The Slurm scheduler upgrade has been completed on Expanse and the machine is available for use. Slurm was upgraded from version 21.08.8 to 23.02.6. Please note that with the upgrade of Slurm, srun default behaviour has changed. Details of release specific changes are available in:

https://github.com/SchedMD/slurm/blob/slurm-23-02-6-1/NEWS

One change in particular might impact some users as srun is no longer reading SLURM_CPUS_PER_TASK. This meas that the --cpus-per-task value set in the #SBATCH specification will not be automatically picked up by any srun command within the script. Users can either add a specific option to their srun command OR set the following variable before the srun commands:

export SRUN_CPUS_PER_TASK=${SLURM_CPUS_PER_TASK}

No changes are required if your script was using Intel MPI and mpirun. Please contact us either via the ACCESS ticketing system or via email to consult@sdsc.edu if you have any questions.

>>>>>>>>

Dear Expanse User,

We will have a maintenance period on Expanse 7AM-Midnight (PT), Dec 18, 2023. During this maintenance, we will be updating the Slurm scheduler. We have a reservation in place to prevent jobs from running during this period. The "squeue" output will show "ReqNodeNotAvail, Reserved for maintenance" for jobs that do not fit in the time period before the maintenance begins. These jobs will run after we release the maintenance reservation.

Thanks

SDSC User Support Staff