Created node never joins the cluster

[fdtmsft]

Following the example for self-hosted I have created my nodepool with Ubuntu 22.04 and I do see Karpenter (0.7,2) creates a node when I scale my pause deployment, but the new node that I see running on Azure never joins the cluster.

Doing some research here I have tried to update using az aks update -n cluster_name -g cluster_resource_group which successfully completed but yet still no node appears.

I have also checked the karpenter-values .yaml file contains the correct bootstrap tokens of the cluster. Any suggestions of what could be the issue?

[MattMencel]

@fdtmsft The az aks update hack only works for NAP.

I switched back to NAP from self-hosted pretty quickly. The bootstrap tokens reset every few days and there is no easy(automated) way to keep them aligned in the self-hosted configuration. So the scaling just stops until a human goes in and updates the bootstrap token in the config.

So IMO.... just use NAP. At least with NAP, if the token expires you can fix it easily and put some automation around the az aks update command. Also for this reason alone... let alone that this is a preview feature, even with NAP, I would not recommend karpenter for production workloads.

Maybe in the future this will change, but for now I don't think self-hosted is an option unless you just want to see the components deployed in their native state.

[Bryce-Soghigian]

You shouldn't need to do that at all with nap now. That was just initially. I agree that without some automated token rotation in place managing the helm values for the bootstrapping token are a major pain for self hosted making it non-viable in production environments.

CC: @charliedmcb