Merge "Handle an older agent with agent_token" into stable/ussuri

ironic/drivers/modules/agent_client.py

@@ -140,8 +140,28 @@ class AgentClient(object):
                    'res': result.get('command_result'),
                    'error': error,
                    'code': response.status_code})
-
         if response.status_code >= http_client.BAD_REQUEST:
+            faultstring = result.get('faultstring')
+            if 'agent_token' in faultstring and agent_token:
+                # NOTE(TheJulia) We have an agent that is out of date.
+                # which means I guess grenade updates the agent image
+                # for upgrades... :(
+                if not CONF.require_agent_token:
+                    LOG.warning('Agent command %(method)s for node %(node)s '
+                                'failed. Expected 2xx HTTP status code, got '
+                                '%(code)d. Error suggests an older ramdisk '
+                                'which does not support ``agent_token``. '
+                                'Removing the token for the next retry.',
+                                {'method': method, 'node': node.uuid,
+                                 'code': response.status_code})
+                    i_info = node.driver_internal_info
+                    i_info.pop('agent_secret_token')
+                    node.driver_internal_info = i_info
+                    node.save()
+                    msg = ('Node {} does not appear to support '
+                           'agent_token and it is not required. Next retry '
+                           'will be without the token.').format(node.uuid)
+                    raise exception.AgentConnectionFailed(reason=msg)
             LOG.error('Agent command %(method)s for node %(node)s failed. '
                       'Expected 2xx HTTP status code, got %(code)d.',
                       {'method': method, 'node': node.uuid,

ironic/tests/unit/drivers/modules/test_agent_client.py

@@ -55,6 +55,9 @@ class MockNode(object):
             'instance_info': self.instance_info
         }
 
+    def save(self):
+        pass
+
 
 class TestAgentClient(base.TestCase):
     def setUp(self):
@@ -489,6 +492,61 @@ class TestAgentClientAttempts(base.TestCase):
             params={'wait': 'false'},
             timeout=60)
 
+    @mock.patch.object(retrying.time, 'sleep', autospec=True)
+    def test__command_succeed_after_agent_token(self, mock_sleep):
+        self.config(require_agent_token=False)
+        mock_sleep.return_value = None
+        error = {'faultstring': 'Unknown Argument: "agent_token"'}
+        response_data = {'status': 'ok'}
+        method = 'standby.run_image'
+        image_info = {'image_id': 'test_image'}
+        params = {'image_info': image_info}
+        i_info = self.node.driver_internal_info
+        i_info['agent_secret_token'] = 'meowmeowmeow'
+        self.client.session.post.side_effect = [
+            MockResponse(json.dumps(error),
+                         status_code=http_client.BAD_REQUEST),
+            MockResponse(json.dumps(response_data)),
+        ]
+
+        response = self.client._command(self.node, method, params)
+        self.assertEqual(2, self.client.session.post.call_count)
+        self.assertEqual(response, response_data)
+        self.client.session.post.assert_called_with(
+            self.client._get_command_url(self.node),
+            data=self.client._get_command_body(method, params),
+            params={'wait': 'false'},
+            timeout=60)
+        self.assertNotIn('agent_secret_token', self.node.driver_internal_info)
+
+    @mock.patch.object(retrying.time, 'sleep', autospec=True)
+    def test__command_fail_agent_token_required(self, mock_sleep):
+        self.config(require_agent_token=True)
+        mock_sleep.return_value = None
+        error = {'faultstring': 'Unknown Argument: "agent_token"'}
+        method = 'standby.run_image'
+        image_info = {'image_id': 'test_image'}
+        params = {'image_info': image_info}
+        i_info = self.node.driver_internal_info
+        i_info['agent_secret_token'] = 'meowmeowmeow'
+        self.client.session.post.side_effect = [
+            MockResponse(json.dumps(error),
+                         status_code=http_client.BAD_REQUEST),
+        ]
+
+        self.assertRaises(exception.AgentAPIError,
+                          self.client._command,
+                          self.node, method, params)
+        self.assertEqual(1, self.client.session.post.call_count)
+        self.client.session.post.assert_called_with(
+            self.client._get_command_url(self.node),
+            data=self.client._get_command_body(method, params),
+            params={'wait': 'false', 'agent_token': 'meowmeowmeow'},
+            timeout=60)
+        self.assertEqual(
+            'meowmeowmeow',
+            self.node.driver_internal_info.get('agent_secret_token'))
+
     @mock.patch.object(retrying.time, 'sleep', autospec=True)
     def test__command_succeed_after_one_timeout(self, mock_sleep):
         mock_sleep.return_value = None

releasenotes/notes/handle-older-agent-command-5930124fd03bb327.yaml

@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    Fixes an issue with agent token handling where the agent has not been
+    upgraded resulting in an AgentAPIError, when the token is not required.
+    The conductor now retries without sending an agent token.