[ironic] Don't remove instance info twice in destroy

During teardown, at the end of the ironic virt driver's destroy method,
we call _cleanup_deploy, which since
https://review.opendev.org/#/c/563722/ removes instance_info from the
ironic node. Given that we're destroying the node, the instance_info
will have been cleared from the node anyway, so we don't need to do
this.

Further, it can cause tear down to fail if automated cleaning is
enabled in ironic. This happens because ironic prevents updates to nodes
that are undergoing a state transition, as is the case during cleaning.
The virt driver will retry this request by default every 2 seconds with
60 attempts. Typically this is not long enough for cleaning to complete,
so tear down fails with the following error:

  Conflict: Node a00696d5-32ba-475e-9528-59bf11cffea6 can not be updated
  while a state transition is in progress. (HTTP 409)

This change skips the instance info update in _cleanup_deploy in the
case of tear down.

Change-Id: Iea337f73c231db2cb9d9f639b92475daaede6793
Closes-Bug: #1815799

Author: markgoddard

nova/tests/unit/virt/ironic/test_driver.py

@@ -1666,7 +1666,8 @@ def fake_set_provision_state(*_):
 
         mock_node.get_by_instance_uuid.assert_called_with(
             instance.uuid, fields=ironic_driver._NODE_FIELDS)
-        mock_cleanup_deploy.assert_called_with(node, instance, network_info)
+        mock_cleanup_deploy.assert_called_with(node, instance, network_info,
+                                               remove_instance_info=False)
 
         # For states that makes sense check if set_provision_state has
         # been called
@@ -1699,7 +1700,8 @@ def test_destroy_trigger_undeploy_fail(self, mock_clean, fake_validate,
         mock_sps.side_effect = exception.NovaException()
         self.assertRaises(exception.NovaException, self.driver.destroy,
                           self.ctx, instance, None, None)
-        mock_clean.assert_called_once_with(node, instance, None)
+        mock_clean.assert_called_once_with(node, instance, None,
+                                           remove_instance_info=False)
 
     @mock.patch.object(FAKE_CLIENT.node, 'update')
     @mock.patch.object(ironic_driver.IronicDriver,
@@ -1718,7 +1720,8 @@ def test_destroy_trigger_remove_info_fail(self, mock_clean, fake_validate,
         mock_update.side_effect = SystemError('unexpected error')
         self.assertRaises(SystemError, self.driver.destroy,
                           self.ctx, instance, None, None)
-        mock_clean.assert_called_once_with(node, instance, None)
+        mock_clean.assert_called_once_with(node, instance, None,
+                                           remove_instance_info=False)
 
     @mock.patch.object(FAKE_CLIENT.node, 'set_provision_state')
     @mock.patch.object(ironic_driver.IronicDriver,
@@ -2549,6 +2552,24 @@ def test__cleanup_deploy(self, mock_call, mock_vol, mock_unvif,
         mock_call.has_calls(
             [mock.call('node.update', node.uuid, expected_patch)])
 
+    @mock.patch.object(ironic_driver.IronicDriver, '_stop_firewall')
+    @mock.patch.object(ironic_driver.IronicDriver, '_unplug_vifs')
+    @mock.patch.object(ironic_driver.IronicDriver,
+                       '_cleanup_volume_target_info')
+    @mock.patch.object(cw.IronicClientWrapper, 'call')
+    def test__cleanup_deploy_no_remove_ii(self, mock_call, mock_vol,
+                                          mock_unvif, mock_stop_fw):
+        # TODO(TheJulia): This REALLY should be updated to cover all of the
+        # calls that take place.
+        node = ironic_utils.get_test_node(driver='fake')
+        instance = fake_instance.fake_instance_obj(self.ctx,
+                                                   node=node.uuid)
+        self.driver._cleanup_deploy(node, instance, remove_instance_info=False)
+        mock_vol.assert_called_once_with(instance)
+        mock_unvif.assert_called_once_with(node, instance, None)
+        mock_stop_fw.assert_called_once_with(instance, None)
+        self.assertFalse(mock_call.called)
+
 
 class IronicDriverSyncTestCase(IronicDriverTestCase):
 

nova/virt/ironic/driver.py

@@ -486,11 +486,13 @@ def _cleanup_volume_target_info(self, instance):
                              'reason': e},
                             instance=instance)
 
-    def _cleanup_deploy(self, node, instance, network_info=None):
+    def _cleanup_deploy(self, node, instance, network_info=None,
+                        remove_instance_info=True):
         self._cleanup_volume_target_info(instance)
         self._unplug_vifs(node, instance, network_info)
         self._stop_firewall(instance, network_info)
-        self._remove_instance_info_from_node(node, instance)
+        if remove_instance_info:
+            self._remove_instance_info_from_node(node, instance)
 
     def _wait_for_active(self, instance):
         """Wait for the node to be marked as ACTIVE in Ironic."""
@@ -1336,7 +1338,12 @@ def destroy(self, context, instance, network_info,
                 #                removed from ironic node.
                 self._remove_instance_info_from_node(node, instance)
         finally:
-            self._cleanup_deploy(node, instance, network_info)
+            # NOTE(mgoddard): We don't need to remove instance info at this
+            # point since we will have already done it. The destroy will only
+            # succeed if this method returns without error, so we will end up
+            # removing the instance info eventually.
+            self._cleanup_deploy(node, instance, network_info,
+                                 remove_instance_info=False)
 
         LOG.info('Successfully unprovisioned Ironic node %s',
                  node.uuid, instance=instance)