diff --git a/nova/scheduler/manager.py b/nova/scheduler/manager.py
index 6472abaa3b45..b6979f88cc4d 100644
--- a/nova/scheduler/manager.py
+++ b/nova/scheduler/manager.py
@@ -130,9 +130,10 @@ class SchedulerManager(manager.Manager):
             raise exception.NoValidHost(reason=e.message)
 
         resources = utils.resources_from_request_spec(spec_obj)
+        is_rebuild = utils.request_is_rebuild(spec_obj)
         alloc_reqs_by_rp_uuid, provider_summaries, allocation_request_version \
             = None, None, None
-        if self.driver.USES_ALLOCATION_CANDIDATES:
+        if self.driver.USES_ALLOCATION_CANDIDATES and not is_rebuild:
             res = self.placement_client.get_allocation_candidates(ctxt,
                                                                   resources)
             if res is None:
diff --git a/nova/tests/functional/test_servers.py b/nova/tests/functional/test_servers.py
index a1aa4303f80e..68e15c39a9bf 100644
--- a/nova/tests/functional/test_servers.py
+++ b/nova/tests/functional/test_servers.py
@@ -1257,6 +1257,15 @@ class ServerRebuildTestCase(integrated_helpers._IntegratedTestBase,
             return self.placement_api.get(
                 '/allocations/%s' % server_uuid).body['allocations']
 
+        def _set_provider_inventory(rp_uuid, resource_class, inventory):
+            # Get the resource provider generation for the inventory update.
+            rp = self.placement_api.get(
+                '/resource_providers/%s' % rp_uuid).body
+            inventory['resource_provider_generation'] = rp['generation']
+            return self.placement_api.put(
+                '/resource_providers/%s/inventories/%s' %
+                (rp_uuid, resource_class), inventory).body
+
         def assertFlavorMatchesAllocation(flavor, allocation):
             self.assertEqual(flavor['vcpus'], allocation['VCPU'])
             self.assertEqual(flavor['ram'], allocation['MEMORY_MB'])
@@ -1285,6 +1294,9 @@ class ServerRebuildTestCase(integrated_helpers._IntegratedTestBase,
 
         flavor = self.api.api_get('/flavors/1').body['flavor']
 
+        # make the compute node full and ensure rebuild still succeed
+        _set_provider_inventory(rp_uuid, "VCPU", {"total": 1})
+
         # There should be usage for the server on the compute node now.
         rp_usages = _get_provider_usages(rp_uuid)
         assertFlavorMatchesAllocation(flavor, rp_usages)
diff --git a/nova/tests/unit/scheduler/test_scheduler.py b/nova/tests/unit/scheduler/test_scheduler.py
index afd83061a633..e5998473d304 100644
--- a/nova/tests/unit/scheduler/test_scheduler.py
+++ b/nova/tests/unit/scheduler/test_scheduler.py
@@ -228,6 +228,22 @@ class SchedulerManagerTestCase(test.NoDBTestCase):
         place_res = ([], {}, None)
         self._test_select_destination(place_res)
 
+    @mock.patch('nova.scheduler.utils.resources_from_request_spec')
+    @mock.patch('nova.scheduler.client.report.SchedulerReportClient.'
+                'get_allocation_candidates')
+    def test_select_destination_is_rebuild(self, mock_get_ac, mock_rfrs):
+        fake_spec = objects.RequestSpec(
+            scheduler_hints={'_nova_check_type': ['rebuild']})
+        fake_spec.instance_uuid = uuids.instance
+        with mock.patch.object(self.manager.driver, 'select_destinations'
+                ) as select_destinations:
+            self.manager.select_destinations(self.context, spec_obj=fake_spec,
+                    instance_uuids=[fake_spec.instance_uuid])
+            select_destinations.assert_called_once_with(
+                self.context, fake_spec,
+                [fake_spec.instance_uuid], None, None, None, False)
+            mock_get_ac.assert_not_called()
+
     @mock.patch('nova.scheduler.utils.resources_from_request_spec')
     @mock.patch('nova.scheduler.client.report.SchedulerReportClient.'
                 'get_allocation_candidates')