cc @starbops @w13915984028

Hi @adonskoy, thank you for reporting this issue. I'm curious, could you share the reason why you needed to disable kube-vip in the harvester-cloud-provider?

Hi @brandboat, we encountered the following problems:

• Stability of integration with kube-vip. In cases where the etcd leader changes in the guest cluster, kube-apiserver is restarted, then harvester-cloud-provider restarted, and for some reason it restarts kube-vip too. After which traffic to the target service stops coming until the pods of this service are recreated

• Inability to schedule kube-vip to the worker node - changed selectors and affinity rules are ignored (also probably a problem in the chart, but I have not yet looked at what exactly the problem is). I would like to use externalTrafficPolicy: Local in services, but with kube-vip I will have to schedule the workload to the master nodes for this

At the moment we have deployed metallb, removed address pools for the built-in balancer from the harvester and everything works, but kube-vip still performs unnecessary actions, trying to reconcile these services

Also, it seems that metallb's performance is slightly higher in this setup than kube-vip's. But this is a subjective opinion, I haven't done any measurements yet.

Back to the issue itself, the enabled option is not handled in the embedded kube-vip chart, we will add a fix.

https://github.com/harvester/charts/blob/4cc0524d6cbf231cecc174baa388829ed66b1796/charts/harvester-cloud-provider/values.yaml#L74

@adonskoy Thanks for reporting this issue.

@w13915984028
@adonskoy has a PR: harvester/charts#279

Stability of integration with kube-vip. In cases where the etcd leader changes in the guest cluster, kube-apiserver is restarted, then harvester-cloud-provider restarted, and for some reason it restarts kube-vip too. After which traffic to the target service stops coming until the pods of this service are recreated

Hi @adonskoy Regarding this one, do you have more information? We'd like to check if it's reproducible.

• Rancher version
• Guest cluster version
• VM OS

Thank you.

Need a condition in Chart.yaml to get it enabled/disabled.

https://helm.sh/docs/topics/charts/#tags-and-condition-fields-in-dependencies

https://github.com/harvester/charts/blob/4cc0524d6cbf231cecc174baa388829ed66b1796/charts/harvester-cloud-provider/Chart.yaml#L47

Hi @adonskoy Regarding this one, do you have more information? We'd like to check if it's reproducible.

• Rancher version
• Guest cluster version
• VM OS

Thank you.

Hi @starbops, sure

• Rancher version: 2.8.5-2.9.1
• Guest cluster version: rke2 from 1.28 to 1.31
• VM OS: Ubuntu 22.04 and 24.04

We had this problem on all these versions