This is almost exactly the issue I’m having too: K8s runners with different job resource requirements