There was a race-condition, that caused crashes.
We could use gproc, but it feels a bit overkill for ~10-20 processes