OpenELB in VIP Mode in k8s cluster causes node-ip confusion

[bdowling]

Describe the bug

When using OpenELB in VIP mode, this adds virtual addresses to the node's interface. If kubelet is not configured with a static --node-ip, it will randomly pick a local IP for the node's InternalIP.

While this is not really a bug in OpenELB, it is an issue that I feel should be documented in your instructions as I'm sure others will come across this issue.

THE FIX

Adding --node-ip [nodes-primary-ip] to respective kubelet systemd startup script for your system, e.g. /etc/systemd/system/kubelet.service does the trick to force kubelet to a fixed node ip.

To Reproduce

Install OpenELB and configure it for VIP mode per instructions:
https://openelb.io/docs/getting-started/usage/use-openelb-in-vip-mode/

Deploy a LoadBalanced service.

run kubectl get nodes repeatedly and you will see Internal IPs randomly match your service IPs.

Or you could run this and grep for IPs in your vip-pool
kubectl get nodes "$@" -o jsonpath='{.items[*].status.addresses[?(@.type=="InternalIP")].address}' | sed -e 's/ /\n/g'; echo

Expected behaviour

NA - now that I know what happened, it was expected behavior :)

Output
NA
Version Info

• Version of Kubernetes: 1.28.4
• Version of OpenELB: 0.5.0

[renyunkang]

Thanks for pointing it out, we will add documentation later.