We'll very soon be moving to a cluster setup where people may kill+reschedule our jobs. We'll want a way to be able to resume runs when this happens.