The extended-exception QOS

Unity jobs are normally limited to a 5-day max wall time, a limit that was reduced from 14 days in October 2026 to support a weekly kernel update schedule. The extended-exception QOS raises that limit back to 14 days for PI groups that are approved by the Unity team.

A QOS is a property of the job rather than a pool of nodes, so the extended-exception QOS changes how long a job may run, not the hardware it gets. It is currently available only in the cpu partition. Jobs in all other partitions remain subject to the 5-day limit.

stylus_note
Requires approval
While all groups are eligible to apply, the extended-exception QOS is not available to everyone by default. Access is granted to PI groups by request; submit the extended-exception request form and wait for it to be approved before submitting a job with the QOS. Jobs submitted without approval are rejected by Slurm.
stylus_note
Limits
The QOS raises the wall time limit to 14 days, but is further restricted to 10 simultaneous jobs per user. Jobs running with this QOS can also be ended at any time, without prior warning, if the Unity team deems it necessary for cluster administration. All other limits still apply, including the per-PI-group core and GPU quotas of the partition the job is submitted to (see Core limits).

How to use the extended-exception QOS

Submit the request form to request access, then add #SBATCH --qos=extended-exception to your job script, with --time set to your expected run time of up to 14 days:

#!/bin/bash
#SBATCH --job-name=my-long-job
#SBATCH --partition=cpu
#SBATCH --qos=extended-exception
#SBATCH --cpus-per-task=8
#SBATCH --mem=32g
#SBATCH --time=14-0:00:00

./my_program
  • --time uses Days-Hours:Minutes:Seconds format; the largest value an approved job may request is 14-0:00:00.
  • The QOS is currently available only in the cpu partition. Set --partition=cpu (or -p cpu).

See the Slurm cheat sheet for a quick reference of job submission options.

Should I use this QOS?

report
Prefer checkpointing
The 5-day limit exists so the cluster can apply kernel updates on a weekly schedule, and a 14-day job runs across at least one of those windows. Before requesting an exception, consider whether your job can be reshaped to run in under 5 days, for example by checkpointing and restarting. If you need help developing a checkpointing strategy, contact the Unity team.

The request form asks you to confirm that you have exhausted strategies to fit your workload within the 5-day limit, such as checkpointing or scaling your requested resources, and that your workload cannot gracefully restart from a checkpoint and requires more than 5 days to execute. If both hold, the extended-exception QOS is the right tool. For the full list of partitions and their limits, see the Partition List, and the Node List for the specific nodes that belong to each partition.