NodeClaim namespace hardcoded to kube-system

[miguelvidex]

Fail to get info about the NodeClaims

Error:

failed to list *v1beta1.NodeClaim: conversion webhook for karpenter.sh/v1, Kind=NodeClaim failed: Post "https://karpenter.kube-system.svc:8443/conversion/karpenter.sh?timeout=30s": service "karpenter" not found

Reason:

NodeClaim namespace hardcoded to kube-system, while here in the blueprints the service is running on karpenter namespace:

conversion:
    strategy: Webhook
    webhook:
      conversionReviewVersions:
        - v1beta1
        - v1
      clientConfig:
        service:
          name: karpenter
          namespace: kube-system
          port: 8443

[jakeskyaws]

Thanks for raising an issue. After reviewing, it seems this issue may be related to the one mentioned in AWS Karpenter Provider Issue #6818. Based on this, it doesn’t appear to be a blueprint issue specifically.

We recommend upgrading to Karpenter 1.0.2, as this version includes fixes that might resolve the problem you're encountering.

Additionally, if you could provide more information on how to replicate the error (e.g., specific steps, configurations, or environment details), we’d be happy to take some time to test the upgrade and confirm whether it resolves the issue.

[chrismld]

@miguelvidex are you seeing this error when using eks-node-viewer? We've seen that error when using that tool. The team has a new release to fix that problem. I'd suggest to install the v0.7.0 of that tool and try again.

Please let us know if that fixes your problem.

Thanks!!

[stemann]

This might only vaguely related, but I encountered a similar issue using the Karpenter MNG blueprint, which instructs to disable the webhook in the Karpenter Helm chart values YAML:

    webhook:
      enabled: false

Which caused issues when applying the NodePool manifest/yaml - failing to find the karpenter service with the same error as above.

Removing webhook.enabled: false (and deploying karpenter to the kube-system namespace) resolved the issue.

[chrismld]

closing it as we haven't heard back from @miguelvidex ... but feel to re-open it if needed.