wait until all processes have finished before exiting

[stdweird]

looks like mpirun in openmpi sometimes finsihes while there is still a srun process running.
so mympirun is done, while the step is not finished.

this can cause problems for the next step.

simply settings export SLURM_OVERLAP=1 is way too dangerous, but might be a good workaround if you know it is safe to do (eg the step terminates in a few seconds)