fix: Terminate, rather than delete, deadlined pods. Fixes #8545 (#8620)

Signed-off-by: Alex Collins <[email protected]>
argoproj · May 6, 2022 · 859ebe9 · 859ebe9
1 parent dd56520
commit 859ebe9
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 40 deletions.
diff --git a/workflow/controller/controller.go b/workflow/controller/controller.go
@@ -468,24 +468,6 @@ func (wfc *WorkflowController) processNextPodCleanupItem(ctx context.Context) bo
 	err := func() error {
 		pods := wfc.kubeclientset.CoreV1().Pods(namespace)
 		switch action {
-		case shutdownPod:
-			// to shutdown a pod, we signal the wait container to terminate, the wait container in turn will
-			// kill the main container (using whatever mechanism the executor uses), and will then exit itself
-			// once the main container exited
-			pod, err := wfc.getPod(namespace, podName)
-			if pod == nil || err != nil {
-				return err
-			}
-			for _, c := range pod.Spec.Containers {
-				if c.Name == common.WaitContainerName {
-					if err := signal.SignalContainer(wfc.restConfig, pod, common.WaitContainerName, syscall.SIGTERM); err != nil {
-						return err
-					}
-					return nil // done
-				}
-			}
-			// no wait container found
-			fallthrough
 		case terminateContainers:
 			if terminationGracePeriod, err := wfc.signalContainers(namespace, podName, syscall.SIGTERM); err != nil {
 				return err

diff --git a/workflow/controller/exec_control.go b/workflow/controller/exec_control.go
@@ -7,7 +7,6 @@ import (
 	"time"
 
 	apiv1 "k8s.io/api/core/v1"
-	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
 	wfv1 "github.com/argoproj/argo-workflows/v3/pkg/apis/workflow/v1alpha1"
 	"github.com/argoproj/argo-workflows/v3/workflow/common"
@@ -37,15 +36,13 @@ func (woc *wfOperationCtx) applyExecutionControl(ctx context.Context, pod *apiv1
 			_, onExitPod := pod.Labels[common.LabelKeyOnExit]
 
 			if !woc.GetShutdownStrategy().ShouldExecute(onExitPod) {
-				woc.log.Infof("Deleting Pending pod %s/%s as part of workflow shutdown with strategy: %s", pod.Namespace, pod.Name, woc.GetShutdownStrategy())
-				err := woc.controller.kubeclientset.CoreV1().Pods(pod.Namespace).Delete(ctx, pod.Name, metav1.DeleteOptions{})
-				if err == nil {
-					msg := fmt.Sprintf("workflow shutdown with strategy:  %s", woc.GetShutdownStrategy())
-					woc.handleExecutionControlError(nodeID, wfNodesLock, msg)
-					return
-				}
-				// If we fail to delete the pod, fall back to setting the annotation
-				woc.log.Warnf("Failed to delete %s/%s: %v", pod.Namespace, pod.Name, err)
+				woc.log.WithField("podName", pod.Name).
+					WithField("shutdownStrategy", woc.GetShutdownStrategy()).
+					Info("Terminating pod as part of workflow shutdown")
+				woc.controller.queuePodForCleanup(pod.Namespace, pod.Name, terminateContainers)
+				msg := fmt.Sprintf("workflow shutdown with strategy:  %s", woc.GetShutdownStrategy())
+				woc.handleExecutionControlError(nodeID, wfNodesLock, msg)
+				return
 			}
 		}
 		// Check if we are past the workflow deadline. If we are, and the pod is still pending
@@ -54,21 +51,20 @@ func (woc *wfOperationCtx) applyExecutionControl(ctx context.Context, pod *apiv1
 			// pods that are part of an onExit handler aren't subject to the deadline
 			_, onExitPod := pod.Labels[common.LabelKeyOnExit]
 			if !onExitPod {
-				woc.log.Infof("Deleting Pending pod %s/%s which has exceeded workflow deadline %s", pod.Namespace, pod.Name, woc.workflowDeadline)
-				err := woc.controller.kubeclientset.CoreV1().Pods(pod.Namespace).Delete(ctx, pod.Name, metav1.DeleteOptions{})
-				if err == nil {
-					woc.handleExecutionControlError(nodeID, wfNodesLock, "Step exceeded its deadline")
-					return
-				}
-				// If we fail to delete the pod, fall back to setting the annotation
-				woc.log.Warnf("Failed to delete %s/%s: %v", pod.Namespace, pod.Name, err)
+				woc.log.WithField("podName", pod.Name).
+					WithField(" workflowDeadline", woc.workflowDeadline).
+					Info("Terminating pod which has exceeded workflow deadline")
+				woc.controller.queuePodForCleanup(pod.Namespace, pod.Name, terminateContainers)
+				woc.handleExecutionControlError(nodeID, wfNodesLock, "Step exceeded its deadline")
+				return
 			}
 		}
 	}
 	if woc.GetShutdownStrategy().Enabled() {
 		if _, onExitPod := pod.Labels[common.LabelKeyOnExit]; !woc.GetShutdownStrategy().ShouldExecute(onExitPod) {
-			woc.log.Infof("Shutting down pod %s", pod.Name)
-			woc.controller.queuePodForCleanup(woc.wf.Namespace, pod.Name, shutdownPod)
+			woc.log.WithField("podName", pod.Name).
+				Info("Terminating on-exit pod")
+			woc.controller.queuePodForCleanup(woc.wf.Namespace, pod.Name, terminateContainers)
 		}
 	}
 }
@@ -101,7 +97,7 @@ func (woc *wfOperationCtx) killDaemonedChildren(nodeID string) {
 		if !childNode.IsDaemoned() {
 			continue
 		}
-		woc.controller.queuePodForCleanup(woc.wf.Namespace, childNode.ID, shutdownPod)
+		woc.controller.queuePodForCleanup(woc.wf.Namespace, childNode.ID, terminateContainers)
 		childNode.Phase = wfv1.NodeSucceeded
 		childNode.Daemoned = nil
 		woc.wf.Status.Nodes[childNode.ID] = childNode

diff --git a/workflow/controller/pod_cleanup_key.go b/workflow/controller/pod_cleanup_key.go
@@ -16,7 +16,6 @@ type (
 
 const (
 	deletePod           podCleanupAction = "deletePod"
-	shutdownPod         podCleanupAction = "shutdownPod"
 	labelPodCompleted   podCleanupAction = "labelPodCompleted"
 	terminateContainers podCleanupAction = "terminateContainers"
 	killContainers      podCleanupAction = "killContainers"