Interruptable Partitions and Jobs

We have configured Slurm to allow users to access additional resources that they normally do not have access to. A Slurm "interruptible queue" isn't a specific queue type but a configuration using Quality of Service (QoS) and partitions, allowing for low-priority jobs that can be paused or canceled (preempted) by higher-priority job. This allows users to access unused hardware, but they must be aware that their jobs could be cancelled by higher priority jobs.

Your jobs can be interrupted

Any job that has higher priority will cause your job to be cancelled and requeued. This may not be suitable for your workflow, or for long jobs. So smaller jobs would be preferrable using this method.

How to submit interruptable jobs

The following partitions allow job submissions with the irq QOS value even if users do not normally have access to that partition.

- genomics
- genomicsb
- fit
- fitc

To access these partitions, please specify the partition you want, with the irq QOS.

For example to access the following partitions for CPU only access.

#SBATCH --partition=genomics,genomicsb,fitc
#SBATCH --qos=irq

To access additional GPU resources you could go:

#SBATCH --partition=fit
#SBATCH --qos=irq
#SBATCH --gres=gpu:1

We recommend that you use show_cluster to check the business of the partitions that you would like to use.