Update DryDock Operator & Shipyard Chart

This Patch Set is meant to expose the 'query_interval' and
'task_timeout' parameters for Drydock tasks in Shipyard.
This will allow us to specify the values for a particular
site.

The corresponding changes for the Helm Chart are included in
this Patch Set as well.

It is also noted that the task has been updated to 'prepare_nodes'
and 'deploy_nodes' instead.

Task State can either be 'completed' or 'terminated'. These new
changes have been captured in this Patch Set as well.

Change-Id: I1b446f7bcf493bc8e5bbfdba842158797f0e3594

charts/shipyard/templates/etc/_shipyard.conf.tpl

@@ -51,7 +51,15 @@
 {{ if not .armada.service_type }}#{{ end }}service_type = {{ .armada.service_type | default "armada" }}
 
 [drydock]
-{{ if not .drydock.service_type }}#{{ end }}service_type = {{ .drydock.service_type | default "drydock" }}
+{{ if not .drydock.service_type }}#{{ end }}service_type = {{ .drydock.service_type | default "physicalprovisioner" }}
+{{ if not .drydock.verify_site_query_interval }}#{{ end }}verify_site_query_interval = {{ .drydock.verify_site_query_interval | default "10" }}
+{{ if not .drydock.verify_site_task_timeout }}#{{ end }}verify_site_task_timeout = {{ .drydock.verify_site_task_timeout | default "60" }}
+{{ if not .drydock.prepare_site_query_interval }}#{{ end }}prepare_site_query_interval = {{ .drydock.prepare_site_query_interval | default "10" }}
+{{ if not .drydock.prepare_site_task_timeout }}#{{ end }}prepare_site_task_timeout = {{ .drydock.prepare_site_task_timeout | default "120" }}
+{{ if not .drydock.prepare_node_query_interval }}#{{ end }}prepare_node_query_interval = {{ .drydock.prepare_node_query_interval | default "30" }}
+{{ if not .drydock.prepare_node_task_timeout }}#{{ end }}prepare_node_task_timeout = {{ .drydock.prepare_node_task_timeout | default "1800" }}
+{{ if not .drydock.deploy_node_query_interval }}#{{ end }}deploy_node_query_interval = {{ .drydock.deploy_node_query_interval | default "30" }}
+{{ if not .drydock.deploy_node_task_timeout }}#{{ end }}deploy_node_task_timeout = {{ .drydock.deploy_node_task_timeout | default "3600" }}
 
 [healthcheck]
 {{ if not .healthcheck.schema }}#{{ end }}schema = {{ .healthcheck.schema | default "<None>" }}

charts/shipyard/values.yaml

@@ -297,7 +297,15 @@ conf:
     armada:
       service_type: armada
     drydock:
-      service_type: drydock
+      service_type: physicalprovisioner
+      verify_site_query_interval: 10
+      verify_site_task_timeout: 60
+      prepare_site_query_interval: 10
+      prepare_site_task_timeout: 120
+      prepare_node_query_interval: 30
+      prepare_node_task_timeout: 1800
+      deploy_node_query_interval: 30
+      deploy_node_task_timeout: 3600
     healthcheck:
       schema: http
       endpoint: /api/v1.0/health

shipyard_airflow/conf/config.py

@@ -130,6 +130,46 @@ SECTIONS = [
                     'the service lookup in the Keystone service catalog. '
                 )
             ),
+            cfg.IntOpt(
+                'verify_site_query_interval',
+                default=10,
+                help='Query interval (in seconds) for verify_site task'
+            ),
+            cfg.IntOpt(
+                'verify_site_task_timeout',
+                default=60,
+                help='Time out (in seconds) for verify_site task'
+            ),
+            cfg.IntOpt(
+                'prepare_site_query_interval',
+                default=10,
+                help='Query interval (in seconds) for prepare_site task'
+            ),
+            cfg.IntOpt(
+                'prepare_site_task_timeout',
+                default=120,
+                help='Time out (in seconds) for prepare_site task'
+            ),
+            cfg.IntOpt(
+                'prepare_node_query_interval',
+                default=30,
+                help='Query interval (in seconds) for prepare_node task'
+            ),
+            cfg.IntOpt(
+                'prepare_node_task_timeout',
+                default=1800,
+                help='Time out (in seconds) for prepare_node task'
+            ),
+            cfg.IntOpt(
+                'deploy_node_query_interval',
+                default=30,
+                help='Query interval (in seconds) for deploy_node task'
+            ),
+            cfg.IntOpt(
+                'deploy_node_task_timeout',
+                default=3600,
+                help='Time out (in seconds) for deploy_node task'
+            ),
         ]
     ),
     ConfigSection(

shipyard_airflow/dags/drydock_deploy_site.py

@@ -24,8 +24,8 @@ config_path = '/usr/local/airflow/plugins/shipyard.conf'
 CREATE_DRYDOCK_CLIENT_DAG_NAME = 'create_drydock_client'
 DRYDOCK_VERIFY_SITE_DAG_NAME = 'verify_site'
 DRYDOCK_PREPARE_SITE_DAG_NAME = 'prepare_site'
-DRYDOCK_PREPARE_NODE_DAG_NAME = 'prepare_node'
-DRYDOCK_DEPLOY_NODE_DAG_NAME = 'deploy_node'
+DRYDOCK_PREPARE_NODE_DAG_NAME = 'prepare_nodes'
+DRYDOCK_DEPLOY_NODE_DAG_NAME = 'deploy_nodes'
 
 
 def get_drydock_subdag_step(parent_dag_name, child_dag_name, args):
@@ -84,14 +84,14 @@ def deploy_site_drydock(parent_dag_name, child_dag_name, args):
         task_id=DRYDOCK_PREPARE_SITE_DAG_NAME,
         dag=dag)
 
