Comment 1 for bug 1723165

So if I'm following correct, you have 3 volumes when you snapshot the instance, then set volume quota to 5, create another instance from the snapshot image which has 3 image-defined BDMs in it.

Since it's boot from volume, nova-compute is what creates the 3 volumes from volume snapshots. 2 of those volumes would get created and the 3rd would be rejected by cinder because of over quota. That makes the _prep_block_devices method in the compute service fail and set the instance to ERROR state. I'm not sure if 2 of those 3 volumes would be attached or not, and I guess that's the bug. Maybe they aren't in a state that compute knows how to properly clean them up. It probably also matters what the delete_on_termination value is in those bdms to see if compute would remove them.