I believe that will only limit the maximum size of any single file, not the aggregate file space usage.
Slurm propagates ulimits from the submit environment to the spawned tasks, but a job_submit plugin could overwrite the environment variable in which the ulimit information is passed or a gres/file that you develop could set the ulimit based upon the user specification. In either case, the exercise is left to the reader.
Quoting Jens Dreger <[email protected]>:
Hi! I'm in the process of migrating a torque/MAUI based cluster to slurm. On the torque system, users can specify '-l file=10gb' to request 10gb of space on the local disk. The prolog creates the corresponding directory for the user and torque sets a 'ulimit -f 10485760'. On my slurm test system I created a generic resource for the local disk which works fine, but I can't establish a ulimit and slurm doesn't set it. If I set ulimit -f in the prolog script, it's not propagated to the jobs since it only affects subshells. Does anyone have an idea how I could prevent users from using more local diskspace than they asked for? Thanks, Jens. -- Jens Dreger Freie Universitaet Berlin [email protected] Fachbereich Physik - ZEDV Tel: +49 30 83854774 Arnimallee 14 Fax: +49 30 83855902 14195 Berlin
