--- a/taskcluster/taskgraph/actions/add-new-jobs.py
+++ b/taskcluster/taskgraph/actions/add-new-jobs.py
@@ -47,13 +47,12 @@ def add_new_jobs_action(parameters, inpu
             task = full_task_graph.tasks[elem]
 
             # fix up the task's dependencies, similar to how optimization would
             # have done in the decision
             dependencies = {name: label_to_taskid[label]
                             for name, label in task.dependencies.iteritems()}
             task_def = resolve_task_references(task.label, task.task, dependencies)
             task_def.setdefault('dependencies', []).extend(dependencies.itervalues())
-            task_def['schedulerId'] = 'gecko-level-{}'.format(parameters['level'])
             # actually create the new task
-            create_task(slugid(), task_def)
+            create_task(slugid(), task_def, parameters['level'])
         else:
             raise Exception('{} was not found in the task-graph'.format(elem))

--- a/taskcluster/taskgraph/actions/backfill.py
+++ b/taskcluster/taskgraph/actions/backfill.py
@@ -81,12 +81,11 @@ def backfill_action(parameters, input, t
                 'public/label-to-taskid.json')
 
         if label in full_task_graph.tasks.keys():
             task = full_task_graph.tasks[label]
             dependencies = {name: label_to_taskid[label]
                             for name, label in task.dependencies.iteritems()}
             task_def = resolve_task_references(task.label, task.task, dependencies)
             task_def.setdefault('dependencies', []).extend(dependencies.itervalues())
-            task_def['schedulerId'] = 'gecko-level-{}'.format(parameters['level'])
-            create_task(slugid(), task_def)
+            create_task(slugid(), task_def, parameters['level'])
         else:
             logging.info('Could not find {} on {}. Skipping.'.format(label, push))

--- a/taskcluster/taskgraph/actions/run_missing_tests.py
+++ b/taskcluster/taskgraph/actions/run_missing_tests.py
@@ -56,12 +56,12 @@ def run_missing_tests(parameters, input,
     for task in to_run:
 
         # fix up the task's dependencies, similar to how optimization would
         # have done in the decision
         dependencies = {name: label_to_taskid[label]
                         for name, label in task.dependencies.iteritems()}
         task_def = resolve_task_references(task.label, task.task, dependencies)
         task_def.setdefault('dependencies', []).extend(dependencies.itervalues())
-        create_task(slugid(), task_def)
+        create_task(slugid(), task_def, parameters['level'])
 
     logger.info('Out of {} test tasks, {} already existed and the action created {}'.format(
         already_run + len(to_run), already_run, len(to_run)))

--- a/taskcluster/taskgraph/actions/util.py
+++ b/taskcluster/taskgraph/actions/util.py
@@ -24,21 +24,22 @@ testing = False
 def find_decision_task(parameters):
     """Given the parameters for this action, find the taskId of the decision
     task"""
     return find_task_id('gecko.v2.{}.pushlog-id.{}.decision'.format(
         parameters['project'],
         parameters['pushlog_id']))
 
 
-def create_task(task_id, task_def):
+def create_task(task_id, task_def, level):
     """Create a new task.  The task definition will have {relative-datestamp':
     '..'} rendered just like in a decision task.  Action callbacks should use
     this function to create new tasks, as it has the additional advantage of
     allowing easy debugging with `mach taskgraph action-callback --test`."""
+    task_def['schedulerId'] = 'gecko-level-{}'.format(level)
     if testing:
         json.dump([task_id, task_def], sys.stdout,
                   sort_keys=True, indent=4, separators=(',', ': '))
         return
     label = task_def['metadata']['name']
     session = get_session()
     create.create_task(session, task_id, label, task_def)