Add built-in support for kube-vip

[cl-rf]

I would like for rke2 to support kube-vip natively during provisioning. There are ways to do this manually using the following gist:

https://gist.github.com/bgulla/7a6a72bdc5df6febb1e22dbc32f0ca4f

Ideally, there would be a an option in the config.yaml like "bind-vip-address=" where if this was set, the helm chart would be installed and configured. This would allow setting up a HA cluster without needing to setup external hardware/software load balancers and would just be highly available out of the box.

[brandond]

You can already enable k3s's ServiceLB if you need a basic load-balancer controller. I know that's not quite the same as what you'd get from kube-vip, but would it meet your needs?

[brandond]

Just to be clear, are you asking to have kube-vip or another LoadBalancer controller put in front of the RKE2 supervisor and apiserver to host a load-balanced virtual IP? Or do you just want a load-balancer controller available in your cluster?

[cl-rf]

I am asking for kube-vip or something similar to provide a built-in load balancer for the control nodes that does not require external resources. FIPS and STIG are the main factors for using rke2 vs k3s.

[brandond]

provide a built-in load balancer for the control nodes

Does this need to be a floating or load-balanced virtual IP, or are you referring to a Kubernetes LoadBalancer controller?

[cl-rf]

Floating IP. It is more for HA than balancing the traffic between the control nodes.

[snivou]

This can also solve the problem of downstream rke2 load balancing

rancher/rancher#39021 (comment)

[flrichar]

The use-case for kube-vip is somewhat limited to onpremises (ie non-cloud native) environments. I use it myself and like it, however there's no recent helm chart, each manifest is unique and dependent on the environment. I can see this as a hurdle to adoption.

[danieljkemp]

I've been running a manually deployed kube-vip setup to provide a VIP for the apiserver via static pods (was using round-robin dns previously). Most (every) times the control plane server restarts the kube-vip pod isn't started properly, and a crictl on the node shows it as exited. Deleteing the dead pod and restarting rke2 fixes things.