How to submit .ksh scripts on HPC using sbatch for WRFDA without staying on the head node?

[haritha1022]

Hi,

I am currently working on generating background error (BE) statistics using WRFDA on an HPC cluster that uses SLURM for job submission.

I was able to successfully run gen_be_stage0, stage1, stage2, and stage3 by directly executing the .exe files through simple shell commands, even though some of those executions happened on the head node (which I understand is not recommended).

However, I am stuck at stage4 (gen_be_stage4). I am unable to run this script properly, and I’m unsure how to write an appropriate SLURM shell script for it. The wrapper script seems to be originally written for an environment like LSF (e.g., using bsub or rsh), but I want to execute it via sbatch on compute nodes, not on the head node.

Additionally, I have a main .ksh script (ge_be_wrapper.ksh) that calls another script (gen_be.ksh), which in turn calls gen_be_stage4_regional.ksh. These scripts execute several stages of the workflow, including calls to .exe binaries.

I want to submit this full workflow using sbatch, without staying on the head node or running interactively there. My goal is to just submit the job and let the compute nodes handle the execution.

Could you please advise:

• How to structure these .ksh scripts for SLURM?
• What needs to be done differently to avoid head node usage?
• Any best practices for running KornShell-based workflows with sbatch?

Thank you for your guidance!

gen_be_stage4_regional.docx

gen_be_wrapper.docx

gen_be.docx please find the attachemets

[saper]

Hello @haritha1022 - are you sure this is the right place to ask? I am afraid most of us have no idea what is a SLURM script and things like that.

If you run into a particular error message that is related to Korn Shell, maybe you can find some help among the Korn Shell users...