always chain failures

* When inv and proj updates trigger from a JT run, if either update
fails then the job template should get marked failed. Before this
commit, the job template would get marked failed ONLY if there was
enough capacity to run all the associated updates within the same
schedule() call. If, instead, the associated updates were ran in another
schedule() call, the failure chain was lost. This changeset fixes that
by saving the necessary data in the dependent_jobs relationship so that
the failure is always chained.

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

@@ -36,6 +36,7 @@ def scheduler_factory(mocker, epoch):
         def no_create_project_update(task):
             raise RuntimeError("create_project_update should not be called")
 
+        mocker.patch.object(sched, 'capture_chain_failure_dependencies')
         mocker.patch.object(sched, 'get_tasks', return_value=tasks)
         mocker.patch.object(sched, 'get_running_workflow_jobs', return_value=[])
         mocker.patch.object(sched, 'get_inventory_source_tasks', return_value=inventory_sources)

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

@@ -87,3 +87,21 @@ class TestCreateDependentInventoryUpdate():
         scheduler._schedule()
 
         scheduler.start_task.assert_called_with(waiting_inventory_update, [pending_job])
+
+
+class TestCaptureChainFailureDependencies():
+    @pytest.fixture
+    def inventory_id_sources(self, inventory_source_factory):
+        return [
+            (1,  [inventory_source_factory(id=1)]),
+        ]
+
+    def test(self, scheduler_factory, pending_job, waiting_inventory_update, inventory_id_sources):
+        scheduler = scheduler_factory(tasks=[pending_job], 
+                                      create_inventory_update=waiting_inventory_update,
+                                      inventory_sources=inventory_id_sources)
+
+        scheduler._schedule()
+
+        scheduler.capture_chain_failure_dependencies.assert_called_with(pending_job, [waiting_inventory_update])
+