K0s control plane High Availability

hey there,

Been working on bringing “real” HA to k0s on my rpi4 cluster, but it seems glitchy at some point.
what would be the easy way to achieve that ?
kube-vip on the CP nodes ? DNS RR ? anything else ?

What kind of setup do you have, how many controller and workers?

And what do you want to achieve in regards to HA?

We’ve got NLLB feature which can tackle HA from cluster components point of view.

To have HA for the Kubernetes API, you need some external solution. And that depends highly on the needs you have.

VIPs in general do not work too well with k0s, due to the fact how konnectivity does HA setup. The agent needs to be able to connect to ALL controllers and with VIP you’ll hit only single controller at a time. So it does need some sort of LB.

Haven’t really tried it but having a VIP for the API and using NLLB for the cluster components could work out.

hey there,

3 Control Plane nodes, 5 Data Plane nodes

I’d like to find the best implementation to have a single endpoint that would connect me to any of the 3 control planes nodes.

i’ve tried to have kube-vip deployed at bootstrap time but it didn’t work well.

for the Data Plane nodes, kube-vip works pretty well so far, don’t know how NLLB could simply the setup