FMR

be/src/util/threadpool.cpp

@@ -46,10 +46,8 @@ DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(thread_pool_queue_size, MetricUnit::NOUNIT);
 DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(thread_pool_max_queue_size, MetricUnit::NOUNIT);
 DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(thread_pool_max_threads, MetricUnit::NOUNIT);
 DEFINE_COUNTER_METRIC_PROTOTYPE_2ARG(thread_pool_submit_failed, MetricUnit::NOUNIT);
-DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(task_execution_time_ns_avg_in_last_1000_times,
-                                   MetricUnit::NANOSECONDS);
-DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(task_wait_worker_ns_avg_in_last_1000_times,
-                                   MetricUnit::NANOSECONDS);
+DEFINE_COUNTER_METRIC_PROTOTYPE_2ARG(task_execution_time_ns_total, MetricUnit::NANOSECONDS);
+DEFINE_COUNTER_METRIC_PROTOTYPE_2ARG(task_wait_worker_time_ns_total, MetricUnit::NANOSECONDS);
 using namespace ErrorCode;
 
 using std::string;
@@ -311,8 +309,8 @@ Status ThreadPool::init() {
     INT_GAUGE_METRIC_REGISTER(_metric_entity, thread_pool_max_threads);
     INT_GAUGE_METRIC_REGISTER(_metric_entity, thread_pool_queue_size);
     INT_GAUGE_METRIC_REGISTER(_metric_entity, thread_pool_max_queue_size);
-    INT_GAUGE_METRIC_REGISTER(_metric_entity, task_execution_time_ns_avg_in_last_1000_times);
-    INT_GAUGE_METRIC_REGISTER(_metric_entity, task_wait_worker_ns_avg_in_last_1000_times);
+    INT_COUNTER_METRIC_REGISTER(_metric_entity, task_execution_time_ns_total);
+    INT_COUNTER_METRIC_REGISTER(_metric_entity, task_wait_worker_time_ns_total);
     INT_COUNTER_METRIC_REGISTER(_metric_entity, thread_pool_submit_failed);
 
     _metric_entity->register_hook("update", [this]() {
@@ -327,14 +325,6 @@ Status ThreadPool::init() {
         thread_pool_queue_size->set_value(get_queue_size());
         thread_pool_max_queue_size->set_value(get_max_queue_size());
         thread_pool_max_threads->set_value(max_threads());
-        task_execution_time_ns_avg_in_last_1000_times->set_value(
-                _task_execution_time_ns_statistic.mean());
-        task_wait_worker_ns_avg_in_last_1000_times->set_value(
-                _task_wait_worker_time_ns_statistic.mean());
-
-        // Remove older data from the histogram so that we can see curve could be more smooth.
-        _task_execution_time_ns_statistic.reduce_size(1);
-        _task_wait_worker_time_ns_statistic.reduce_size(1);
     });
     return Status::OK();
 }
@@ -606,7 +596,7 @@ void ThreadPool::dispatch_thread() {
         DCHECK_EQ(ThreadPoolToken::State::RUNNING, token->state());
         DCHECK(!token->_entries.empty());
         Task task = std::move(token->_entries.front());
-        _task_wait_worker_time_ns_statistic.add(task.submit_time_wather.elapsed_time());
+        task_wait_worker_time_ns_total->increment(task.submit_time_wather.elapsed_time());
         token->_entries.pop_front();
         token->_active_threads++;
         --_total_queued_tasks;
@@ -624,7 +614,7 @@ void ThreadPool::dispatch_thread() {
         // with this threadpool, and produce a deadlock.
         task.runnable.reset();
         l.lock();
-        _task_execution_time_ns_statistic.add(task_execution_time_watch.elapsed_time());
+        task_execution_time_ns_total->increment(task_execution_time_watch.elapsed_time());
         // Possible states:
         // 1. The token was shut down while we ran its task. Transition to QUIESCED.
         // 2. The token has no more queued tasks. Transition back to IDLE.

be/src/util/threadpool.h

@@ -40,7 +40,6 @@
 
 #include "agent/cgroup_cpu_ctl.h"
 #include "common/status.h"
-#include "util/interval_histogram.h"
 #include "util/metrics.h"
 #include "util/uid_util.h"
 #include "util/work_thread_pool.hpp"
@@ -414,11 +413,8 @@ class ThreadPool {
     IntGauge* thread_pool_queue_size = nullptr;
     IntGauge* thread_pool_max_queue_size = nullptr;
     IntGauge* thread_pool_max_threads = nullptr;
-    IntGauge* task_execution_time_ns_avg_in_last_1000_times = nullptr;
-    IntGauge* task_wait_worker_ns_avg_in_last_1000_times = nullptr;
-
-    IntervalHistogramStat<int64_t> _task_execution_time_ns_statistic {1000};
-    IntervalHistogramStat<int64_t> _task_wait_worker_time_ns_statistic {1000};
+    IntCounter* task_execution_time_ns_total = nullptr;
+    IntCounter* task_wait_worker_time_ns_total = nullptr;
 
     IntCounter* thread_pool_submit_failed = nullptr;
 };