run parsl inside a theta cobalt job, rather than have parsl submit the cobalt jobs.
This fits the model we've been using for runs a bit better, of acquiring resources fairly manually with a reservation and then running parsl when that allocation is available.
It would also remove the need to have a persistent parsl process which outlives any one job, and which must stay alive (or be restarted in a suitable fashion) over login node reboots.
run parsl inside a theta cobalt job, rather than have parsl submit the cobalt jobs.
This fits the model we've been using for runs a bit better, of acquiring resources fairly manually with a reservation and then running parsl when that allocation is available.
It would also remove the need to have a persistent parsl process which outlives any one job, and which must stay alive (or be restarted in a suitable fashion) over login node reboots.