when a singularity run fails fast on a broken node, that node can then end up absorbing all the work by executing-and-completing(failing) much faster than the other nodes
possibly parsl can be made to rate limit per worker node to make this less of a problem
when a singularity run fails fast on a broken node, that node can then end up absorbing all the work by executing-and-completing(failing) much faster than the other nodes
possibly parsl can be made to rate limit per worker node to make this less of a problem