If by "slurm agent" you mean the slurmd daemon on compute nodes, then
it generally needs to run as user root in order to spawn processes as
various users.
Quoting Anna Kostikova <[email protected]>:
Dear list,
I have a following question: when slurm-agent is running under a user
'slurm', then when I submit a task I have following error:
chown(/tmp/slurmd/job00556): Operation not permitted. If slurm-agent
is run under 'root' user, I don't have the error, but it's really not
ideal from the security perspective. I've tried to change option
SlurmdSpoolDir, but it didn't help. The user 'slurm' has an access to
the directory set in the option SlurmdSpoolDir. Here are my logs:
[2014-11-02T21:45:38.273] Launching batch job 559 for UID 10002
[2014-11-02T21:45:38.276] [559] chown(/tmp/slurmd/job00559): Operation
not permitted
[2014-11-02T21:45:38.276] [559] batch script setup failed for job
559.4294967294
[2014-11-02T21:45:38.276] [559] _step_setup: no job returned
[2014-11-02T21:45:38.276] [559] done with job
What can I do to fix it?
Anna
--
Morris "Moe" Jette
CTO, SchedMD LLC