Comment 5 for bug 1311182

What you are seeing here is the result of:

1. Snapshot delete operation started
2. Cinder times out the snapshot delete operation and fails it because it took longer than expected. (This is not ideal but is how things currently work.)
3. Nova continues processing the snapshot delete operation and commits the changes. This means the snapshot actually is deleted from the storage/qcow2 point of view.
4. Cinder's view is now out of sync because it didn't update the info file -- what should have happened at this point is the snapshot entry is removed from the .info file and the snapshot is deleted from Cinder.
5. Cinder tries to delete the snapshot on this run but it no longer exists so the delete operation fails.

I have some changes in the works to address (2) above but another approach here would be to have the delete snapshot operation handle this case cleanly. (It probably should.)