Skip to content

A basic job

Before you start

To run a job you will need:

  • Research data on the cluster
  • An application to run
  • Your login credentials
  • A text editor — nano is a good choice if you are new to the command line

Make sure you are logged in before continuing. If you are new to SLURM, the SLURM documentation is a useful reference.

The job script and sbatch

You submit a program to SLURM for execution using the sbatch command. The easiest way to do this is with a job script (also called a job description file).

A minimal job script looks like this:

#!/bin/bash
#SBATCH -t 00:05:00

echo "hello"

Write this to a file — in the examples below it is named echo_script.sh, but any name will do. Submit it with sbatch:

sbatch echo_script.sh

You should see output similar to:

[fred@cosmos ~]$ sbatch echo_script.sh
Submitted batch job 7185

SLURM assigns the job a number (7185 here). Once the job has run, you will find a file named slurm-7185.out in your working directory containing the program's output and any error messages.

The three parts of a job script

The echo_script.sh example shows the three sections every job script must have:

  1. Shell specification — the #!/bin/bash line tells the system which shell to use
  2. Resource statement(s) — lines starting with #SBATCH declare what resources the job needs; here -t 00:05:00 requests 5 minutes of compute time. If the job has not finished by then, SLURM will terminate it.
  3. Body — the UNIX commands to run; a job script is a normal shell script, so anything valid in bash is valid here

Real job scripts typically include several #SBATCH lines to specify things like the number of cores, memory, and a job name. These are covered in Specifying job requirements.

Author: (LUNARC)

Last Updated: 2022-10-05