On 23/02/18 21:50, Loris Bennett wrote:
OK, I'm confused now. Our main culprit for producing processes with
incorrect affinity is ORCA [1]. It uses OpenMPI but also likes to
start processes asynchronously via SSH within the node set.
In that case (and for the general case where there are wrappers
involved) I'd leave it as is. As long as the OpenMPI it is using
is compiled against Slurm you're probably OK for those components.
You can also use pam_slurm_adopt to ensure SSH sessions are captured
into a users job cgroup too, which might help to limit damage.
Best of luck!
Chris
