Controller deletes hardware nodes from cluster

[vladimirtiukhtin]

TL;DR

I run cloud plus hardware nodes (through vswitch). Each time kubelet on hardware nodes register a node, controller deletes it from the API. I found a way around it, but controller ensures node.kubernetes.io/network-unavailable taint which is really pain in the known place

Expected behavior

Controller does not attempt to manage HW servers

Observed behavior

As described above

Minimal working example

No response

Log output

No response

Additional information

No response

[apricote]

Hey @vladimirtiukhtin,

did you try setting up hcloud-cloud-controller-manager according to the Robot Docs?

[vladimirtiukhtin]

Hi @apricote. Thanks for the prompt response. Yes, I tried, however there is a critical feature missing Cloud provider could not be initialized: could not init cloud provider "hcloud": using Routes with Robot is not supported. What would be helpful is for the controller to not attempt to manage robot servers at all. Through a label possibly

[apricote]

This is unfortunately the way CCMs work.

Is there a specific reason why you do not want to manage the Robot servers through HCCM?

[apricote]

Other question, are you using the routes features properly? If not, you could disable the routes controller and enable the Robot support instead.

To use the routes feature you need to tell your CNI to use native routing, ie routing-mode: native in Cilium.