-    drydock_prepare_node = SubDagOperator(
+    drydock_prepare_nodes = SubDagOperator(
         subdag=get_drydock_subdag_step(dag.dag_id,
                                        DRYDOCK_PREPARE_NODE_DAG_NAME,
                                        args),
         task_id=DRYDOCK_PREPARE_NODE_DAG_NAME,
         dag=dag)
 
-    drydock_deploy_node = SubDagOperator(
+    drydock_deploy_nodes = SubDagOperator(
         subdag=get_drydock_subdag_step(dag.dag_id,
                                        DRYDOCK_DEPLOY_NODE_DAG_NAME,
                                        args),
@@ -101,7 +101,7 @@ def deploy_site_drydock(parent_dag_name, child_dag_name, args):
     # DAG Wiring
     drydock_verify_site.set_upstream(drydock_client)
     drydock_prepare_site.set_upstream(drydock_verify_site)
-    drydock_prepare_node.set_upstream(drydock_prepare_site)
-    drydock_deploy_node.set_upstream(drydock_prepare_node)
+    drydock_prepare_nodes.set_upstream(drydock_prepare_site)
+    drydock_deploy_nodes.set_upstream(drydock_prepare_nodes)
 
     return dag

shipyard_airflow/plugins/drydock_operators.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import configparser
 import logging
 import os
 import time
@@ -135,32 +136,57 @@ class DryDockOperator(BaseOperator):
         else:
             raise AirflowException("Unable to Retrieve Genesis Node IP!")
 
+        # Read shipyard.conf
+        config = configparser.ConfigParser()
+        config.read(self.shipyard_conf)
+
+        if not config.read(self.shipyard_conf):
+            raise AirflowException("Unable to read content of shipyard.conf")
+
         # Create Task for verify_site
         if self.action == 'verify_site':
 
             # Default settings for 'verify_site' execution is to query
             # the task every 10 seconds and to time out after 60 seconds
-            # TODO: Need to decide if we want to make polling interval and
-            # time out a variable in the Dags
-            self.drydock_action(drydock_client, context, self.action, 10, 60)
+            query_interval = config.get('drydock',
+                                        'verify_site_query_interval')
+            task_timeout = config.get('drydock', 'verify_site_task_timeout')
+
+            self.drydock_action(drydock_client, context, self.action,
+                                query_interval, task_timeout)
 
         # Create Task for prepare_site
         elif self.action == 'prepare_site':
             # Default settings for 'prepare_site' execution is to query
             # the task every 10 seconds and to time out after 120 seconds
-            self.drydock_action(drydock_client, context, self.action, 10, 120)
+            query_interval = config.get('drydock',
+                                        'prepare_site_query_interval')
+            task_timeout = config.get('drydock', 'prepare_site_task_timeout')
+
+            self.drydock_action(drydock_client, context, self.action,
+                                query_interval, task_timeout)
 
         # Create Task for prepare_node
-        elif self.action == 'prepare_node':
+        elif self.action == 'prepare_nodes':
             # Default settings for 'prepare_node' execution is to query
             # the task every 30 seconds and to time out after 1800 seconds
-            self.drydock_action(drydock_client, context, self.action, 30, 1800)
+            query_interval = config.get('drydock',
+                                        'prepare_node_query_interval')
+            task_timeout = config.get('drydock', 'prepare_node_task_timeout')
+
+            self.drydock_action(drydock_client, context, self.action,
+                                query_interval, task_timeout)
 
         # Create Task for deploy_node
-        elif self.action == 'deploy_node':
+        elif self.action == 'deploy_nodes':
             # Default settings for 'deploy_node' execution is to query
             # the task every 30 seconds and to time out after 3600 seconds
-            self.drydock_action(drydock_client, context, self.action, 30, 3600)
+            query_interval = config.get('drydock',
+                                        'deploy_node_query_interval')
+            task_timeout = config.get('drydock', 'deploy_node_task_timeout')
+
+            self.drydock_action(drydock_client, context, self.action,
+                                query_interval, task_timeout)
 
         # Do not perform any action
         else:
@@ -247,7 +273,10 @@ class DryDockOperator(BaseOperator):
     def drydock_query_task(self, drydock_client, interval, time_out, task_id):
 
         # Calculate number of times to execute the 'for' loop
-        end_range = int(time_out / interval)
+        # Convert 'time_out' and 'interval' from string into integer
+        # The result from the division will be a floating number which
+        # We will round off to nearest whole number
+        end_range = round(int(time_out) / int(interval))
 
         # Query task status
         for i in range(0, end_range + 1):
@@ -264,11 +293,12 @@ class DryDockOperator(BaseOperator):
             if task_status == 'running' and i == end_range:
                 raise AirflowException("Task Execution Timed Out!")
 
-            # Exit 'for' loop if task is in 'complete' state
-            if task_status == 'complete':
+            # Exit 'for' loop if the task is in 'complete' or 'terminated'
+            # state
+            if task_status in ['complete', 'terminated']:
                 break
             else:
-                time.sleep(interval)
+                time.sleep(int(interval))
 
         # Get final task state
         # NOTE: There is a known bug in Drydock where the task result