diff --git a/tempest/api/compute/servers/test_device_tagging.py b/tempest/api/compute/servers/test_device_tagging.py
index 58d4d7d..56456f4 100644
--- a/tempest/api/compute/servers/test_device_tagging.py
+++ b/tempest/api/compute/servers/test_device_tagging.py
@@ -35,6 +35,8 @@
 
 class DeviceTaggingBase(base.BaseV2ComputeTest):
 
+    credentials = ['primary', 'admin']
+
     @classmethod
     def skip_checks(cls):
         super(DeviceTaggingBase, cls).skip_checks()
@@ -54,6 +56,7 @@
         cls.ports_client = cls.os_primary.ports_client
         cls.subnets_client = cls.os_primary.subnets_client
         cls.interfaces_client = cls.os_primary.interfaces_client
+        cls.servers_admin_client = cls.os_admin.servers_client
 
     @classmethod
     def setup_credentials(cls):
@@ -422,11 +425,13 @@
         self.servers_client.detach_volume(server['id'], volume['id'])
         waiters.wait_for_volume_resource_status(self.volumes_client,
                                                 volume['id'], 'available')
-        self.interfaces_client.delete_interface(server['id'],
-                                                interface['port_id'])
-        waiters.wait_for_interface_detach(self.interfaces_client,
+        req_id = self.interfaces_client.delete_interface(
+            server['id'], interface['port_id']
+        ).response['x-openstack-request-id']
+        waiters.wait_for_interface_detach(self.servers_admin_client,
                                           server['id'],
-                                          interface['port_id'])
+                                          interface['port_id'],
+                                          req_id)
         # FIXME(mriedem): The assertion that the tagged devices are removed
         # from the metadata for the server is being skipped until bug 1775947
         # is fixed.
diff --git a/tempest/common/waiters.py b/tempest/common/waiters.py
index f6a4555..33ed153 100644
--- a/tempest/common/waiters.py
+++ b/tempest/common/waiters.py
@@ -489,18 +489,34 @@
     return body
 
 
-def wait_for_interface_detach(client, server_id, port_id):
+def wait_for_interface_detach(client, server_id, port_id, detach_request_id):
     """Waits for an interface to be detached from a server."""
-    body = client.list_interfaces(server_id)['interfaceAttachments']
-    ports = [iface['port_id'] for iface in body]
+    def _get_detach_event_results():
+        # NOTE(gibi): The obvious choice for this waiter would be to wait
+        # until the interface disappears from the client.list_interfaces()
+        # response. However that response is based on the binding status of the
+        # port in Neutron. Nova deallocates the port resources _after the port
+        # was  unbound in Neutron. This can cause that the naive waiter would
+        # return before the port is fully deallocated. Wait instead of the
+        # os-instance-action to succeed as that is recorded after both the
+        # port is fully deallocated.
+        events = client.show_instance_action(
+            server_id, detach_request_id)['instanceAction'].get('events', [])
+        return [
+            event['result'] for event in events
+            if event['event'] == 'compute_detach_interface'
+        ]
+
+    detach_event_results = _get_detach_event_results()
+
     start = int(time.time())
 
-    while port_id in ports:
+    while "Success" not in detach_event_results:
         time.sleep(client.build_interval)
-        body = client.list_interfaces(server_id)['interfaceAttachments']
-        ports = [iface['port_id'] for iface in body]
-        if port_id not in ports:
-            return body
+        detach_event_results = _get_detach_event_results()
+        if "Success" in detach_event_results:
+            return client.show_instance_action(
+                server_id, detach_request_id)['instanceAction']
 
         timed_out = int(time.time()) - start >= client.build_timeout
         if timed_out:
diff --git a/tempest/tests/common/test_waiters.py b/tempest/tests/common/test_waiters.py
index 5cdbfbf..a5016e2 100755
--- a/tempest/tests/common/test_waiters.py
+++ b/tempest/tests/common/test_waiters.py
@@ -186,37 +186,94 @@
                                          mock.call('server_id', 'port_id')])
         sleep.assert_called_once_with(client.build_interval)
 
