Suppose you have a test that create a task but not run it with run_all_tasks(),
then in a second test run_all_tasks() was running the task of the previous
test...
Fix this by ensuring the task_id (an unique hash generated by celery) is
reachable within the current test database.