Hi, Am 25.10.2016 um 13:20 schrieb Patrice Peterson: > is there a build-in way to queue LoadLeveler-like job steps in SLURM? > Something like this: > > #!/bin/bash > #SBATCH --num-tasks=1 > echo "prepping data, simple stuff" > #SBATCH --- END STEP > > #SBATCH --num-tasks=4 > echo "main run, needs lots of resources" > srun main_program > #SBATCH --- END STEP > > #SBATCH --num-tasks=1 > echo "cleaning up data, simple stuff" > > which would then submit three more-or-less "separate" jobs with > dependencies on each other?
Hmm, interesting question! Made me overthink how dependencies get modeled here... And I found https://hpc.nih.gov/docs/job_dependencies.html gives a better overview than the slurm docs -- you already know all of that? What I'm even more unsure of is if the subtleties of (not) using per job prolog/epilog is something to consider too... like a possible requeue after a failed prolog and the difference between a dedicated cleaning job and an epilog that doesn't care to much in case of errors. Not really an answer. Just idea I think might be related. If not: sorry for the noise. Regards, Benjamin -- FSU Jena | JULIELab.de/Staff/Benjamin+Redling.html vox: +49 3641 9 44323 | fax: +49 3641 9 44321
