Comment by figassis
I have yet to see a guide to automate k8s on Hetzner's beefy bare metal instances. True, you want cattle, but being able to include some bare metal instance with amazing CPUs and memory would be great, and I do just that. My clusters include both cloud and bare metal instances. In the past I had used Hetzner virtual switch to create a shared L2 network between cloud and bare metal nodes. Now I just use tailscale.
But the TF and other tools are using the API to add and kill nodes, if you could pass a class of nodes to those tools that they know can't create but are able to wipe and rebuild, this would be ideal.
Would CAPH solve your issue? https://github.com/syself/cluster-api-provider-hetzner you would wire up cluster autoscaler to point to it using clusterapi https://github.com/kubernetes/autoscaler/tree/a9cb59fdd2dd88...