It's just a hunch, but also we should do something about it. For example if the failure is critical (say a dependency isn't found) we should at least log an ERROR, and no point in keeping rescheduling new ones..