v2 volume: prevent from errors in spdk_tgt when deleting a volume

To prevent from the "no such device" error in spdk_tgt when deleting a volume, remove the raid bdev before tearing down the replicas. Longhorn 7568 Signed-off-by: Derek Su <derek.su@suse.com>
c3y1huang · Jan 8, 2024 · 1a206d9 · 1a206d9
1 parent e5f7071
commit 1a206d9
Show file tree

Hide file tree

Showing 2 changed files with 24 additions and 1 deletion.
diff --git a/controller/setting_controller.go b/controller/setting_controller.go
@@ -840,7 +840,7 @@ func (sc *SettingController) cleanupInstanceManager(dataEngine longhorn.DataEngi
 	}
 
 	for _, im := range imMap {
-		sc.logger.Infof("Cleaning up the instance manager for %v data engine", dataEngine)
+		sc.logger.Infof("Cleaning up the instance manager %v for %v data engine", im.Name, dataEngine)
 		if err := sc.ds.DeleteInstanceManager(im.Name); err != nil {
 			return err
 		}

diff --git a/controller/volume_controller.go b/controller/volume_controller.go
@@ -350,6 +350,29 @@ func (c *VolumeController) syncVolume(key string) (err error) {
 				}
 			}
 		}
+		if datastore.IsDataEngineV2(volume.Spec.DataEngine) {
+			// To prevent from the "no such device" error in spdk_tgt,
+			// remove the raid bdev before tearing down the replicas.
+			engines, err := c.ds.ListVolumeEnginesRO(volume.Name)
+			if err != nil {
+				return err
+			}
+			if len(engines) > 0 {
+				replicaDeleted := false
+				for _, r := range replicas {
+					if r.DeletionTimestamp != nil {
+						replicaDeleted = true
+						break
+					}
+				}
+				if !replicaDeleted {
+					c.logger.Infof("Volume (%s) %v still has engines %v, so skip deleting its replicas until the engines have been deleted",
+						volume.Spec.DataEngine, volume.Name, engines)
+					return nil
+				}
+			}
+		}
+
 		for _, r := range replicas {
 			if r.DeletionTimestamp == nil {
 				if err := c.ds.DeleteReplica(r.Name); err != nil {