[slurm-users] Partition Hold/Release

[Nicolas Sonoda]

Hi Marcus and Kevin,

I'm sorry, I forgot to set DefMemPerCPU on my partitions so my preempt was not working.

But after I set, the preempt worked fine and my jobs with low priority can suspend.

Thank you very much!

Regards,
Nícolas

________________________________
De: slurm-users <slurm-users-bounces at lists.schedmd.com> em nome de Kevin Broch <kbroch at rivosinc.com>
Enviado: quarta-feira, 15 de março de 2023 12:51
Para: Slurm User Community List <slurm-users at lists.schedmd.com>
Assunto: Re: [slurm-users] Partition Hold/Release

Nicolas,

It looks like for the partition named "test" you still have PreemptMode=off ?

On Wed, Mar 15, 2023 at 7:35 AM Wagner, Marcus <wagner at itc.rwth-aachen.de<mailto:wagner at itc.rwth-aachen.de>> wrote:

Hi Nicolas,


sorry to say, but we have no experience with preemption.


Best

Marcus


Am 14.03.2023 um 22:07 schrieb Nicolas Sonoda:
Hi Marcus,

Thank you very much for the response.

I set the PriorityTier for my partitions and also set PreemptType=preempt/partition_prio and PreemptMode=SUSPEND,GANG. But the job in the low priority partition does not change it state to SUSPEND. Have any idea?

Following are some information:

slurm.conf:
PreemptType=preempt/partition_prio
PreemptMode=SUSPEND,GANG
NodeName=n[24] Sockets=2 CoresPerSocket=24 ThreadsPerCore=1 RealMemory=185000 State=UNKNOWN
PartitionName=test Nodes=n[24] Default=NO MaxTime=INFINITE State=UP PriorityTier=300 PriorityJobFactor=100 OverSubscribe=FORCE PreemptMode=off AllowGroups=teste
PartitionName=test2 Nodes=n[24] Default=NO MaxTime=INFINITE State=UP PriorityTier=20 PriorityJobFactor=1 OverSubscribe=FORCE PreemptMode=suspend,gang AllowGroups=teste

$ squeue -u vhpc
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
             46003      test     test     vhpc PD       0:00      1 (Resources)
             46002     test2    test2     vhpc  R       0:05      1 n24

Thank you,
Nícolas



________________________________
De: slurm-users em nome de Wagner, Marcus
Enviadas: Terça-feira, 14 de Março de 2023 07:25
Para: slurm-users at lists.schedmd.com<mailto:slurm-users at lists.schedmd.com>
Assunto: Re: [slurm-users] Partition Hold/Release


Hi Nicolas,


you could use the prioritytier for partitions:


       PriorityTier
              Jobs submitted to a partition with a higher PriorityTier value will be evaluated by the scheduler before pending jobs in a partition with a lower PriorityTier value.  They  will
              also  be  considered  for preemption of running jobs in partition(s) with lower PriorityTier values if PreemptType=preempt/partition_prio.  The value may not exceed 65533.  Also
              see PriorityJobFactor.



Best

Marcus


Am 06.03.2023 um 19:33 schrieb Nicolas Sonoda:
Hi!

Can I create a partition with a capacity of hold and release jobs when another partition jobs is submited? For example, the partition one and two can hold their jobs when some job of partition three is submited, and after this job completes the partition one and two releases their jobs again.

Thank you.
Nícolas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20230315/0aa7edfd/attachment-0001.htm>