A way to specify in config that if a job fails (ideally only because of an infrastructure failure), automatically retry it on a different Runner of the same resource class.