VIP/Load balancer for worker nodes for Ingress

[violuke]

We're really excited to start building our new on-prem Kubernetes cluster using Talos (I've previously used RKE).

I see that a Virtual IP can be set up and used to ensure reliable access to the control plane, but is there a similar feature for worker nodes (to use as the ingress IP)? If not, is the recommended approach a load balancer for all worker nodes (e.g. HAProxy)?

If so, for an HA setup, would you recommend two HAProxy nodes that also have a VIP? And is there any advice on how HAProxy can fetch dynamically the worker nodes to save hard-coding the possible IPs?

Edit: I've also seen kube-vip. Would this, in conjunction with controller.service.type: LoadBalancer (assuming ingress-nginx) be a good solution (which avoids extra hardware in front of the Talos cluster)?

Or is there a better/alternative approach altogether?

Thanks for your help 😄

[smira]

I don't think this question is specific to Talos Linux, but in general any solution would work, e.g. I know people successfully use MetalLB.

[violuke]

Ok, thank you. Appreciate the advice 🙏

[jfmatth]

I just finished building my home cluster, was using metallb and nginx, moving to using Cilium CNI as the Talos docs point to.

I'm also using the Controlplane VIP, man that is soooo cool.

Another cool feature with Talos, since the OS and K8s are so connected, doing maintenance on the node for Talos, puts the k8s in non-scheduling mode too, very coo.