Retry agent get_command_status upon failures

The agent command status code lacks any retry mechanism
which meant if any intermittent failure such as a dropped
packet or an overloaded firewall could potentially begin
to cause the entire deployment or cleaning process to derail
and fail.

This fix addes logic to ensure we retry upon such failures.

Worth noting, the exact same logic has been used elsewhere
in the agent client code for the exact same problem when
issuing commands.

Change-Id: I4f6581b7fb895ed2b1d505b9947e363665551b57
Story: 2007470
Task: 39158
(cherry picked from commit 242775ae18)

ironic/drivers/modules/agent_client.py

@@ -133,6 +133,10 @@ class AgentClient(object):
         return result
 
     @METRICS.timer('AgentClient.get_commands_status')
+    @retrying.retry(
+        retry_on_exception=(
+            lambda e: isinstance(e, exception.AgentConnectionFailed)),
+        stop_max_attempt_number=CONF.agent.max_command_attempts)
     def get_commands_status(self, node):
         """Get command status from agent.
 
@@ -162,7 +166,16 @@ class AgentClient(object):
         """
         url = self._get_command_url(node)
         LOG.debug('Fetching status of agent commands for node %s', node.uuid)
-        resp = self.session.get(url, timeout=CONF.agent.command_timeout)
+        try:
+            resp = self.session.get(url, timeout=CONF.agent.command_timeout)
+        except (requests.ConnectionError, requests.Timeout) as e:
+            msg = (_('Failed to connect to the agent running on node %(node)s '
+                     'to collect commands status. '
+                     'Error: %(error)s') %
+                   {'node': node.uuid, 'error': e})
+            LOG.error(msg)
+            raise exception.AgentConnectionFailed(reason=msg)
+
         result = resp.json()['commands']
         status = '; '.join('%(cmd)s: result "%(res)s", error "%(err)s"' %
                            {'cmd': r.get('command_name'),

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

@@ -192,6 +192,17 @@ class TestAgentClient(base.TestCase):
                     'api_version': CONF.agent.agent_api_version},
                 timeout=CONF.agent.command_timeout)
 
+    def test_get_commands_status_retries(self):
+        with mock.patch.object(self.client.session, 'get',
+                               autospec=True) as mock_get:
+            res = mock.MagicMock(spec_set=['json'])
+            res.json.return_value = {'commands': []}
+            mock_get.side_effect = [
+                requests.ConnectionError('boom'),
+                res]
+            self.assertEqual([], self.client.get_commands_status(self.node))
+            self.assertEqual(2, mock_get.call_count)
+
     def test_prepare_image(self):
         self.client._command = mock.MagicMock(spec_set=[])
         image_info = {'image_id': 'image'}

releasenotes/notes/agent-command-status-retry-f9b6f53a823c6b01.yaml

@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    Fixes an issue with the agent client code where checks of the agent
+    command status had no logic to prevent an intermittent or transient
+    connection failure from causing the entire operation to fail.