From 2c0587d6daaf6a47b19091084057be5f21fd6868 Mon Sep 17 00:00:00 2001
From: shuangkun <tsk2013uestc@163.com>
Date: Tue, 9 Apr 2024 20:41:01 +0800
Subject: [PATCH] fix: retry status

Signed-off-by: shuangkun <tsk2013uestc@163.com>
---
 workflow/controller/controller.go      | 12 ++++++++++++
 workflow/controller/operator.go        | 17 ++++++++---------
 workflow/controller/pod_cleanup_key.go | 10 +++++-----
 3 files changed, 25 insertions(+), 14 deletions(-)

diff --git a/workflow/controller/controller.go b/workflow/controller/controller.go
index 0db3adee61ce..0f99464afb36 100644
--- a/workflow/controller/controller.go
+++ b/workflow/controller/controller.go
@@ -567,6 +567,18 @@ func (wfc *WorkflowController) processNextPodCleanupItem(ctx context.Context) bo
 			if err != nil && !apierr.IsNotFound(err) {
 				return err
 			}
+		case batchDeletePods:
+			workflowName := podName
+			wfClient := wfc.wfclientset.ArgoprojV1alpha1().Workflows(namespace)
+			wf, err := wfClient.Get(ctx, workflowName, metav1.GetOptions{})
+			if err != nil && !apierr.IsNotFound(err) {
+				return err
+			}
+			wf.ObjectMeta.Labels[common.LabelKeyWorkflowRetryStatus] = "Retried"
+			wf, err = wfClient.Update(ctx, wf, metav1.UpdateOptions{})
+			if err != nil && !apierr.IsNotFound(err) {
+				return err
+			}
 		}
 		return nil
 	}()
diff --git a/workflow/controller/operator.go b/workflow/controller/operator.go
index 1d37df185ebf..aa4c44740b43 100644
--- a/workflow/controller/operator.go
+++ b/workflow/controller/operator.go
@@ -2379,7 +2379,6 @@ func (woc *wfOperationCtx) markWorkflowPhase(ctx context.Context, phase wfv1.Wor
 			}
 		}
 		woc.updated = true
-		woc.wf.Status.RetryStatus = nil
 		woc.controller.queuePodForCleanup(woc.wf.Namespace, woc.getAgentPodName(), deletePod)
 	}
 }
@@ -3842,15 +3841,11 @@ func (woc *wfOperationCtx) shouldRetry() bool {
 	if !ok || retryStatus == "Retried" {
 		return false
 	}
-	if retryStatus == "Retrying" {
-		// TODO make sure all pod in podsToDelete deleted, avoid "create pod exists"
-		return false
-	}
 	return true
 }
 
 func (woc *wfOperationCtx) IsRetried() bool {
-	return woc.wf.Labels[common.LabelKeyWorkflowRetried] != "Retried"
+	return woc.wf.ObjectMeta.Labels[common.LabelKeyWorkflowRetryStatus] != "Retried"
 }
 
 func (woc *wfOperationCtx) retryWorkflow(ctx context.Context) error {
@@ -3864,7 +3859,11 @@ func (woc *wfOperationCtx) retryWorkflow(ctx context.Context) error {
 	if err != nil {
 		return fmt.Errorf("fail to unmarshaling parameters: %v", err)
 	}
-	restartSuccessful := woc.wf.Labels[common.LabelKeyRetryRestartSuccessful]
+	restartSuccessful := false
+	restartSuccessfulStr := woc.wf.Labels[common.LabelKeyRetryRestartSuccessful]
+	if restartSuccessfulStr == "true" {
+		restartSuccessful = true
+	}
 
 	// Clean up remaining pods in the workflow
 	wf, podsToDelete, err := wfutil.FormulateRetryWorkflow(ctx, woc.wf, restartSuccessful, nodeFiledSelector, parameters)
@@ -3874,9 +3873,9 @@ func (woc *wfOperationCtx) retryWorkflow(ctx context.Context) error {
 	for _, podName := range podsToDelete {
 		woc.controller.queuePodForCleanup(wf.Namespace, podName, deletePod)
 	}
-	woc.controller.queuePodForCleanup(wf.Namespace, wf.Name, deletedAllPodsFlag)
+	woc.controller.queuePodForCleanup(wf.Namespace, wf.Name, batchDeletePods)
 	woc.wf = wf
-	woc.wf.labels[common.LabelKeyWorkflowRetryStatus] = "Retrying"
+	woc.wf.ObjectMeta.Labels[common.LabelKeyWorkflowRetryStatus] = "Retrying"
 	woc.updated = true
 	return nil
 }
diff --git a/workflow/controller/pod_cleanup_key.go b/workflow/controller/pod_cleanup_key.go
index 42328c67d71f..e68a4e3f1130 100644
--- a/workflow/controller/pod_cleanup_key.go
+++ b/workflow/controller/pod_cleanup_key.go
@@ -15,11 +15,11 @@ type (
 )
 
 const (
-	deletePod              podCleanupAction = "deletePod"
-	labelPodCompleted      podCleanupAction = "labelPodCompleted"
-	terminateContainers    podCleanupAction = "terminateContainers"
-	killContainers         podCleanupAction = "killContainers"
-	showDeletedAllPodsFlag podCleanupAction = "deletedAllPods"
+	deletePod           podCleanupAction = "deletePod"
+	labelPodCompleted   podCleanupAction = "labelPodCompleted"
+	terminateContainers podCleanupAction = "terminateContainers"
+	killContainers      podCleanupAction = "killContainers"
+	batchDeletePods     podCleanupAction = "batchDeletePods"
 )
 
 func newPodCleanupKey(namespace string, podName string, action podCleanupAction) podCleanupKey {