Provisioning the cluster should of course be automated. Does that mean you'll never need to ever access a node for any reason? Of course not.