Merge pull request #5568 from chrismeyersfsu/fix-5562

revert waiting task stuck fix

awx/main/models/unified_jobs.py

@@ -930,8 +930,7 @@ class UnifiedJob(PolymorphicModel, PasswordFieldsModel, CommonModelNameNotUnique
 
     def start_celery_task(self, opts, error_callback, success_callback):
         task_class = self._get_task_class()
-        async_result = task_class().apply_async((self.pk,), opts, link_error=error_callback, link=success_callback)
+        task_class().apply_async((self.pk,), opts, link_error=error_callback, link=success_callback)
-        return async_result.id
 
     def start(self, error_callback, success_callback, **kwargs):
         '''

awx/main/scheduler/__init__.py

@@ -60,7 +60,8 @@ class TaskManager():
     '''
     Tasks that are running and SHOULD have a celery task.
     '''
-    def get_running_tasks(self, status_list=('running',)):
+    def get_running_tasks(self):
+        status_list = ('running',)
 
         jobs = JobDict.filter_partial(status=status_list)
         inventory_updates = InventoryUpdateDict.filter_partial(status=status_list)
@@ -215,15 +216,15 @@ class TaskManager():
         else:
             if type(job_obj) is WorkflowJob:
                 job_obj.status = 'running'
-            else:
+
-                celery_task_id = job_obj.start_celery_task(opts, error_callback=error_handler, success_callback=success_handler)
+            job_obj.save()
-                job_obj.celery_task_id = celery_task_id
 
             self.consume_capacity(task)
-            job_obj.save()
 
         def post_commit():
             job_obj.websocket_emit_status(job_obj.status)
+            if job_obj.status != 'failed':
+                job_obj.start_celery_task(opts, error_callback=error_handler, success_callback=success_handler)
 
         connection.on_commit(post_commit)
 

awx/main/scheduler/dependency_graph.py

@@ -83,6 +83,11 @@ class DependencyGraph(object):
     '''
     def should_update_related_project(self, job):
         now = self.get_now()
+
+        # Already processed dependencies for this job
+        if job.data['dependent_jobs__id'] is not None:
+            return False
+
         latest_project_update = self.data[self.LATEST_PROJECT_UPDATES].get(job['project_id'], None)
         if not latest_project_update:
             return True

awx/main/scheduler/tasks.py

@@ -52,7 +52,7 @@ def run_fail_inconsistent_running_jobs():
                 # TODO: Failed to contact celery. We should surface this.
                 return None
 
-            all_running_sorted_tasks = scheduler.get_running_tasks(status_list=('running', 'waiting',))
+            all_running_sorted_tasks = scheduler.get_running_tasks()
             scheduler.process_celery_tasks(active_tasks, all_running_sorted_tasks)
         except DatabaseError:
             return

awx/main/tasks.py

@@ -665,7 +665,7 @@ class BaseTask(Task):
         '''
         Run the job/task and capture its output.
         '''
-        instance = self.update_model(pk, status='running')
+        instance = self.update_model(pk, status='running', celery_task_id='' if self.request.id is None else self.request.id)
 
         instance.websocket_emit_status("running")
         status, rc, tb = 'error', None, ''

awx/main/tests/unit/scheduler/test_dependency_graph.py

@@ -17,8 +17,10 @@ def graph():
 
 
 @pytest.fixture
-def job():
+def job(job_factory):
-    return dict(project_id=1)
+    j = job_factory()
+    j.project_id = 1
+    return j
 
 
 @pytest.fixture
@@ -36,13 +38,11 @@ def unsuccessful_last_project(graph, job):
 
 
 @pytest.fixture
-def last_dependent_project(graph):
+def last_dependent_project(graph, job):
     now = tz_now()
 
-    job = {
+    job['project_id'] = 1
-        'project_id': 1,
+    job['created'] = now
-        'created': now,
-    }
     pu = ProjectUpdateDict(dict(id=1, project_id=1, status='waiting',
                                 project__scm_update_cache_timeout=0,
                                 launch_type='dependency',
@@ -57,10 +57,8 @@ def last_dependent_project(graph):
 def timedout_project_update(graph, job):
     now = tz_now()
 
-    job = {
+    job['project_id'] = 1
-        'project_id': 1,
+    job['created'] = now
-        'created': now,
-    }
     pu = ProjectUpdateDict(dict(id=1, project_id=1, status='successful',
                                 project__scm_update_cache_timeout=10,
                                 launch_type='dependency',
@@ -76,10 +74,8 @@ def timedout_project_update(graph, job):
 def not_timedout_project_update(graph, job):
     now = tz_now()
 
-    job = {
+    job['project_id'] = 1
-        'project_id': 1,
+    job['created'] = now
-        'created': now,
-    }
     pu = ProjectUpdateDict(dict(id=1, project_id=1, status='successful',
                                 project__scm_update_cache_timeout=3600,
                                 launch_type='dependency',