Implement force_detach to allow safe cleanup of stuck volumes
For volumes stuck in 'attaching' or 'detaching' , there is no safe way to cleanup that involves the backend storage. Using python-cinderclient 'reset-state' will only change the Cinder database, and may leave the volume exported to the compute host, and may leave an entry in Nova's database that prevents the volume from being re-used. This blueprint+spec outlines the changes to the cinder side to allow a force_detach. In addtion, python-novaclient will need to have a '--force' option added to 'nova volume-detach' so that Nova BlockDeviceMapping table is cleaned up after calling cinder force_detach().
Blueprint information
- Status:
- Complete
- Approver:
- Mike Perez
- Priority:
- Medium
- Drafter:
- Scott DAngelo
- Direction:
- Approved
- Assignee:
- Scott DAngelo
- Definition:
- Superseded
- Series goal:
- Proposed for liberty
- Implementation:
-
Started
- Milestone target:
-
next
- Started by
- Scott DAngelo
- Completed by
- Sean McGinnis
Related branches
Related bugs
Sprints
Whiteboard
Hi Scott,
I'm very pleasure to hear about that, and the Nova spec has been restored for L.
Thanks for your effort to this feature.:)
Hi Wanghao,
I've spoken with Nova folk at the Vancouver Summit. They are in favor of this change, which requires a Nova spec. I'd like to un-abandon your spec from Juno and add some use cases, etc.
Let's get together and sync up on this sometime (perhaps after the Summit is over).
Scott
hi Scott,
This feature looks like very similar with this bp:https:/
So I wonder to know is the implement of this bp in Nova side? If so, maybe we need to discuss this with nova folks too.
This fix does not need any changes in Cinder other than:
https:/
