fix: backup failed during the cluster creation phase (#9705)

(cherry picked from commit d9d5a9e)

Author: 1aal

controllers/dataprotection/backup_controller.go

@@ -283,10 +283,12 @@ func (r *BackupReconciler) handleDeletingPhase(reqCtx intctrlutil.RequestCtx, ba
 func (r *BackupReconciler) handleNewPhase(
 	reqCtx intctrlutil.RequestCtx,
 	backup *dpv1alpha1.Backup) (ctrl.Result, error) {
+
 	request, err := r.prepareBackupRequest(reqCtx, backup)
 	if err != nil {
 		return r.updateStatusIfFailed(reqCtx, backup.DeepCopy(), backup, err)
 	}
+
 	// record the status.target/status.targets infos for continuous backup.
 	if err = r.recordBackupStatusTargets(reqCtx, request); err != nil {
 		return r.updateStatusIfFailed(reqCtx, backup, request.Backup, err)
@@ -1083,6 +1085,10 @@ func setClusterSnapshotAnnotation(request *dpbackup.Request, cluster *appsv1alph
 
 // validateContinuousBackup validates the continuous backup.
 func validateContinuousBackup(backup *dpv1alpha1.Backup, reqCtx intctrlutil.RequestCtx, cli client.Client) error {
+	// filter the cluster that is creating when the backup is continuous
+	if clusterIsCreating(reqCtx, backup, cli) {
+		return intctrlutil.NewErrorf(intctrlutil.ErrorTypeRequeue, "requeue to wait for cluster %s creation to finish", backup.Labels[constant.AppInstanceLabelKey])
+	}
 	// validate if the continuous backup is created by a backupSchedule.
 	if _, ok := backup.Labels[dptypes.BackupScheduleLabelKey]; !ok {
 		return fmt.Errorf("continuous backup is only allowed to be created by backupSchedule")
@@ -1136,3 +1142,20 @@ func prepare4Incremental(request *dpbackup.Request) (*dpbackup.Request, error) {
 	}
 	return request, nil
 }
+
+// clusterIsCreating will return true when the backup is Continuous backup and the cluster is creating
+func clusterIsCreating(reqCtx intctrlutil.RequestCtx, backup *dpv1alpha1.Backup, cli client.Client) bool {
+	if backup.Labels[constant.AppInstanceLabelKey] != "" {
+		ownerCluster := &kbappsv1.Cluster{}
+		if err := cli.Get(reqCtx.Ctx, types.NamespacedName{
+			Namespace: backup.Namespace,
+			Name:      backup.Labels[constant.AppInstanceLabelKey],
+		}, ownerCluster); err != nil {
+			return false
+		}
+		if ownerCluster.Status.Phase == kbappsv1.CreatingClusterPhase {
+			return true
+		}
+	}
+	return false
+}

pkg/dataprotection/backup/utils.go

@@ -30,6 +30,7 @@ import (
 	"golang.org/x/mod/semver"
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/utils/pointer"
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -335,8 +336,10 @@ func StopStatefulSetsWhenFailed(ctx context.Context, cli client.Client, backup *
 	}
 	sts := &appsv1.StatefulSet{}
 	stsName := GenerateBackupStatefulSetName(backup, targetName, BackupDataJobNamePrefix)
-	if err := cli.Get(ctx, client.ObjectKey{Name: stsName, Namespace: backup.Namespace}, sts); client.IgnoreNotFound(err) != nil {
+	if err := cli.Get(ctx, client.ObjectKey{Name: stsName, Namespace: backup.Namespace}, sts); apierrors.IsNotFound(err) {
 		return nil
+	} else if err != nil {
+		return err
 	}
 	sts.Spec.Replicas = pointer.Int32(0)
 	return cli.Update(ctx, sts)