Hi Magnus, On 15/1/19 4:15 am, Magnus Jonsson wrote:
We have a user that in some way have managed to get both jobs to end up on the same GPU (verified via nvidia-smi).
So this was with nvidia-smi run by root from outside, showing both processes on the same GPU and the other with none? That would be really strange and I've not noticed it before.
All I can think would be to check the /proc/$pid/cgroup file of them to see what cgroups are set and then go poking around in the cgroup filesystem to see what restrictions are set for them.
You don't have Docker installed by some chance either? That could allow users to escape their cgroup settings as it can set up its own.
All the best, Chris -- Chris Samuel : http://www.csamuel.org/ : Berkeley, CA, USA
