Can we bringup on k8s cluster with initial CP node before kube-vip can start to realise VIP?

General Discussions
,
1

We are planning to use kubeadm to bring up a k8s cluster. Its a bare metal k8s cluster (well, these are VMs in vCenter). Plan to use L2 mode of kube-vip to get stable VIP for CP HA.

Can we get the first CP node to be fully configured with VIP passed as config param to “kubeadm init” with “–control-plane-endpoint” before kube-vip is deployed in the cluster.

I can see that all of the kubelet/controller manager/scheduler kubeconf file refer to local IP (or the one passed in --apiserver-advertise-address), not VIP. admin.conf and super-admin.conf refers to VIP, but we may not need to talk to api-server before cluster comes up with this one node.

Therefore, can we do the below without any issues:

  • Run kubeadm init passing VIP to control-plane-endpoint
  • Let cluster come to life
  • Now, deploy kube-vip as daemonset and let the first CP node capture the VIP

And can we do the same thing for additional CP nodes?

Thanks for your help in advance,

vchanna

2

Hello,

How do you plan to bootstrap nodes with kube-vip, static-pod or daemonSet?

The steps you mentioned, I think you should be using static-pod at least until the cluster is stable, before switching to ds only.

Run kubeadm init with:

--control-plane-endpoint = virtualIP:6443
--apiserver-advertise-address = first node real IP

Once this is done, add other CP nodes.

To join other control plan nodes, VIP must already be active and reachable before running kubeadm join --control-plane.

Let me know if this helps or if there are any specific issues or errors you see.

Thanks,
Manan

3

Thanks for the response.

The idea is to configure first CP node in the cluster by running “kubeadm init” with --control-plane-endpoint = virtualIP:6443. Once the first node is fully configured, deploy kube-vip as daemonSet.

And now, continue with kubeadm join on other nodes.

The actual question is should we have the kube-vip running as static pod on first CP node before we can fully configure first node, or we can pass yet not active control-plan-endpoint to kubeadm init?

Once the first node of the k8s is fully operational, we can deploy kube-vip as daemonSet which will make first CP node to own the VIP and then that is reachable when “kubeadm join” is run on other nodes.

Thanks,

4

Yes, have it running as a static pod. Don’t pass the not-yet-active VIP; make it active first via a static Pod.

5

Thanks for the response. Helps.