[slurm-users] strange resource allocation issue - thoughts?

[Sharma, M D]

Hi Prentice,

Thanks, that was one of the many different ways we had tested but still had the same issue.

Best Regards,
MD Sharma(sent via a mobile device.. please blame the AI driven autocorrect algorithm for any errors)

________________________________
From: slurm-users <slurm-users-bounces at lists.schedmd.com> on behalf of Prentice Bisbal <pbisbal at pppl.gov>
Sent: Wednesday, March 27, 2019 3:02:30 PM
To: slurm-users at lists.schedmd.com
Subject: Re: [slurm-users] strange resource allocation issue - thoughts?


On 3/23/19 2:16 PM, Sharma, M D wrote:
Hi folks,

By default slurm allocates the whole node for a job (even if it specifically requested a single core). This is usually taken care of by adding SelectType=select/cons_res along with an appropriate parameter such as  SelectTypeParameters=CR_Core_Memory.

When testing the job submission and resource allocation, we can see things work as intended when using srun:

srun -N1 -n1 -p fxq --mem=1000 sleep 60 &

# A command as above, submitted 20 times would launch 20 jobs on a single 40 core node as intended.

However, if the same request is submitted via sbatch, the entire node gets into an "allocated" state and does not accept any other jobs until completion of the single core job.

Has anyone else seen this behaviour / have thoughts on a fix?



I believe by requesting -N1, you are requesting the entire node be allocation to that job. Remove -N1 from that command, like this:

srun -n1 -p fxq --mem=1000 sleep 60 &

That should allow you to run multiple jobs on the same node.


--
Prentice

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20190327/84fd3b14/attachment.html>