Merge branch 'v1.5.x' into mergify/bp/v1.5.x/pr-2635

controller/volume_controller.go

@@ -3116,18 +3116,24 @@ func (c *VolumeController) checkAndFinishVolumeRestore(v *longhorn.Volume, e *lo
 			return errors.Wrapf(err, "failed to get backup name from volume %s backup URL %v", v.Name, v.Spec.FromBackup)
 		}
 		bv, err := c.ds.GetBackupVolumeRO(bvName)
-		if err != nil {
+		if err != nil && !apierrors.IsNotFound(err) {
 			return err
 		}
-		if !bv.Status.LastSyncedAt.IsZero() &&
-			bv.Spec.SyncRequestedAt.After(bv.Status.LastSyncedAt.Time) {
-			log.Infof("Restore/DR volume needs to wait for backup volume %s update", bvName)
-			return nil
-		}
-		if e.Status.LastRestoredBackup != bv.Status.LastBackupName {
-			log.Infof("Restore/DR volume needs to restore the latest backup %s, and the current restored backup is %s", bv.Status.LastBackupName, e.Status.LastRestoredBackup)
-			c.enqueueVolume(v)
-			return nil
+		if bv != nil {
+			if bv.Status.LastBackupName != "" {
+				// If the backup CR does not exist, the Longhorn system may be still syncing up the info with the remote backup target.
+				// If the backup is removed already, the backup volume should receive the notification and update bv.Status.LastBackupName.
+				// Hence we cannot continue the activation when the backup get call returns error IsNotFound.
+				b, err := c.ds.GetBackup(bv.Status.LastBackupName)
+				if err != nil {
+					return err
+				}
+				if b.Name != e.Status.LastRestoredBackup {
+					log.Infof("Restore/DR volume needs to restore the latest backup %s, and the current restored backup is %s", b.Name, e.Status.LastRestoredBackup)
+					c.enqueueVolume(v)
+					return nil
+				}
+			}
 		}
 	}
 

manager/volume.go

@@ -11,6 +11,7 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 
 	corev1 "k8s.io/api/core/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
 	"github.com/longhorn/longhorn-manager/datastore"
@@ -447,6 +448,12 @@ func (m *VolumeManager) triggerBackupVolumeToSync(volume *longhorn.Volume) error
 
 	backupVolume, err := m.ds.GetBackupVolume(backupVolumeName)
 	if err != nil {
+		// The backup volume may be deleted already.
+		// hence it's better not to block the caller to continue the handlings like DR volume activation.
+		if apierrors.IsNotFound(err) {
+			logrus.Infof("Cannot find backup volume %v to trigger the sync-up, will skip it", backupVolumeName)
+			return nil
+		}
 		return errors.Wrapf(err, "failed to get backup volume: %v", backupVolumeName)
 	}
 	requestSyncTime := metav1.Time{Time: time.Now().UTC()}