More info for @f.uhlig: in recent weeks the weekly slurm jobs were almost always killed during the filtered rsync process from /tmp to lustre and even a few times earlier (once even before submitting to CDASH). I hope that by making the reserved time at the end 3x larger it will improve.