Comment by flowerthoughts

My biggest issue with k8s on Hetzner is that there is no way of going from tiny (1 machine) to medium (10-100 machines) purely on bare metal. I was able to get Gateway API to provide ingress nodes after a bit of Envoy tinkering, but storage is an issue. You can't run Ceph Rook on a single machine (too heavy,) and I couldn't find a think local volume wrapper that lets me easily migrate from local volume management to distributed.

Feels like there should be a PV daemon that can do local, and transparent migrations and is lightweight enough to run on a single machine. Once my PV has been migrated to Ceph, the proxy could configure itself away from that job.