[slurm-users] Slurm not enforcing gres requests at submission

[Stephan Schott]

Well, I might be wrong, but afaik, the SelectType is just telling your
cluster which resources are going to be consumables (in this case,
trackable resources, which is not only gpus). If you want to specify which
conditions have to be met when using the cluster or a determined partition,
then you have to take a look into quality of service (QOS) rules. There you
can set that from the TRES, always at least a gpu has to be assigned, for
example.

El mié., 21 oct. 2020 a las 20:19, Jason Macklin (<Jason.Macklin at jax.org>)
escribió:

> Good day everyone!
>
> We have a GPU-only cluster with Slurm 19.05.5 installed.  My expectation
> with respect to the current configuration is that users submitting jobs
> must include at least one of the following header options:
>
> *--gres*
> *--gpus*
> *--gpus-per-node*
> *--gpus-per-socket*
> *--gpus-per-task*
>
> The current config includes:
>
> SelectType = select/cons_tres
>
> According to the admin guide, this should keep jobs without one of the
> above options from starting.
>
> Am I misinterpreting how this should work?  If so, can someone help me
> understand the issue?
>
> Best regards,
> Jason
> ---
>
> The information in this email, including attachments, may be confidential
> and is intended solely for the addressee(s). If you believe you received
> this email by mistake, please notify the sender by return email as soon as
> possible.
>


-- 
Stephan Schott Verdugo
Biochemist

Heinrich-Heine-Universitaet Duesseldorf
Institut fuer Pharm. und Med. Chemie
Universitaetsstr. 1
40225 Duesseldorf
Germany
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20201024/4db5353a/attachment-0001.htm>