Comment 5 for bug 1308405

So I think the real problem here is that the device_id was never unassigned before the instance rescheduled. I suspect this review that landed 5 days ago happens to fix this bug also:

https://review.openstack.org/#/c/99400/