From the Marconi HPC User Support:
“srun will no longer read in SLURM_CPUS_PER_TASK. This means you will
explicitly have to specify --cpus-per-task on your srun calls, or set the new
SRUN_CPUS_PER_TASK env var to accomplish the same thing.”
Such modification affects hybrid (MPI/OpenMP) jobs launched with srun, since
the consequent mapping of the multithreaded tasks will result in an
overallocation of the cpus. As reported in the slurm Highlights, you can
- define the SRUN_CPUS_PER_TASK env var in your scripts:
export SRUN_CPUS_PER_TASK=$SLURM_CPUS_PER_TASK
srun your_exe
- or your can repeat the --cpus-per-task directive as option of srun:
srun --cpus-per-task=$SLURM_CPUS_PER_TASK your_exe
This will likely break the cpus option for the Slurm-Runner.
From the Marconi HPC User Support:
This will likely break the
cpusoption for the Slurm-Runner.