fix: Allow deleting orphans while retaining the data

controller/node_controller.go

@@ -1326,7 +1326,7 @@ func (nc *NodeController) syncOrphans(node *longhorn.Node, collectedDataInfo map
 	return nc.deleteOrphansForEngineAndReplicaInstances(node)
 }
 
-func (nc *NodeController) getNewAndMissingOrphanedReplicaDataStores(diskName, diskUUID, diskPath string, replicaDataStores map[string]string) (map[string]string, map[string]string) {
+func (nc *NodeController) getNewAndMissingOrphanedReplicaDataStores(diskName, diskUUID, diskPath string, orphanedReplicaDataStores map[string]string) (map[string]string, map[string]string) {
 	newOrphanedReplicaDataStores := map[string]string{}
 	missingOrphanedReplicaDataStores := map[string]string{}
 
@@ -1343,13 +1343,19 @@ func (nc *NodeController) getNewAndMissingOrphanedReplicaDataStores(diskName, di
 		}
 	}
 
-	for dataStore := range replicaDataStores {
+	for dataStore := range orphanedReplicaDataStores {
 		orphanName := types.GetOrphanChecksumNameForOrphanedDataStore(nc.controllerID, diskName, diskPath, diskUUID, dataStore)
 		if _, ok := orphanMap[orphanName]; !ok {
 			newOrphanedReplicaDataStores[dataStore] = ""
 		}
 	}
 
+	replicas, err := nc.ds.ListReplicasByDiskUUIDRO(diskUUID)
+	if err != nil {
+		nc.logger.WithError(err).Warnf("Failed to list replicas for disk %v", diskUUID)
+		return map[string]string{}, map[string]string{}
+	}
+
 	for _, orphan := range orphanMap {
 		if orphan.Spec.Parameters[longhorn.OrphanDiskName] != diskName ||
 			orphan.Spec.Parameters[longhorn.OrphanDiskUUID] != diskUUID ||
@@ -1358,7 +1364,19 @@ func (nc *NodeController) getNewAndMissingOrphanedReplicaDataStores(diskName, di
 		}
 
 		dataStore := orphan.Spec.Parameters[longhorn.OrphanDataName]
-		if _, ok := replicaDataStores[dataStore]; !ok {
+		for _, r := range replicas {
+			if r.Spec.DataDirectoryName == dataStore {
+				if err := datastore.AddOrphanDeleteCustomResourceOnlyLabel(nc.ds, orphan.Name); err != nil {
+					nc.logger.Infof("failed to add label delete-custom-resource-only to orphan %v ", orphan.Name)
+					return map[string]string{}, map[string]string{}
+				}
+
+				break
+			}
+		}
+		_, ok := orphanedReplicaDataStores[dataStore]
+
+		if !ok {
 			missingOrphanedReplicaDataStores[dataStore] = ""
 		}
 	}

controller/orphan_controller.go

@@ -622,7 +622,6 @@ func (oc *OrphanController) updateDataCleanableCondition(orphan *longhorn.Orphan
 	if orphan.Spec.Type == longhorn.OrphanTypeReplicaData {
 		reason = oc.checkOrphanedReplicaDataCleanable(node, orphan)
 	}
-
 	return nil
 }
 
@@ -647,6 +646,11 @@ func (oc *OrphanController) checkOrphanedReplicaDataCleanable(node *longhorn.Nod
 		return longhorn.OrphanConditionTypeDataCleanableReasonDiskEvicted
 	}
 
+	exists, _ := datastore.IsLabelLonghornDeleteCustomResourceOnlyExisting(orphan)
+	if exists {
+		return longhorn.OrphanConditionTypeDataCleanableReasonSkipped
+	}
+
 	return ""
 }
 

datastore/longhorn.go

@@ -3484,6 +3484,18 @@ func (s *DataStore) ListReplicasByDiskUUID(uuid string) (map[string]*longhorn.Re
 	return s.listReplicas(diskSelector)
 }
 
+func (s *DataStore) ListReplicasByDiskUUIDRO(uuid string) ([]*longhorn.Replica, error) {
+	diskSelector, err := metav1.LabelSelectorAsSelector(&metav1.LabelSelector{
+		MatchLabels: map[string]string{
+			types.LonghornDiskUUIDKey: uuid,
+		},
+	})
+	if err != nil {
+		return nil, err
+	}
+	return s.replicaLister.Replicas(s.namespace).List(diskSelector)
+}
+
 func getBackingImageSelector(backingImageName, diskUUID string) (labels.Selector, error) {
 	matchLabels := map[string]string{
 		types.GetLonghornLabelKey(types.LonghornLabelBackingImage): backingImageName,
@@ -3732,6 +3744,27 @@ func AddSystemBackupDeleteCustomResourceOnlyLabel(ds *DataStore, systemBackupNam
 	return nil
 }
 
+// AddOrphanDeleteCustomResourceOnlyLabel adds the label `longhorn.io/delete-custom-resource-only: true` to the Orphan
+func AddOrphanDeleteCustomResourceOnlyLabel(ds *DataStore, OrphanName string) error {
+	orphan, err := ds.GetOrphan(OrphanName)
+	if err != nil {
+		return err
+	}
+	if exists, err := IsLabelLonghornDeleteCustomResourceOnlyExisting(orphan); err != nil {
+		return err
+	} else if exists {
+		return nil
+	}
+	if err := labelLonghornDeleteCustomResourceOnly(orphan); err != nil {
+		return err
+	}
+	if _, err = ds.UpdateOrphan(orphan); err != nil {
+		return err
+	}
+
+	return nil
+}
+
 func FixupRecurringJob(v *longhorn.Volume) error {
 	if v.Spec.CloneMode == longhorn.CloneModeLinkedClone {
 		// Do not add recurring job for linked-clone volume as these volumes do not support snapshot/backup operations

k8s/pkg/apis/longhorn/v1beta2/orphan.go

@@ -20,6 +20,7 @@ const (
 	OrphanConditionTypeDataCleanableReasonDiskInvalid     = "DiskInvalid"
 	OrphanConditionTypeDataCleanableReasonDiskEvicted     = "DiskEvicted"
 	OrphanConditionTypeDataCleanableReasonDiskChanged     = "DiskChanged"
+	OrphanConditionTypeDataCleanableReasonSkipped         = "Skipped"
 )
 
 const (