glusterFS_native_driver: snapshot delete doesn't delete snapshot entries that are in error state
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Shared File Systems Service (Manila) |
Won't Fix
|
Medium
|
Unassigned |
Bug Description
Description:
In openstack-manila running kilo release, if snapshot creation fails for some reason, an entry for snapshot is still created with 'error' state. Attempting to delete this invalid entry of snapshot doesn't succeed.
Possible scenarios which could create a snapshot in error state are when,
- gluster daemon is down in the backend gluster node
- gluster node is down
- gluster brick process is down
- gluster volume is down
- snapshot created by manila is deleted from backend gluster
While it is okay to create a snapshot with error state, attempting to delete or force-delete should clear the entry from database. Administrator has no means to remove these snapshots in error state.
Error msg from the logs:
GlusterfsException: Failed to identify backing GlusterFS object for snapshot 0a6e89a3-
CLI output:
# manila snapshot-list
+------
| ID | Share ID | Status | Name | Share Size |
+------
| fac9835a-
+------
# manila snapshot-delete fac9835a-
# manila snapshot-list
+------
| ID | Share ID | Status | Name | Share Size |
+------
| fac9835a-
+------
# manila snapshot-
# manila snapshot-list
+------
| ID | Share ID | Status | Name | Share Size |
+------
| fac9835a-
+------
# manila snapshot-
# manila snapshot-list
+------
| ID | Share ID | Status | Name | Share Size |
+------
| fac9835a-
+------
Steps to reproduce:
1) Try to create a snapshot when volume is down (one of the possible ways to create a snapshot that would end up in error state)
2) A snapshot in error state is created
3) Attempt to delete or force-delete the snapshot - Snapshot doesn't gets deleted
summary: |
glusterFS_native_driver: snapshot delete doesn't delete snapshot entries - that are not available + that are in error state |
tags: | added: driver glusterfs |
tags: | added: liberty-rc-potential |
Changed in manila: | |
importance: | Undecided → High |
status: | New → Triaged |
Changed in manila: | |
importance: | High → Medium |
We discussed this in the channel and it's actually a behavior of the manager not the driver. We're open to considering changing the behavior of the manager, but not as a bugfix for Liberty -- it would a change for Mitaka.