Refactor node-local control plane loadbalancer
#12.631 geöffnet am 16. Okt. 2025
Repository-Metriken
- Stars
- (10.380 Stars)
- PR-Merge-Metriken
- (Durchschn. Merge 25T) (26 gemergte PRs in 30 T)
Beschreibung
What would you like to be added
We currently have 3 implementations for node-local LB to the control plane (HA endpoint for kubelet to talks to the control plane).
- nginx
- haproxy
- kube-vip
I'm not sure if we need that much choice and what this actually brings, besides more complexity.
I propose to keep only one option, and furthemore investigate if we have to use static pods or can use a daemonset (probably not, I think kubeadm needs access to the control plane during node join). Ideally, the chosen implementation should have a superset of the features of the other.
The relevant code and template are in roles/kubernetes/node/{tasks/main.yml,templates/}.
If the different options do have non-overlapping interesting features, please speak up.
Why is this needed
Reduce maintenance and complexity.
/help /good-first-issue