alloc_runner: stop sidecars in separate loop

Author: DerekStrickland

client/allocrunner/alloc_runner.go

@@ -604,56 +604,80 @@ func (ar *allocRunner) handleTaskStateUpdates() {
 func (ar *allocRunner) killTasks() map[string]*structs.TaskState {
 	var mu sync.Mutex
 	states := make(map[string]*structs.TaskState, len(ar.tasks))
-
+ 
 	// run alloc prekill hooks
 	ar.preKillHooks()
-
+ 
 	// Kill leader first, synchronously
 	for name, tr := range ar.tasks {
-		if !tr.IsLeader() {
-			continue
-		}
-
-		taskEvent := structs.NewTaskEvent(structs.TaskKilling)
-		taskEvent.SetKillTimeout(tr.Task().KillTimeout)
-		err := tr.Kill(context.TODO(), taskEvent)
-		if err != nil && err != taskrunner.ErrTaskNotRunning {
-			ar.logger.Warn("error stopping leader task", "error", err, "task_name", name)
-		}
-
-		taskState := tr.TaskState()
-		states[name] = taskState
-		break
+	   if !tr.IsLeader() {
+		  continue
+	   }
+ 
+	   taskEvent := structs.NewTaskEvent(structs.TaskKilling)
+	   taskEvent.SetKillTimeout(tr.Task().KillTimeout)
+	   err := tr.Kill(context.TODO(), taskEvent)
+	   if err != nil && err != taskrunner.ErrTaskNotRunning {
+		  ar.logger.Warn("error stopping leader task", "error", err, "task_name", name)
+	   }
+ 
+	   taskState := tr.TaskState()
+	   states[name] = taskState
+	   break
 	}
-
+ 
 	// Kill the rest non-sidecar or poststop tasks concurrently
 	wg := sync.WaitGroup{}
 	for name, tr := range ar.tasks {
-		// Filter out poststop and sidecar tasks so that they stop after all the other tasks are killed
-		if tr.IsLeader() || tr.IsPoststopTask() || tr.IsSidecarTask() {
-			continue
-		}
-
-		wg.Add(1)
-		go func(name string, tr *taskrunner.TaskRunner) {
-			defer wg.Done()
-			taskEvent := structs.NewTaskEvent(structs.TaskKilling)
-			taskEvent.SetKillTimeout(tr.Task().KillTimeout)
-			err := tr.Kill(context.TODO(), taskEvent)
-			if err != nil && err != taskrunner.ErrTaskNotRunning {
-				ar.logger.Warn("error stopping task", "error", err, "task_name", name)
-			}
-
-			taskState := tr.TaskState()
-			mu.Lock()
-			states[name] = taskState
-			mu.Unlock()
-		}(name, tr)
+	   // Filter out poststop and sidecar tasks so that they stop after all the other tasks are killed
+	   if tr.IsLeader() || tr.IsPoststopTask() || tr.IsSidecarTask() {
+		  continue
+	   }
+ 
+	   wg.Add(1)
+	   go func(name string, tr *taskrunner.TaskRunner) {
+		  defer wg.Done()
+		  taskEvent := structs.NewTaskEvent(structs.TaskKilling)
+		  taskEvent.SetKillTimeout(tr.Task().KillTimeout)
+		  err := tr.Kill(context.TODO(), taskEvent)
+		  if err != nil && err != taskrunner.ErrTaskNotRunning {
+			 ar.logger.Warn("error stopping task", "error", err, "task_name", name)
+		  }
+ 
+		  taskState := tr.TaskState()
+		  mu.Lock()
+		  states[name] = taskState
+		  mu.Unlock()
+	   }(name, tr)
 	}
 	wg.Wait()
-
+ 
+	// Kill the sidecar tasks last.
+	for name, tr := range ar.tasks {
+	   if !tr.IsSidecarTask() || tr.IsLeader() || tr.IsPoststopTask() {
+		  continue
+	   }
+ 
+	   wg.Add(1)
+	   go func(name string, tr *taskrunner.TaskRunner) {
+		  defer wg.Done()
+		  taskEvent := structs.NewTaskEvent(structs.TaskKilling)
+		  taskEvent.SetKillTimeout(tr.Task().KillTimeout)
+		  err := tr.Kill(context.TODO(), taskEvent)
+		  if err != nil && err != taskrunner.ErrTaskNotRunning {
+			 ar.logger.Warn("error stopping task", "error", err, "task_name", name)
+		  }
+ 
+		  taskState := tr.TaskState()
+		  mu.Lock()
+		  states[name] = taskState
+		  mu.Unlock()
+	   }(name, tr)
+	}
+	wg.Wait()
+ 
 	return states
-}
+ }
 
 // clientAlloc takes in the task states and returns an Allocation populated
 // with Client specific fields