Runs an executable file. It is really intended to be used from within sbatch scripts. Used within a script, each invocation of srun starts a new job step (terminology you will see in the slurm documentation).

Used from a terminal, srun runs the job in the foreground (and blocks unless you use “&”).

If you use srun directly (in a login terminal) your job will end if you close your terminal, unless you redirect IO and disown the job. You can also use a “terminal multiplexer” like screen or byobu. However, even if you take steps to protect against “hang-ups”, if the head node needs to be rebooted (to pick up kernel updates, for instance) that will terminate any jobs running through srun. A reboot of the head node will not affect jobs submitted using sbatch.

srun hostname

srun -n3 hostname (runs 3 tasks: copies of the program specified, may run on the same node)

srun -n30 hostname (runs 30 tasks, will run on multiple nodes)

srun -N3 hostname (runs on 3 different nodes)

srun –pty bash -i

• Runs an interactive shell on a compute node.
• You can specify a specific node with -w node5 (for instance).
  • Will only run if there is a slot available on the node.

srun long-job > srun.out 2>&1 &
disown

But it's much better to use sbatch.