Hi,
We're experiencing trouble with slurm and sssd users & groups. When running a job on a node, we get the following error in /var/log/slurm-llnl/slurmd.log [2016-09-09T15:16:46.888] _run_prolog: run job script took usec=487177 [2016-09-09T15:16:46.888] _run_prolog: prolog with lock for job 298 ran for 0 seconds [2016-09-09T15:16:46.888] Launching batch job 298 for UID 89532 [2016-09-09T15:16:46.894] uid 89532 is not a member of gid 84762 [2016-09-09T15:16:46.894] batch_stepd_step_rec_create() failed: Group ID not found on host [2016-09-09T15:16:46.894] _step_setup: no job returned [2016-09-09T15:16:46.894] done with job [2016-09-09T15:16:46.894] error: slurmstepd return code 4006 Our idea was to add a prolog script that does the "getent passwd username" and "getent group groupname" so that the check that is run afterwards will pass... It doesn't resolve the issue. By the way, it seems to be a mater of time (cache?) ... because after a small break (~10min), when we launch a new job, it passes : [2016-09-09T15:50:38.801] _run_prolog: run job script took usec=185203 [2016-09-09T15:50:38.801] _run_prolog: prolog with lock for job 303 ran for 0 seconds [2016-09-09T15:50:38.801] Launching batch job 303 for UID 89532 [2016-09-09T15:50:43.941] [303] sending REQUEST_COMPLETE_BATCH_SCRIPT, error:0 status 0 [2016-09-09T15:50:43.944] [303] done with job We used to compile slurm from sources to disable the check (that was with slurm version 2.6.1). But now, except this issue, we'd prefer to stay with a package (Ubuntu 16.04 offers slurm 15.08.7). By the way, we've read http://slurm.schedmd.com/faq.html#sssd . But it's not possible to "enumerate" our school's ldap on the nodes. Also interesting, it says that 16.05 "does not need enumeration". That's cool ... but not yet packaged. Any idea to get this fixed? Regards, Samuel Bancal -- Samuel Bancal ENAC-IT EPFL
