fix: opj should failed if ops not found pod (#308)

* ops-not-found-pod

* 优化e2e

* 优化ut

* 优化ut

* 优化ut

Author: ColdsteelRail

pkg/controllers/collaset/collaset_controller_test.go

@@ -119,7 +119,7 @@ var _ = Describe("collaset controller", func() {
 				return len(podList.Items) == 1
 			}, 5*time.Second, 1*time.Second).Should(BeTrue())
 			Expect(c.Get(context.TODO(), types.NamespacedName{Namespace: cs.Namespace, Name: cs.Name}, cs)).Should(BeNil())
-			Expect(expectedStatusReplicas(c, cs, 0, 0, 0, 1, 1, 0, 0, 0)).Should(BeNil())
+			Eventually(func() error { return expectedStatusReplicas(c, cs, 0, 0, 0, 1, 1, 0, 0, 0) }, 5*time.Second, 1*time.Second).Should(BeNil())
 
 			// label pod to trigger replace
 			originPod := podList.Items[0]

pkg/controllers/operationjob/operationjob_manager.go

@@ -118,6 +118,8 @@ func (r *ReconcileOperationJob) filterAndOperateAllowOpsTargets(
 		}
 
 		if enablePodOpsLifecycle && candidate.Pod == nil {
+			MarkCandidateFailed(candidate)
+			ojutils.SetOpsStatusError(candidate, appsv1alpha1.ReasonPodNotFound, "failed to operate a non-exist pod")
 			return nil
 		}
 

test/e2e/apps/collaset.go

@@ -1058,7 +1058,7 @@ var _ = SIGDescribe("CollaSet", func() {
 					return true
 				}
 				return false
-			}, 12*time.Second, time.Second).Should(BeTrue())
+			}, 12*time.Second, time.Microsecond).Should(BeTrue())
 			Expect(duration > time.Duration(*cls.Spec.UpdateStrategy.OperationDelaySeconds)*time.Second-4).Should(BeTrue())
 
 			By("Wait for update finished")