Skip to main content Link Menu Expand (external link) Document Search Copy Copied
Arjuna

Submitting Your First Job

This tutorial will guide you through the process of submitting your first job on Arjuna. This tutorial assumes you’ve already done the following items

  1. Login to Arjuna via ssh
  2. Some familiarity with entering commands at a bash shell
  3. Some familiarity with vim or another command-line file editor

Documentation for the above tasks can be found under Getting Started

Hello World

For this example, create a file called ~/hello_world.sh with the following content:

#!/bin/bash
#SBATCH -n 1
#SBATCH --mem=2G
echo "Hello World!"

What do the lines mean?

1) #!/bin/bash This is a shebang it tells Linux how to handle the file. Here’s we’re saying use bash to run this file 2) #SBATCH -n 1 This is a SLURM directive asking for 1 task 3) #SBATCH --mem=2G Now we’re asking for 2 gigabytes of memory. Notice the suffix G, the default suffix is M, but we can also use K or T 4) echo "Hello World!" This is the command that gets run on the compute node

For more SLURM directives, see man sbatch or the documentation for sbatch.

Submitting the Job

To submit the job, run the following command: sbatch ~/hello_world.sh.

> sbatch ~/test.sh
Submitted batch job 2970359

Check the job’s status in the queue using: squeue -u $(whoami).

> squeue -u $(whoami)
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
           2970359       gpu hello_wo  awadell PD       0:00      1 (None)

The -u $(whoami) flag cause squeue to only show jobs for the current user. Omitting this will show the entire queue.

The squeue command displays the current state of SLURM’s queue and by default the following columns:

Name Description
JOBID The numerical id SLURM uses to identify a job
PARTITION The partition the job was submitted to
USER Who submitted the job
ST The state of the job, see the docs for a list of abbreviations
TIME How long the job has been running for
Nodes How many nodes the job is running on
NODELIST(REASON) Why the job isn’t running or a list of the nodes it’s running on

For more information on squeue checkout man squeue or it’s documentation.

Getting Results

Once the job has completed (It will disappear from the output of squeue) we should be left with a slurm-<jobid>.out file (Replace <jobid> with your JOBID). Displaying this file should give the following:

> cat slurm-2970359.log
Hello World!

By default SLURM will create this file, but you can change it’s name via #SBATCH -output directives or sbatch’s command line flags --output. See sbatch for more information.

This file will contain the standard output of the script ~/hello_world.sh. Job scripts are not limited to just this file. Output can be saved to other files like this:

#!/bin/bash
#SBATCH -n 1
#SBATCH --mem=2G
echo "Hello World!" > ~/another_file.txt

This will instead write “Hello World!” to ~/another_file.txt instead of slurm-<jobid>.out. Slurm will still create a slurm-<jobid>.out file, but now it will now be empty.

The > tells bash we what to redirect the output of echo to ~/another_file.txt. Check out Redirection for more information

What’s Next?