SLURM

Commands

• Use the sbatch command to submit a job script

• Check job status: squeue

• Cancel a job: scancel <job_id>

• View detailed job info: scontrol show job <job_id>

• Start an interactive session: srun --pty bash

• parallel jobs:

  • srun with specifying #SBATCH --ntasks-per-node=<int>

Arguments

• Specify CPU cores: --ntasks=<number> or --cpus-per-task=<number>
• Request memory: --mem=<amount>[K|M|G|T]
• Set time limit: --time=<days>-<hours>:<minutes>:<seconds>
• Specify custom output file: --output=<filename>
  • • By default, SLURM creates output files named slurm-<job_id>.out
• Run a job after another completes: --dependency=afterok:<job_id>