[slurm-users] "--batch" option of the sbatch command

Loris Bennett loris.bennett at fu-berlin.de
Wed Oct 2 06:23:29 UTC 2019


Hi Tomo,

"Uemoto, Tomoki" <fj2770fj at aa.jp.fujitsu.com> writes:

> Hi,all
> I'm working with slurm 18.08.6 on RHEL7.6
>   manager : 1node
>   computes: 2nodes (c001:haswell,c002:broadwell)
>
> I am checking the --batch option of the sbatch command.
> The following Features were set for testing.
>
>   # scontrol update nodename=c001 Features=haswell
>   # scontrol update nodename=c002 Features=broadwell
>
> And submitted a sleep job.
>
> $ cat sleep_60.sh
> #!/bin/bash
>
> #SBATCH -J sleep_60           # Job name
> #SBATCH -o job.%j.out         # Name of stdout output file (%j expands to jobId)
>
> prun sleep 60
> $
>
> $ sbatch --batch=broadwell --constraint="haswell|broadwell" sleep_60.sh
> $ squeue -l
> Wed Oct  2 13:50:40 2019
>              JOBID PARTITION     NAME     USER    STATE       TIME TIME_LIMI  NODES NODELIST(REASON)
>                 28    normal sleep_60     test  RUNNING       0:03 1-00:00:00      1 c001
> $
>
> I thought the job would be executed in "c002(broadwell)" by the designation of "--batch=broadwell".
> However the job was executed in "c001(haswell)"
> Why isn't it running on "c002(broadwell)" ?
>
> Regards,
> Tomo

I haven't come across this before, but the documentation seems to
indicate that the option '--batch' applies only to the batch step.  The
actual job step, in your case the 'sleep' can then run on any node
satisfying the condition given via the option '--constraint', so in your
case either your Broadwell or you Haswell node.

However, I would have thought that, all things being equal, the batch
step and the job step would run on the same node.  However, I'm not sure
how a constrain with 'or' is resolved if multiple solutions are
available.

What happens if you write the constraint as 

  --constraint="broadwell|haswell"

?

Cheers,

Loris

-- 
Dr. Loris Bennett (Mr.)
ZEDAT, Freie Universität Berlin         Email loris.bennett at fu-berlin.de



More information about the slurm-users mailing list