On Mon, 6 Jun 2011 13:29:38 -0700, Bill Johnstone <[email protected]> wrote: > Hello all. > > > I've read the manpage on slurm.conf, as well as the FAQ on this topic ( > https://computing.llnl.gov/linux/slurm/faq.html#rlimit ) and I still have a > few questions. > > For background, I'm running SLURM 2.2.1 on Debian Squeeze (6.0). > > 1. How are the initial limits on the compute node slurmd's established/set? > The /etc/security/limits.d mechanism is PAM-based. How are the limits that > apply to slurmd launched from the init script established in the first place?
Initial slurmd limits are inherited from whatever runs the init script. Typically this is init, but if you start by hand as root it will be the limits of your shell. > > > 2. I don't understand the documentation's mention of "setting ulimit" in the > init script config file (/etc/sysconfig/slurm for RH). Where would the > ulimit invocation need to occur, and how should this be done on non-Redhat > distributions? The equivalent on debian is /etc/default/slurm-llnl. This file is sourced by the startup script, so you would put ulimit commands directly in the file, e.g.: ulimit -Hn 120000 to adjust the limit on number of open file descriptors. > > 3. If using PAM, I'd prefer to fill in some sort of access control in the > "<domain>" field, rather than "*" as indicated in the FAQ. What will the > effective user be that PAM enforces the limits on, when UsePAM=1? Would it > be root? The submitting user? SlurmUser? SLURM should be running through the pam stack with the USER set to the submitting user (the user of the job). If not, that would be a bug. mark > Thank you very much for insight on these. > >
