Comment 5 for bug 1719915

Nova destroys the instance object in the compute/manager [1] _before_ removes the allocation [2]. This is the root cause of the race condition at instance delete. Fortunately we have the instance.delete notification emitted[3] _after_ the allocation is deleted so we can modify the _wait_until_deleted to wait for the deleted notification as well.

[1] https://github.com/openstack/nova/blob/faede889d3620f8ff0131a7a4c6b9c1bc844cd06/nova/compute/manager.py#L2425
[2] https://github.com/openstack/nova/blob/faede889d3620f8ff0131a7a4c6b9c1bc844cd06/nova/compute/manager.py#L734
[3]https://github.com/openstack/nova/blob/faede889d3620f8ff0131a7a4c6b9c1bc844cd06/nova/compute/manager.py#L738