Merge "libvirt: handle DiskNotFound during update_available_resource" into stable/pike

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

@@ -13449,6 +13449,50 @@ class LibvirtConnTestCase(test.NoDBTestCase,
         drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False)
         self.assertEqual(0, drvr._get_disk_over_committed_size_total())
 
+    @mock.patch('nova.virt.libvirt.host.Host.list_instance_domains')
+    @mock.patch('nova.virt.libvirt.driver.LibvirtDriver.'
+                '_get_instance_disk_info_from_config',
+                side_effect=exception.DiskNotFound(location='/opt/stack/foo'))
+    @mock.patch('nova.objects.BlockDeviceMappingList.bdms_by_instance_uuid',
+                return_value=objects.BlockDeviceMappingList())
+    @mock.patch('nova.objects.InstanceList.get_by_filters',
+                return_value=objects.InstanceList(objects=[
+                    objects.Instance(uuid=uuids.instance,
+                                     task_state=task_states.DELETING)]))
+    def test_disk_over_committed_size_total_disk_not_found_ignore(
+            self, mock_get, mock_bdms, mock_get_disk_info, mock_list_domains):
+        """Tests that we handle DiskNotFound gracefully for an instance that
+        is undergoing a task_state transition.
+        """
+        mock_dom = mock.Mock()
+        mock_dom.XMLDesc.return_value = "<domain/>"
+        mock_dom.UUIDString.return_value = uuids.instance
+        mock_list_domains.return_value = [mock_dom]
+        drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False)
+        self.assertEqual(0, drvr._get_disk_over_committed_size_total())
+
+    @mock.patch('nova.virt.libvirt.host.Host.list_instance_domains')
+    @mock.patch('nova.virt.libvirt.driver.LibvirtDriver.'
+                '_get_instance_disk_info_from_config',
+                side_effect=exception.DiskNotFound(location='/opt/stack/foo'))
+    @mock.patch('nova.objects.BlockDeviceMappingList.bdms_by_instance_uuid',
+                return_value=objects.BlockDeviceMappingList())
+    @mock.patch('nova.objects.InstanceList.get_by_filters',
+                return_value=objects.InstanceList(objects=[
+                    objects.Instance(uuid=uuids.instance, task_state=None)]))
+    def test_disk_over_committed_size_total_disk_not_found_reraise(
+            self, mock_get, mock_bdms, mock_get_disk_info, mock_list_domains):
+        """Tests that we handle DiskNotFound gracefully for an instance that
+        is NOT undergoing a task_state transition and the error is re-raised.
+        """
+        mock_dom = mock.Mock()
+        mock_dom.XMLDesc.return_value = "<domain/>"
+        mock_dom.UUIDString.return_value = uuids.instance
+        mock_list_domains.return_value = [mock_dom]
+        drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False)
+        self.assertRaises(exception.DiskNotFound,
+                          drvr._get_disk_over_committed_size_total)
+
     def test_cpu_info(self):
         drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), True)
 

nova/tests/unit/virt/test_images.py

@@ -32,7 +32,7 @@ class QemuTestCase(test.NoDBTestCase):
 
     @mock.patch.object(os.path, 'exists', return_value=True)
     def test_qemu_info_with_errors(self, path_exists):
-        self.assertRaises(exception.InvalidDiskInfo,
+        self.assertRaises(exception.DiskNotFound,
                           images.qemu_img_info,
                           '/fake/path')
 
@@ -65,6 +65,24 @@ class QemuTestCase(test.NoDBTestCase):
                                 '/fake/path')
         self.assertIn('qemu-img aborted by prlimits', six.text_type(exc))
 
+    @mock.patch.object(utils, 'execute')
+    @mock.patch.object(os.path, 'exists', return_value=True)
+    def test_qemu_img_info_with_disk_not_found(self, exists, mocked_execute):
+        """Tests that the initial os.path.exists check passes but the qemu-img
+        command fails because the path is gone by the time the command runs.
+        """
+        path = '/opt/stack/data/nova/instances/some-uuid/disk'
+        stderr = (u"qemu-img: Could not open "
+                  "'/opt/stack/data/nova/instances/some-uuid/disk': "
+                  "Could not open '/opt/stack/data/nova/instances/some-uuid/"
+                  "disk': No such file or directory\n")
+        mocked_execute.side_effect = (
+            processutils.ProcessExecutionError(
+                exit_code=1, stderr=stderr))
+        self.assertRaises(exception.DiskNotFound, images.qemu_img_info, path)
+        exists.assert_called_once_with(path)
+        mocked_execute.assert_called_once()
+
     @mock.patch.object(images, 'convert_image',
                        side_effect=exception.ImageUnacceptable)
     @mock.patch.object(images, 'qemu_img_info')

nova/virt/images.py

@@ -72,10 +72,15 @@ def qemu_img_info(path, format=None):
             cmd = cmd + ('--force-share',)
         out, err = utils.execute(*cmd, prlimit=QEMU_IMG_LIMITS)
     except processutils.ProcessExecutionError as exp:
-        # this means we hit prlimits, make the exception more specific
         if exp.exit_code == -9:
+            # this means we hit prlimits, make the exception more specific
             msg = (_("qemu-img aborted by prlimits when inspecting "
                     "%(path)s : %(exp)s") % {'path': path, 'exp': exp})
+        elif exp.exit_code == 1 and 'No such file or directory' in exp.stderr:
+            # The os.path.exists check above can race so this is a simple
+            # best effort at catching that type of failure and raising a more
+            # specific error.
+            raise exception.DiskNotFound(location=path)
         else:
             msg = (_("qemu-img failed to execute on %(path)s : %(exp)s") %
                    {'path': path, 'exp': exp})

nova/virt/libvirt/driver.py

@@ -7511,6 +7511,24 @@ class LibvirtDriver(driver.ComputeDriver):
                             'by concurrent operations such as resize. '
                             'Error: %(error)s',
                             {'i_name': guest.name, 'error': e})
+            except exception.DiskNotFound:
+                with excutils.save_and_reraise_exception() as err_ctxt:
+                    # If the instance is undergoing a task state transition,
+                    # like moving to another host or is being deleted, we
+                    # should ignore this instance and move on.
+                    if guest.uuid in local_instances:
+                        inst = local_instances[guest.uuid]
+                        if inst.task_state is not None:
+                            LOG.info('Periodic task is updating the host '
+                                     'stats; it is trying to get disk info '
+                                     'for %(i_name)s, but the backing disk '
+                                     'was removed by a concurrent operation '
+                                     '(task_state=%(task_state)s)',
+                                     {'i_name': guest.name,
+                                      'task_state': inst.task_state},
+                                     instance=inst)
+                            err_ctxt.reraise = False
+
             # NOTE(gtt116): give other tasks a chance.
             greenthread.sleep(0)
         return disk_over_committed_size