Description of problem:

When instances on a compute node are evacuated, allocations about these evacuated instances are still left on that source node.
These allocations are deleted when the nova-compute service on that node is again started.

However if the compute node is completely broken, it is not possible to start nova-compute service again.
In this situation a user is likely to remove the compute node.

Because allocation records are left on the node, "openstack compute service delete" removes only the service record and leave the resource provider record[1].
https://github.com/openstack/nova/blob/1aa571103f90228ddf3dc27386486196ad58ba0e/nova/scheduler/client/report.py#L2152-L2156

If a user tries to delete the remaining resource provider record, the delete request is rejected because allocations are still left.
The resource provider record left prevents adding back nova-compute service when a user tries to readd a new compute node with the same hostname.

Version-Release number of selected component (if applicable):

How reproducible:
Always

Steps to Reproduce:
1. Deploy overcloud with multiple compute nodes and create instances on one of them
2. Shutoff one of the compute nodes
3. Evacuate instances from the compute node
4. Delete nova-compute service of the compute node
5. Delete resource provider record of the compute node

Actual results:
the resource provider record is left after step 4, and can't be deleted in step 5.

Expected results:
the resource provider record is deleted by step 4 or can be deleted in step 5.

Additional info: