Deploy of VM occasionally fails with OSError

Deploy of VM occasionally fails due to file system related error
during config drive creation for the VM. Hence, adding a retry
logic for vopt create as part of deploy. In case, if there's
an OSError exception we'll attempt a retry of config drive
image creation.

Change-Id: I89b15bcd8c974c5be12e30ca220526647d45e879
Closes-Bug: #1664176

nova_powervm/tests/virt/powervm/test_media.py

@@ -1,4 +1,4 @@
-# Copyright 2015 IBM Corp.
+# Copyright 2015, 2017 IBM Corp.
 #
 # All Rights Reserved.
 #
@@ -63,6 +63,18 @@ class TestConfigDrivePowerVM(test.TestCase):
         self.assertEqual('/tmp/cfgdrv/cfg_fake_instance_with_name_that_.iso',
                          iso_path)
         self.assertTrue(self.validate_vopt.called)
+        # Test retry vopt create
+        mock_mkdrv.reset_mock()
+        mock_mkdrv.side_effect = [OSError, mock_mkdrv]
+        mock_instance.name = 'fake-instance-2'
+        iso_path, file_name = cfg_dr_builder._create_cfg_dr_iso(mock_instance,
+                                                                mock_files,
+                                                                mock_net)
+        self.assertEqual('cfg_fake_instance_2.iso', file_name)
+        self.assertEqual('/tmp/cfgdrv/cfg_fake_instance_2.iso',
+                         iso_path)
+        self.assertTrue(self.validate_vopt.called)
+        self.assertEqual(mock_mkdrv.call_count, 2)
 
     @mock.patch('nova_powervm.virt.powervm.media.ConfigDrivePowerVM.'
                 '_create_cfg_dr_iso')

nova_powervm/virt/powervm/media.py

@@ -1,4 +1,4 @@
-# Copyright 2015, 2016 IBM Corp.
+# Copyright 2015, 2017 IBM Corp.
 #
 # All Rights Reserved.
 #
@@ -19,6 +19,7 @@ from nova.api.metadata import base as instance_metadata
 from nova.network import model as network_model
 from nova.virt import configdrive
 import os
+import retrying
 from taskflow import task
 
 from oslo_log import log as logging
@@ -122,8 +123,26 @@ class ConfigDrivePowerVM(object):
                          "building to path %(iso_path)s."),
                      {'inst': instance.name, 'iso_path': iso_path},
                      instance=instance)
-            cdb.make_drive(iso_path)
+            # In case, if there's an OSError related failure while
-            return iso_path, file_name
+            # creating config drive, retry make drive operation.
+
+            def _retry_on_oserror(exc):
+                return isinstance(exc, OSError)
+
+            @retrying.retry(retry_on_exception=_retry_on_oserror,
+                            stop_max_attempt_number=2)
+            def _make_cfg_drive(iso_path):
+                cdb.make_drive(iso_path)
+            try:
+                _make_cfg_drive(iso_path)
+                return iso_path, file_name
+            except OSError:
+                # If we get here, that means there's an exception during
+                # second attempt, log the same and fail the deploy
+                # operation.
+                LOG.exception("Config drive ISO could not be built",
+                              instance=instance)
+                raise
 
     def create_cfg_drv_vopt(self, instance, injected_files, network_info,
                             lpar_uuid, admin_pass=None, mgmt_cna=None,