I am running hourly job on a cluster using p3.2xlarge GPU instance, but sometimes cluster couldn't start due to instance unavailability. I wander is there is any fallback mechanism to, for example, try a different instance type if one is not availabl...