Block deleting compute services which are hosting instances

This change makes "DELETE /os-services/{service_id}" fail
with a 409 response when attempting to delete a nova-compute
service which is still hosting instances.

Deleting a compute service also results in deleting the
related compute_nodes table entry for that service host.
The compute node resource provider in placement is tied
to the compute node via the UUID, and if we allow deleting
the compute service and node then the resource provider for
that node is effectively orphaned in Placement, along with
the instances which have allocations against that resource
provider.

Furthermore, restarting the compute service will create a
new service and compute_nodes record, and the compute node
would have a new UUID and resource provider. This will
affect scheduling for that host since Placement will be
reporting it as having available capacity which in reality
is not accurate.

A release note is included for the (justified) behavior
change in the API. A new microversion should not be required
for this since admins should not have to opt out of broken
behavior. Since this API did not previously expect to return
a 409 response, the "expected_errors" decorator is updated
and again, should not require a microversion per the
guidelines:

https://docs.openstack.org/nova/latest/contributor/microversions.html#when-a-microversion-is-not-needed

Conflicts:
      nova/tests/functional/wsgi/test_services.py

NOTE(mriedem): This is due to the rc_fields move from
change Iea182341f9419cb514a044f76864d6bec60a3683 in Rocky.

Change-Id: I0bd63b655ad3d3d39af8d15c781ce0a45efc8e3a
Closes-Bug: #1763183
(cherry picked from commit 42f62f1ed2)

api-ref/source/os-services.inc

@@ -318,6 +318,10 @@ Delete Compute Service
 Deletes a service. If it's a ``nova-compute`` service, then the
 corresponding host will be removed from all the host aggregates as well.
 
+Attempts to delete a ``nova-compute`` service which is still hosting instances
+will result in a 409 HTTPConflict response. The instances will need to be
+migrated or deleted before a compute service can be deleted.
+
 .. important:: Be sure to stop the actual ``nova-compute`` process on the
                physical host *before* deleting the service with this API.
                Failing to do so can lead to the running service re-creating
@@ -325,7 +329,8 @@ corresponding host will be removed from all the host aggregates as well.
 
 Normal response codes: 204
 
-Error response codes: badRequest(400), unauthorized(401), forbidden(403), itemNotFound(404)
+Error response codes: badRequest(400), unauthorized(401), forbidden(403),
+itemNotFound(404), conflict(409)
 
 Request
 -------

nova/api/openstack/compute/services.py

@@ -24,6 +24,7 @@ from nova import availability_zones
 from nova import compute
 from nova import exception
 from nova.i18n import _
+from nova import objects
 from nova.policies import services as services_policies
 from nova import servicegroup
 from nova import utils
@@ -189,7 +190,7 @@ class ServiceController(wsgi.Controller):
         return action(body, context)
 
     @wsgi.response(204)
-    @wsgi.expected_errors((400, 404))
+    @wsgi.expected_errors((400, 404, 409))
     def delete(self, req, id):
         """Deletes the specified service."""
         context = req.environ['nova.context']
@@ -211,6 +212,21 @@ class ServiceController(wsgi.Controller):
             service = self.host_api.service_get_by_id(context, id)
             # remove the service from all the aggregates in which it's included
             if service.binary == 'nova-compute':
+                # Check to see if there are any instances on this compute host
+                # because if there are, we need to block the service (and
+                # related compute_nodes record) delete since it will impact
+                # resource accounting in Placement and orphan the compute node
+                # resource provider.
+                # TODO(mriedem): Use a COUNT SQL query-based function instead
+                # of InstanceList.get_uuids_by_host for performance.
+                instance_uuids = objects.InstanceList.get_uuids_by_host(
+                    context, service['host'])
+                if instance_uuids:
+                    raise webob.exc.HTTPConflict(
+                        explanation=_('Unable to delete compute service that '
+                                      'is hosting instances. Migrate or '
+                                      'delete the instances first.'))
+
                 aggrs = self.aggregate_api.get_aggregates_by_host(context,
                                                                   service.host)
                 for ag in aggrs:

nova/tests/functional/wsgi/test_services.py

@@ -10,9 +10,12 @@
 # License for the specific language governing permissions and limitations
 # under the License.
 
+import six
+
 from nova import context as nova_context
 from nova import objects
 from nova.objects import fields as rc_fields
+from nova.tests.functional.api import client as api_client
 from nova.tests.functional import test_servers
 
 
@@ -73,12 +76,24 @@ class TestServicesAPI(test_servers.ProviderUsageBaseTestCase):
         # update_available_resource periodic task.
         self.admin_api.put_service(service['id'], {'forced_down': True})
         compute.stop()
-        # FIXME(mriedem): This is bug 1763183 where the compute node has
-        # an instance running on it but we allow you to delete the service
-        # and compute node anyway, which will affect the allocations for the
-        # instance and orphans the compute node resource provider in Placement.
-        # Once the bug is fixed, this should fail until the instance is either
-        # migrated or deleted.
+        # The first attempt should fail since there is an instance on the
+        # compute host.
+        ex = self.assertRaises(api_client.OpenStackApiException,
+                               self.admin_api.api_delete,
+                               '/os-services/%s' % service['id'])
+        self.assertIn('Unable to delete compute service that is hosting '
+                      'instances.', six.text_type(ex))
+        self.assertEqual(409, ex.response.status_code)
+
+        # Now delete the instance and wait for it to be gone.
+        # Note that we can't use self._delete_and_check_allocations here
+        # because of bug 1679750 where allocations are not deleted when
+        # an instance is deleted and the compute service it's running on
+        # is down.
+        self.api.delete_server(server['id'])
+        self._wait_until_deleted(server)
+
+        # Now we can delete the service.
         self.admin_api.api_delete('/os-services/%s' % service['id'])
 
         # Make sure the service is deleted.

releasenotes/notes/bug-1763183-service-delete-with-instances-d7c5c47e4ce31239.yaml

@@ -0,0 +1,9 @@
+---
+fixes:
+  - |
+    The ``DELETE /os-services/{service_id}`` compute API will now return a
+    ``409 HTTPConflict`` response when trying to delete a ``nova-compute``
+    service which is still hosting instances. This is because doing so would
+    orphan the compute node resource provider in the placement service on
+    which those instances have resource allocations, which affects scheduling.
+    See https://bugs.launchpad.net/nova/+bug/1763183 for more details.