For some reason it runs massively slower in k8s, we should at least try to run stuff in parallel where we can.