fix(linstor): handle volume creation failure in clone rollback

A clone rollback may be initiated because the new volume could not be
physically created. This can be detected by checking the `not-exists`
property of this volume.

If this property tells us that the new volume doesn't actually exist,
then we mark it for later deletion by the GC so that rolling back the
UUID of the base VDI doesn't fail later on.

Signed-off-by: Alexandre Sollier <alexandre.sollier@vates.tech>

Author: Kuruyia

drivers/LinstorSR.py

@@ -1481,6 +1481,18 @@ def _undo_clone(self, volume_names, vdi_uuid, base_uuid, snap_uuid):
                 # properly the base VDI below this line, so we must change the
                 # UUID of this bad VDI before.
                 self._linstor.mark_volume_for_deletion(vdi_uuid, force=True)
+        else:
+            not_exists_state = self._linstor.get_volume_not_exists_state(vdi_uuid)
+
+            if not_exists_state and not_exists_state != LinstorVolumeManager.STATE_EXISTS:
+                # At this point, some of the new volume properties were created,
+                # but volume creation was interrupted before the volume itself
+                # could be physically created.
+                # We mark this new volume for deletion so that the base VDI UUID
+                # can be restored back to its original value.
+                util.SMlog(f"VDI {vdi_uuid} was partially created, marking it for deletion")
+
+                self._linstor.mark_volume_for_deletion(vdi_uuid, force=True)
 
         # Rename!
         self._linstor.update_volume_uuid(base_uuid, vdi_uuid)

drivers/linstorvolumemanager.py

@@ -1344,6 +1344,20 @@ def update_volume_metadata(self, volume_uuid, metadata):
             current_metadata[key] = value
         volume_properties[self.PROP_METADATA] = json.dumps(current_metadata)
 
+    def get_volume_not_exists_state(self, volume_uuid):
+        """
+        Get the "not-exists" state of a volume.
+        :param str volume_uuid: The target volume.
+        :return: The "not-exists" state.
+        :rtype: Optional[str]
+        """
+
+        # We do not ensure that the volume exists since it might not
+        # physically exist (for example, the volume creation failed), but it
+        # might still have had its "not-exists" property set.
+        volume_properties = self._get_volume_properties(volume_uuid)
+        return volume_properties.get(self.PROP_NOT_EXISTS)
+
     def shallow_clone_volume(self, volume_uuid, clone_uuid, persistent=True):
         """
         Clone a volume. Do not copy the data, this method creates a new volume