Enhanced Drydock Health Check

- Added a database connection check to the health check.
- Added a MaaS connection check to the health check.

Change-Id: I6c771a9ed6a278eb75dbc0f2e503088ff9046149

drydock_provisioner/control/api.py

@@ -55,7 +55,8 @@ def start_api(state_manager=None, ingester=None, orchestrator=None):
     # v1.0 of Drydock API
     v1_0_routes = [
         # API for managing orchestrator tasks
-        ('/health', HealthResource()),
+        ('/health', HealthResource(state_manager=state_manager,
+                                   orchestrator=orchestrator)),
         ('/tasks',
          TasksResource(state_manager=state_manager,
                        orchestrator=orchestrator)),

drydock_provisioner/control/health.py

@@ -12,8 +12,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import falcon
+import json
 
 from drydock_provisioner.control.base import BaseResource
+from drydock_provisioner.drivers.node.maasdriver.actions.node import ValidateNodeServices
+from drydock_provisioner.objects.fields import ActionResult
+import drydock_provisioner.objects.fields as hd_fields
 
 
 class HealthResource(BaseResource):
@@ -21,9 +25,49 @@ class HealthResource(BaseResource):
     Return empty response/body to show
     that Drydock is healthy
     """
+    def __init__(self, state_manager=None, orchestrator=None, **kwargs):
+        """Object initializer.
+
+        :param state_manager: instance of Drydock state_manager
+        """
+        super().__init__(**kwargs)
+        self.state_manager = state_manager
+        self.orchestrator = orchestrator
 
     def on_get(self, req, resp):
         """
-        It really does nothing right now. It may do more later
+        Returns 204 on success, otherwise 500 with a response body.
         """
+        healthy = True
+        # Test database connection
+        try:
+            now = self.state_manager.get_now()
+            if now is None:
+                raise Exception('None received from database for now()')
+        except Exception as ex:
+            healthy = False
+            resp.body = json.dumps({
+                'type': 'error',
+                'message': 'Database error',
+                'retry': True
+            })
+            resp.status = falcon.HTTP_500
+
+        # Test MaaS connection
+        try:
+            task = self.orchestrator.create_task(action=hd_fields.OrchestratorAction.Noop)
+            maas_validation = ValidateNodeServices(task, self.orchestrator, self.state_manager)
+            maas_validation.start()
+            if maas_validation.task.get_status() == ActionResult.Failure:
+                raise Exception('MaaS task failure')
+        except Exception as ex:
+            healthy = False
+            resp.body = json.dumps({
+                'type': 'error',
+                'message': 'MaaS error',
+                'retry': True
+            })
+            resp.status = falcon.HTTP_500
+
+        if healthy:
             resp.status = falcon.HTTP_204

drydock_provisioner/drivers/node/maasdriver/actions/node.py

@@ -66,7 +66,7 @@ class ValidateNodeServices(BaseMaasAction):
                     ctx_type='NA')
                 self.task.success()
                 if self.maas_client.test_authentication():
-                    self.logger.info("Able to authenitcate with MaaS API.")
+                    self.logger.info("Able to authenticate with MaaS API.")
                     self.task.add_status_msg(
                         msg='Able to authenticate with MaaS API.',
                         error=False,

drydock_provisioner/statemgmt/state.py

@@ -667,3 +667,24 @@ class DrydockState(object):
         except Exception as ex:
             self.logger.error("Error selecting build data.", exc_info=ex)
             raise errors.BuildDataError("Error selecting build data.")
+
+    def get_now(self):
+        """Query the database for now() from dual.
+        """
+        try:
+            with self.db_engine.connect() as conn:
+                query = sql.text("SELECT now()")
+                rs = conn.execute(query)
+
+                r = rs.first()
+
+                if r is not None and r.now:
+                    return r.now
+                else:
+                    return None
+        except Exception as ex:
+            self.logger.error(str(ex))
+            self.logger.error(
+                "Error querying for now()",
+                exc_info=True)
+            return None

tests/integration/postgres/test_api_health.py

@@ -18,8 +18,8 @@ from drydock_provisioner.control.health import HealthResource
 import falcon
 
 
-def test_get_health(mocker):
-    api = HealthResource()
+def test_get_health(mocker, deckhand_orchestrator, drydock_state):
+    api = HealthResource(state_manager=drydock_state, orchestrator=deckhand_orchestrator)
 
     # Configure mocked request and response
     req = mocker.MagicMock(spec=falcon.Request)