[slurm-users] howto limit the cpu resource for each user

[Colas Rivière]

Hello Arielle,

I don't have a full answer, but here is a start:

Yes, you first need at least 
"AccountingStorageEnforce=associations,limits" (and qos is you want to 
use it) so that the limits you set are enforced (see 
https://slurm.schedmd.com/resource_limits.html)

Then you can set limits for each user eg with "sacctmgr modify account 
where account=<account> set MaxJobs=50" to set the max number of jobs 
for each of the users of this account (replace <account> as needed). 
Then maybe set it for a single user: "sacctmgr modify account where 
account=<account> user=<user> set MaxJobs=500".

You can check what is set with "sacctmgr show associations", and remove 
what has been set by setting it to -1.

I hope this helps,
Colas

On 2018-01-18 12:22, Arielle Willm wrote:
>
> Hi,
>
> slurm is installed in a minimal configuration for a cluster of 
> 3000cores/170 nodes.We have 4 partitions, one for each type of nodes; 
> each partition is available for all users.
> We want to prevent each user from taking more than 1000 cores running 
> on up to 50 jobs on all the cluster, and I'm looking for guidance on 
> how we should do so.
>
> I have already read a lot of documentation about it , i tried but i 
> have no result
>
> - Below a list of some config parameters :
>
>     ClusterName=cluster
>     # Accounting.
>     AccountingStorageType=accounting_storage/slurmdbd
>     AccountingStorageUser=slurm
>     #AccountingStorageEnforce=qos,limits
>     #AccountingStoragePass=
>     #AccountingStoragePort=
>     EnforcePartLimits=YES
>     # SCHEDULING
>     SchedulerType=sched/backfill
>     # Consumable Resource
>     SelectType=select/cons_res
>     SelectTypeParameters=CR_Core
>
> - about the config in the slurmDB , there are not many things, only 
> account root and only user root
>
> sacctmgr show account
>    Account                Descr                  Org
> ---------- -------------------- --------------------
>       root default root account                 root
>
> I have tried : sacctmgr modify cluster set GrpTRES=cpu=1040
>
> but it does not work !
>
> What is wrong ?
>
>
> Did we need to uncomment the parameter AccountingStorageEnforce ?
>
> Did we need to define all the users ?
>
> Can someone show the way to proceed?
>
> Thanks for your help
>
> A Willm
>
> -- 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20180118/65f3561a/attachment.html>