Is there any world where it makes sense to set the cluster service IP range as global IPv6 addresses?
It _sortof_ works with a range in my router's delegated prefix but I'm suspecting Cilium's native routing option is treating it as North-South traffic since it's inside the native routing range.
Which I think means it's YOLOing CoreDNS's East-West attempts to reach the default Kubernetes service across the host interface (and to the gateway, which has no idea what to do with it).
Relatedly, I think I can put to bed my idea of service implementation being DNS.