dxf: fix the task num count incorrectly (#57127)

close #57172
pingcap · Nov 7, 2024 · 559f218 · 559f218
1 parent 120b36b
commit 559f218
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 9 deletions.
diff --git a/pkg/disttask/framework/scheduler/scheduler.go b/pkg/disttask/framework/scheduler/scheduler.go
@@ -29,7 +29,6 @@ import (
 	"github.com/pingcap/tidb/pkg/disttask/framework/storage"
 	"github.com/pingcap/tidb/pkg/domain/infosync"
 	"github.com/pingcap/tidb/pkg/kv"
-	"github.com/pingcap/tidb/pkg/metrics"
 	"github.com/pingcap/tidb/pkg/sessionctx"
 	"github.com/pingcap/tidb/pkg/util/backoff"
 	disttaskutil "github.com/pingcap/tidb/pkg/util/disttask"
@@ -170,6 +169,8 @@ func (s *BaseScheduler) scheduleTask() {
 				if errors.Cause(err) == storage.ErrTaskNotFound {
 					// this can happen when task is reverted/succeed, but before
 					// we reach here, cleanup routine move it to history.
+					s.logger.Debug("task not found, might be reverted/succeed/failed", zap.Int64("task_id", s.GetTask().ID),
+						zap.String("task_key", s.GetTask().Key))
 					return
 				}
 				s.logger.Error("refresh task failed", zap.Error(err))
@@ -399,7 +400,6 @@ func (s *BaseScheduler) onRunning() error {
 
 func (s *BaseScheduler) onFinished() {
 	task := s.GetTask()
-	metrics.UpdateMetricsForFinishTask(task)
 	s.logger.Debug("schedule task, task is finished", zap.Stringer("state", task.State))
 }
 

diff --git a/pkg/disttask/framework/scheduler/scheduler_manager.go b/pkg/disttask/framework/scheduler/scheduler_manager.go
@@ -285,8 +285,7 @@ func (sm *Manager) startSchedulers(schedulableTasks []*proto.TaskBase) error {
 				zap.Int64("task-id", task.ID), zap.Stringer("state", task.State))
 		}
 
-		metrics.DistTaskGauge.WithLabelValues(task.Type.String(), metrics.SchedulingStatus).Inc()
-		metrics.UpdateMetricsForScheduleTask(task.ID, task.Type)
+		metrics.UpdateMetricsForScheduleTask(task)
 		sm.startScheduler(task, allocateSlots, reservedExecID)
 	}
 	return nil
@@ -437,6 +436,7 @@ func (sm *Manager) cleanupFinishedTasks(tasks []*proto.Task) error {
 			// if task doesn't register cleanup function, mark it as cleaned.
 			cleanedTasks = append(cleanedTasks, task)
 		}
+		metrics.UpdateMetricsForFinishTask(task)
 	}
 	if firstErr != nil {
 		sm.logger.Warn("cleanup routine failed", zap.Error(errors.Trace(firstErr)))

diff --git a/pkg/metrics/disttask.go b/pkg/metrics/disttask.go
@@ -80,10 +80,11 @@ func UpdateMetricsForAddTask(task *proto.TaskBase) {
 }
 
 // UpdateMetricsForScheduleTask update metrics when a task is added
-func UpdateMetricsForScheduleTask(id int64, taskType proto.TaskType) {
-	DistTaskGauge.WithLabelValues(taskType.String(), WaitingStatus).Dec()
-	DistTaskStartTimeGauge.DeleteLabelValues(taskType.String(), WaitingStatus, fmt.Sprint(id))
-	DistTaskStartTimeGauge.WithLabelValues(taskType.String(), SchedulingStatus, fmt.Sprint(id)).SetToCurrentTime()
+func UpdateMetricsForScheduleTask(task *proto.TaskBase) {
+	DistTaskGauge.WithLabelValues(task.Type.String(), WaitingStatus).Dec()
+	DistTaskGauge.WithLabelValues(task.Type.String(), SchedulingStatus).Inc()
+	DistTaskStartTimeGauge.DeleteLabelValues(task.Type.String(), WaitingStatus, fmt.Sprint(task.ID))
+	DistTaskStartTimeGauge.WithLabelValues(task.Type.String(), SchedulingStatus, fmt.Sprint(task.ID)).SetToCurrentTime()
 }
 
 // UpdateMetricsForRunTask update metrics when a task starts running

diff --git a/pkg/metrics/grafana/tidb.json b/pkg/metrics/grafana/tidb.json
@@ -16487,7 +16487,7 @@
           "targets": [
             {
               "exemplar": true,
-              "expr": "sum(tidb_disttask_subtasks{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", status=~\"pending|running|reverting|revert_pending|paused\"}) by (exec_id)",
+              "expr": "sum(tidb_disttask_subtasks{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", status=~\"pending|running|failed|canceled|paused\"}) by (exec_id)",
               "interval": "",
               "legendFormat": "",
               "queryType": "randomWalk",