-    one_interface = {'interfaceAttachments': [{'port_id': 'port_one'}]}
-    two_interfaces = {'interfaceAttachments': [{'port_id': 'port_one'},
-                                               {'port_id': 'port_two'}]}
-
     def test_wait_for_interface_detach(self):
-        list_interfaces = mock.MagicMock(
-            side_effect=[self.two_interfaces, self.one_interface])
-        client = self.mock_client(list_interfaces=list_interfaces)
+        no_event = {
+            'instanceAction': {
+                'events': []
+            }
+        }
+        one_event_without_result = {
+            'instanceAction': {
+                'events': [
+                    {
+                        'event': 'compute_detach_interface',
+                        'result': None
+                    }
+
+                ]
+            }
+        }
+        one_event_successful = {
+            'instanceAction': {
+                'events': [
+                    {
+                        'event': 'compute_detach_interface',
+                        'result': 'Success'
+                    }
+                ]
+            }
+        }
+
+        show_instance_action = mock.MagicMock(
+            # there is an extra call to return the result from the waiter
+            side_effect=[
+                no_event,
+                one_event_without_result,
+                one_event_successful,
+                one_event_successful,
+            ]
+        )
+        client = self.mock_client(show_instance_action=show_instance_action)
         self.patch('time.time', return_value=0.)
         sleep = self.patch('time.sleep')
 
         result = waiters.wait_for_interface_detach(
-            client, 'server_id', 'port_two')
+            client, mock.sentinel.server_id, mock.sentinel.port_id,
+            mock.sentinel.detach_request_id
+        )
 
-        self.assertIs(self.one_interface['interfaceAttachments'], result)
-        list_interfaces.assert_has_calls([mock.call('server_id'),
-                                          mock.call('server_id')])
-        sleep.assert_called_once_with(client.build_interval)
+        self.assertIs(one_event_successful['instanceAction'], result)
+        show_instance_action.assert_has_calls(
+            # there is an extra call to return the result from the waiter
+            [
+                mock.call(
+                    mock.sentinel.server_id, mock.sentinel.detach_request_id)
+            ] * 4
+        )
+        sleep.assert_has_calls([mock.call(client.build_interval)] * 2)
 
     def test_wait_for_interface_detach_timeout(self):
-        list_interfaces = mock.MagicMock(return_value=self.one_interface)
-        client = self.mock_client(list_interfaces=list_interfaces)
+        one_event_without_result = {
+            'instanceAction': {
+                'events': [
+                    {
+                        'event': 'compute_detach_interface',
+                        'result': None
+                    }
+
+                ]
+            }
+        }
+
+        show_instance_action = mock.MagicMock(
+            return_value=one_event_without_result)
+        client = self.mock_client(show_instance_action=show_instance_action)
         self.patch('time.time', side_effect=[0., client.build_timeout + 1.])
         sleep = self.patch('time.sleep')
 
-        self.assertRaises(lib_exc.TimeoutException,
-                          waiters.wait_for_interface_detach,
-                          client, 'server_id', 'port_one')
+        self.assertRaises(
+            lib_exc.TimeoutException,
+            waiters.wait_for_interface_detach,
+            client, mock.sentinel.server_id, mock.sentinel.port_id,
+            mock.sentinel.detach_request_id
+        )
 
-        list_interfaces.assert_has_calls([mock.call('server_id'),
-                                          mock.call('server_id')])
+        show_instance_action.assert_has_calls(
+            [
+                mock.call(
+                    mock.sentinel.server_id, mock.sentinel.detach_request_id)
+            ] * 2
+        )
         sleep.assert_called_once_with(client.build_interval)
 
     def test_wait_for_guest_os_boot(self):