Fix idrac-wsman having Completed with Errors jobs

iDRAC jobs can finish in 'Completed', 'Failed' and also
'Completed with Errors' state. This fix adds handling of
'Completed with Errors' as finished failed job otherwise node
stays in wait state as it does not consider such jobs
as finished.

Change-Id: I5018bf8ef6c86c6d303258f1497fa83d33b3cb76
(cherry picked from commit 6e0c0e7fd025dbfdc4f2e6070ad5cdb3079c2cef)

5 files changed, 73 insertions, 2 deletions

diff --git a/ironic/drivers/modules/drac/bios.py b/ironic/drivers/modules/drac/bios.py
index 2ec6660a9..75d487bdf 100644
--- a/ironic/drivers/modules/drac/bios.py
+++ b/ironic/drivers/modules/drac/bios.py
@@ -324,7 +324,8 @@ class DracWSManBIOS(base.BIOSInterface):
 
             if config_job is None or config_job.status == 'Completed':
                 finished_job_ids.append(config_job_id)
-            elif config_job.status == 'Failed':
+            elif (config_job.status == 'Failed'
+                    or config_job.status == 'Completed with Errors'):
                 finished_job_ids.append(config_job_id)
                 job_failed = True
 
diff --git a/ironic/drivers/modules/drac/raid.py b/ironic/drivers/modules/drac/raid.py
index 4bb41c1f6..0be3a29fe 100644
--- a/ironic/drivers/modules/drac/raid.py
+++ b/ironic/drivers/modules/drac/raid.py
@@ -1521,7 +1521,8 @@ class DracWSManRAID(base.RAIDInterface):
 
             if config_job is None or config_job.status == 'Completed':
                 finished_job_ids.append(config_job_id)
-            elif config_job.status == 'Failed':
+            elif (config_job.status == 'Failed'
+                    or config_job.status == 'Completed with Errors'):
                 finished_job_ids.append(config_job_id)
                 self._set_raid_config_job_failure(node)
 
diff --git a/ironic/tests/unit/drivers/modules/drac/test_bios.py b/ironic/tests/unit/drivers/modules/drac/test_bios.py
index 8aae25be2..5e160caed 100644
--- a/ironic/tests/unit/drivers/modules/drac/test_bios.py
+++ b/ironic/tests/unit/drivers/modules/drac/test_bios.py
@@ -355,6 +355,34 @@ class DracWSManBIOSConfigurationTestCase(test_utils.BaseDracTest):
             mock_cleaning_error_handler.assert_called_once_with(
                 task, mock.ANY, "Failed config job: 123. Message: 'Invalid'.")
 
+    @mock.patch.object(manager_utils, 'cleaning_error_handler', autospec=True)
+    @mock.patch.object(drac_job, 'get_job', spec_set=True,
+                       autospec=True)
+    def test__check_node_bios_jobs_completed_with_errors(
+            self, mock_get_job, mock_cleaning_error_handler):
+        mock_job = mock.Mock()
+        mock_job.status = 'Completed with Errors'
+        mock_job.id = '123'
+        mock_job.message = 'PR31: Completed with Errors'
+        mock_get_job.return_value = mock_job
+
+        with task_manager.acquire(self.context, self.node.uuid) as task:
+            driver_internal_info = task.node.driver_internal_info
+            driver_internal_info['bios_config_job_ids'] = ['123']
+            task.node.driver_internal_info = driver_internal_info
+            task.node.clean_step = {'priority': 100, 'interface': 'bios',
+                                    'step': 'factory_reset', 'argsinfo': {}}
+            task.node.save()
+
+            task.driver.bios._check_node_bios_jobs(task)
+
+            self.assertEqual([],
+                             task.node.driver_internal_info.get(
+                                 'bios_config_job_ids'))
+            mock_cleaning_error_handler.assert_called_once_with(
+                task, mock.ANY, "Failed config job: 123. Message: "
+                                "'PR31: Completed with Errors'.")
+
     def test__check_last_system_inventory_changed_different_inventory_time(
             self):
         with task_manager.acquire(self.context, self.node.uuid,
diff --git a/ironic/tests/unit/drivers/modules/drac/test_periodic_task.py b/ironic/tests/unit/drivers/modules/drac/test_periodic_task.py
index ec7f4e27c..6ba590fcd 100644
--- a/ironic/tests/unit/drivers/modules/drac/test_periodic_task.py
+++ b/ironic/tests/unit/drivers/modules/drac/test_periodic_task.py
@@ -236,6 +236,39 @@ class DracPeriodicTaskTestCase(db_base.DbTestCase):
         self.assertEqual({}, self.node.raid_config)
         mock_cleaning_error_handler.assert_called_once_with(task, mock.ANY)
 
+    @mock.patch.object(manager_utils, 'cleaning_error_handler', autospec=True)
+    @mock.patch.object(drac_common, 'get_drac_client', spec_set=True,
+                       autospec=True)
+    def test__check_node_raid_jobs_with_completed_with_errors_job(
+            self, mock_get_drac_client, mock_cleaning_error_handler):
+        # mock node.driver_internal_info and node.clean_step
+        driver_internal_info = {'raid_config_job_ids': ['42']}
+        self.node.driver_internal_info = driver_internal_info
+        self.node.clean_step = {'foo': 'bar'}
+        self.node.save()
+        # mock task
+        task = mock.Mock(node=self.node, context=self.context)
+        # mock dracclient.get_job
+        self.job['status'] = 'Completed with Errors'
+        self.job['message'] = 'PR31: Completed with Errors'
+        mock_client = mock.Mock()
+        mock_get_drac_client.return_value = mock_client
+        mock_client.get_job.return_value = test_utils.dict_to_namedtuple(
+            values=self.job)
+        # mock dracclient.list_virtual_disks
+        mock_client.list_virtual_disks.return_value = [
+            test_utils.dict_to_namedtuple(values=self.virtual_disk)]
+
+        self.raid._check_node_raid_jobs(task)
+
+        mock_client.get_job.assert_called_once_with('42')
+        self.assertEqual(0, mock_client.list_virtual_disks.call_count)
+        self.node.refresh()
+        self.assertEqual([],
+                         self.node.driver_internal_info['raid_config_job_ids'])
+        self.assertEqual({}, self.node.raid_config)
+        mock_cleaning_error_handler.assert_called_once_with(task, mock.ANY)
+
     @mock.patch.object(manager_utils, 'deploying_error_handler', autospec=True)
     @mock.patch.object(manager_utils, 'cleaning_error_handler', autospec=True)
     @mock.patch.object(drac_common, 'get_drac_client', spec_set=True,
diff --git a/releasenotes/notes/idrac-wsman-completed-with-errors-f65c9a48ed4c02d4.yaml b/releasenotes/notes/idrac-wsman-completed-with-errors-f65c9a48ed4c02d4.yaml
new file mode 100644
index 000000000..1e6747327
--- /dev/null
+++ b/releasenotes/notes/idrac-wsman-completed-with-errors-f65c9a48ed4c02d4.yaml
@@ -0,0 +1,8 @@
+---
+fixes:
+  - |
+    Fixes ``idrac-wsman`` BIOS and RAID interface steps to correctly check
+    status of iDRAC job that completed with errors. Now these jobs are treated
+    as failures. Before this fix node stayed in wait state as it was only
+    checking for "Completed" or "Failed" job status, but not "Completed
+    with Errors".