Specifying memory requirements for jobs doesn't have effect about cloudman HOT 9 CLOSED

No problem. Forgot to mention but the new code is in cloudman-test bucket.

from cloudman.

@natefoo, would you have some ideas/suggestions why this might be the case?

from cloudman.

@afgane can you point me at Cloudman's slurm.conf template?

@hackdna If you're using a relatively stock slurm.conf, it will use the linear scheduling plugin. To control memory/core usage you need to enable the consumable resources plugin: http://slurm.schedmd.com/cons_res.html

from cloudman.

@natefoo Here it is: https://github.com/galaxyproject/cloudman/blob/master/cm/conftemplates/slurm.conf.default
As nodes get added, the file is edited to add node info under the COMPUTE NODES 'section': https://github.com/galaxyproject/cloudman/blob/master/cm/services/apps/jobmanagers/slurmctld.py#L115

from cloudman.

I'm using unmodified CM slurm.conf which contains:

SchedulerType=sched/backfill
SelectType=select/cons_res

I've tried adding SelectTypeParameters=CR_Core_Memory as the Slurm docs suggested but slurm.conf keeps getting reverted for some reason after slurmctld is restarted.

from cloudman.

Can you try restarting slurmctld via the command line vs. using CloudMan? CloudMan rewrites Slurm's conf when the service is restarted (the refactored CloudMan will support user configuration changes but that's months away). The command CloudMan is using should be in its log file.

from cloudman.

Thanks, that worked. Would you mind adding SelectTypeParameters=CR_Core_Memory to slurm.conf as a default?
Speaking of restarting services via command line, what is the right way to do that for Galaxy in CM?

from cloudman.

Done in a64eb57.

For Galaxy, take a look at this page, I just added the exact command: https://wiki.galaxyproject.org/CloudMan/Services/Galaxy

from cloudman.

Thank you @afgane and @natefoo.

from cloudman.