #include "util/stop_watch.h"
 
@@ -13,10 +14,10 @@ namespace rocksdb {
 class PerfStepTimer {
  public:
  explicit PerfStepTimer(
- uint64_t* metric, Env* env = nullptr, bool use_cpu_time = false,
+ uint64_t* metric, bool enable_flag = false, Env* env = 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),
  env_((perf_counter_enabled_ || statistics != nullptr)
  ? ((env != nullptr) ? env : Env::Default())