Is it a good idea to separate cluster-critical deployments, e.g. cluster-autoscaler, load-balancer-controller, calico, traefik, prometheus etc., from applications by running them on dedicated node groups?
We use EKS, so the control plane is separated anyway. On the first sight, it seems to be a good solution because of the separation of concerns and better reliability. On the other hand it can be an overkill, since this system-critical node group might become underutilized and the considered deployments should be stateless anyway, not to forget the effort of additional configuration.
What are the best practices in this case?
As for now, we are running everything in the same node group.
