fix scheduler ha bug (#443)

pingcap · May 1, 2019 · f9166c3 · f9166c3
1 parent 4280261
commit f9166c3
Show file tree

Hide file tree

Showing 2 changed files with 51 additions and 7 deletions.
diff --git a/pkg/scheduler/predicates/ha.go b/pkg/scheduler/predicates/ha.go
@@ -200,13 +200,18 @@ func (h *ha) realAcquireLock(pod *apiv1.Pod) (*apiv1.PersistentVolumeClaim, *api
 	if schedulingPVC == currentPVC {
 		return schedulingPVC, currentPVC, nil
 	}
-	schedulingPodName := getPodNameFromPVC(schedulingPVC)
-	schedulingPod, err := h.podGetFn(ns, schedulingPodName)
-	if err != nil {
-		return schedulingPVC, currentPVC, err
-	}
-	if schedulingPVC.Status.Phase != apiv1.ClaimBound || schedulingPod.Spec.NodeName == "" {
-		return schedulingPVC, currentPVC, fmt.Errorf("waiting for Pod %s/%s scheduling", ns, strings.TrimPrefix(schedulingPVC.GetName(), component))
+
+	// if pvc is not defer deleting(has AnnPVCDeferDeleting annotation means defer deleting), we must wait for its scheduling
+	// else clear its AnnPVCPodScheduling annotation and acquire the lock
+	if schedulingPVC.Annotations[label.AnnPVCDeferDeleting] == "" {
+		schedulingPodName := getPodNameFromPVC(schedulingPVC)
+		schedulingPod, err := h.podGetFn(ns, schedulingPodName)
+		if err != nil {
+			return schedulingPVC, currentPVC, err
+		}
+		if schedulingPVC.Status.Phase != apiv1.ClaimBound || schedulingPod.Spec.NodeName == "" {
+			return schedulingPVC, currentPVC, fmt.Errorf("waiting for Pod %s/%s scheduling", ns, strings.TrimPrefix(schedulingPVC.GetName(), component))
+		}
 	}
 
 	delete(schedulingPVC.Annotations, label.AnnPVCPodScheduling)

diff --git a/pkg/scheduler/predicates/ha_test.go b/pkg/scheduler/predicates/ha_test.go
@@ -379,6 +379,45 @@ func TestHARealAcquireLockFn(t *testing.T) {
 				g.Expect(currentPVC.Annotations[label.AnnPVCPodScheduling]).NotTo(BeEmpty())
 			},
 		},
+		{
+			name:  "scheduling pvc is defer deleting, current pvc acquire lock",
+			podFn: newHAPDPod,
+			pvcListFn: func(ns, instanceName, component string) (*corev1.PersistentVolumeClaimList, error) {
+				return &corev1.PersistentVolumeClaimList{
+					TypeMeta: metav1.TypeMeta{Kind: "PersistentVolumeClaimList", APIVersion: "v1"},
+					Items: []corev1.PersistentVolumeClaim{
+						{
+							TypeMeta: metav1.TypeMeta{Kind: "PersistentVolumeClaim", APIVersion: "v1"},
+							ObjectMeta: metav1.ObjectMeta{
+								Namespace: metav1.NamespaceDefault,
+								Name:      "pd-cluster-1-pd-0",
+							},
+						},
+						{
+							TypeMeta: metav1.TypeMeta{Kind: "PersistentVolumeClaim", APIVersion: "v1"},
+							ObjectMeta: metav1.ObjectMeta{
+								Namespace: metav1.NamespaceDefault,
+								Name:      "pd-cluster-1-pd-1",
+								Annotations: map[string]string{
+									label.AnnPVCPodScheduling: "true",
+									label.AnnPVCDeferDeleting: "true",
+								},
+							},
+							Status: corev1.PersistentVolumeClaimStatus{Phase: corev1.ClaimBound},
+						},
+					},
+				}, nil
+			},
+			podGetFn: podGetErr(),
+			updatePVCFn: func(claim *corev1.PersistentVolumeClaim) error {
+				return nil
+			},
+			expectFn: func(schedulingPVC, currentPVC *apiv1.PersistentVolumeClaim, err error) {
+				g.Expect(err).NotTo(HaveOccurred())
+				g.Expect(schedulingPVC.Annotations[label.AnnPVCPodScheduling]).To(BeEmpty())
+				g.Expect(currentPVC.Annotations[label.AnnPVCPodScheduling]).NotTo(BeEmpty())
+			},
+		},
 	}
 
 	for i := range tests {