Introduction to Slurm: The job scheduler

Slurm is the job scheduler we use in Unity. The following guide will go into depth about some introductory elements of Slurm. There are many features of Slurm that go beyond the scope of this guide, but everything you need to know should be available on this page. For an in-depth cheat sheet on Slurm, see the Slurm cheat sheet.

Core limits

There is currently a 1000 CPU core and 64 GPU limit to be shared by the users of each lab.

If you try to go over this limit, you are denied for MaxCpuPerAccount.

To check the resources currently in use by your PI group, use the unity-slurm-account-usage command.

Partitions or queues

Our cluster has a number of slurm partitions defined, also known as a queue. You can request to use a specific partition based on what resources your job needs. To find out which partition is best for your job, see Partitions.

Job submission overview

A job is an operation which users submit to the cluster to run under allocated resources. There are two commands for submitting jobs: salloc and sbatch.

salloc is tied to your current terminal session which allows you to interact with your job, however, once you close your terminal, the job loses its allocated resources and stops running.

sbatch on the other hand, is not tied to your current session, so you can start it and walk away, but you cannot interact with your job. If you want to interact with your job and be able to walk away, you can use tmux to make a detachable session. For more information, see Use tmux with salloc to keep a session open.

Use salloc to submit jobs

A salloc job is tied to your ssh session. If you break (ctrl+C) or close your ssh session during a salloc job, the job is killed.

Highly recommended: You can also create an interactive job, which allows your job to take input from your keyboard. You can run bash in an interactive job to resume your work on a compute node just as you would on a login node.

See SALLOC Jobs for more information.

Use sbatch to submit jobs

An sbatch job is submitted to the cluster with no information returned to the user other than a Job ID. An sbatch job will try to create a file in your current working directory that contains the results of your job.

See Introduction to batch jobs for more information.

Use tmux with salloc to keep a session open

To keep a session open even if your ssh command disconnects, use tmux. This can be useful on spotty Wi-Fi so you don’t lose your work.

The following is an example of how to use tmux and salloc to keep a session open even if your ssh command disconnects:

# Open tmux session:

tmux

# Open an interactive job on compute node with one cpu core:

salloc -c 1

# Make the interactive job  have a blinking cursor for an hour:

sleep 3600; echo "done"

# Open tmux keyboard-shortcut command mode:

# > ctrl+b

# Detach tmux session and go back to login node:

# > d

# At this point, you can log off and log back in without killing the job.

# Print a list of tmux sessions:

tmux ls

# The first number on the left (let's call it X) is needed to re-attach the session:

tmux attach-session -t X

# This brings us back to the interactive job

Other resources

For an in-depth cheat sheet on Slurm, see the Slurm cheat sheet.