Moe Jette <je...@schedmd.com> writes: > Quoting Bjørn-Helge Mevik <b.h.me...@usit.uio.no>: > >> >> Moe Jette <je...@schedmd.com> writes: >> >>> * Changes in Slurm 13.12.0pre1 >>> ============================== >> >> Just curious: Why the sudden jump in version numbering? year.month? > > Correct. We're Ubuntu fans.
:) >>> -- Add support for load-based scheduling, allocate jobs to nodes with the >>> largest number of available CPUs. >> >> Will it be possible to do the opposite, i.e., allocate jobs to nodes >> with the _fewest_ number of available CPUs? That is something we would >> really like to see. (Our load is a combination of single-threaded jobs >> and parallel jobs, so packing nodes as full as possible keeps nodes free >> for jobs that need to use whole nodes.) > > That is not in the current code or in our plans, but would be pretty simple > to add. It would be a relatively minor variation of the code Stephen > Trofinoff (CSCS) added in the commit below. You are welcome to work on that > and send a patch. > https://github.com/SchedMD/slurm/commit/5e88bc42dca1c6a4f3d29c88e3072ddab344e337 Ok. >>> -- Add mechanism for job_submit plugin to generate error message for srun, >>> salloc or sbatch to log. >> >> But still not to stderr, right? > > The message is logged to stderr. Then I misunderstood; I thought you meant to log file (i.e. slurmctld.log). This is very good news; we've missed a way to generate custom error messages to the user. :) -- Regards, Bjørn-Helge Mevik, dr. scient, Department for Research Computing, University of Oslo