[slurm-users] Restrict user not use node without GPU

Paul Raines raines at nmr.mgh.harvard.edu
Tue Aug 17 13:40:36 UTC 2021


I do this via code in  job_submit.lua

function slurm_job_submit(job_desc, part_list, submit_uid)
     -- Must allocate GPU if on a GPU partition
     if (job_desc.tres_per_job == nil and job_desc.tres_per_node == nil) then
       if (string.find(job_desc.partition, "rtx8000")) then
         slurm.log_info("slurm_job_submit: for user %u, no GPU requested in rtx8000",
           job_desc.user_id)
         slurm.log_user("FATAL: you must request at least one GPU in the rtx8000 partition");
         return slurm.ERROR
       end
     end
     return slurm.SUCCESS
end


Then set JobSubmitPlugins=lua in slurm.conf

I cannot find any documentation about what really should be in 
tres_per_job and tres_per_node as I would expect the cpu and memory
requests in there but it is still "nil" even when those are given.
For our cluster I have only seen it non-nil when GPUs are requested.



-- Paul Raines (http://help.nmr.mgh.harvard.edu)



On Tue, 17 Aug 2021 6:20am, pravin wrote:

> Dear all,
>
> I am trying to restrict users not to use nodes without GPU but no luck.
> Please help me with this.
> Some users are accessing the GPU nodes and submitting the job(without GPU
> ), so I want to restrict such users not use nodes without GPU.
>
> Thanks
> Pravin
>



More information about the slurm-users mailing list