For some projects, we use group passwords and have users authenticate into the
group when they need to access those files. The password is set with gpasswd
and stored in /etc/gshadow. However after a user authenticates into a group,
they can no longer run SLURM jobs, their jobs go into status "(launch failed
requeued held)".
in slurmd.log I see:
[2016-07-05T13:28:55.167] Launching batch job 2844 for UID 1000
[2016-07-05T13:28:55.175] uid 1000 is not a member of gid 5000
[2016-07-05T13:28:55.175] batch_stepd_step_rec_create() failed: Group ID not
found on host
[2016-07-05T13:28:55.175] _step_setup: no job returned
[2016-07-05T13:28:55.175] done with job
It looks like this is a sanity check looking if a user is a static member of
the group - which they're not, so it fails. Is there any way to turn off this
sanity check?
Below is an example of the commands to reach this point:
[prout@login-0 ~]$ id
uid=1000(prout) gid=1000(prout) groups=1000(prout),10(wheel)
[prout@login-0 ~]$ newgrp ProjectX
Password: abc123
[prout@login-0 ~]$ id
uid=1000(prout) gid=5000(ProjectX) groups=1000(prout),10(wheel),5000(ProjectX)
[prout@login-0 ~]$ sbatch --wrap="srun /bin/sleep 300"
Submitted batch job 2844
[prout@login-0 ~]$ squeue
JOBID PARTITION NAME USER ST TIME NODES
NODELIST(REASON)
2844 normal wrap prout PD 0:00 1 (launch
failed requeued held)
