Comment 8 for bug 1714248

Oh wait, its totally the compute node uuid. RP uuid=computeNode and name=ironic node uuid

Nova creates a new compute node uuid for the same ironic node uuid, because its on a different nova-compute node. That works fine see:
https://github.com/openstack/nova/blob/master/nova/db/sqlalchemy/models.py#L117

The ComputeNode unique constraint simply checks ('host', 'hypervisor_hostname', 'deleted') is unique, so we totally allow the ironic node uuid to be registered twice in the ComputeNode's table.

If you restarted the dead n-cpu node (yet, I know, that is stupid), it would delete the orphan compute nodes, but in this case it is dead, so that left in the DB.

But on placement side, we have the (new-uuid, existing-node-uuid) pair, and it fails the unique constraint on the name:
https://github.com/openstack/nova/blob/master/nova/db/sqlalchemy/api_models.py#L295

... So someone needs to delete the old resource provider, I guess the logic that creates the new one could check to see if the name is used by some other node, and delete it as required, and re-assign all the allocations to the new node, etc... Yuck.