Use NodeRepair featureGate

[fe80]

Description

How can the docs be improved?

Hello,

I try to use the featureGate NodeRepair, this is correctly enable on my controller pod (FEATURE_GATES: SpotToSpotConsolidation=false,NodeRepair=true), but that doesn't work like I've expected.

We just need to enable the feature gate for work ? I pretty sure we need to wait 30min before the node was considerate as unready but I don't find any documentation on this point.

Also it's look on node object or nodeclaims.karpenter.sh ? Because my node is ready but not my node claim:

╰─➤ k get node ip-10-34-6-212.eu-west-3.compute.internal
NAME                                        STATUS   ROLES    AGE    VERSION
ip-10-34-6-212.eu-west-3.compute.internal   Ready    <none>   129m   v1.30.6-eks-94953ac

╰─➤ k get nodeclaims.karpenter.sh std-linux-cpu-h4g4s                       
NAME                  TYPE          CAPACITY   ZONE         NODE                                        READY     AGE
std-linux-cpu-h4g4s   m7i.8xlarge   spot       eu-west-3a   ip-10-34-6-212.eu-west-3.compute.internal   Unknown   129m

Regards,

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• Please do not leave "+1" or "me too" comments, they generate extra noise for issue followers and do not help prioritize the request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment

[engedaam]

The node repair feature works by looking at the node readiness. Karpenter only act against nodes that are not ready for 30 min. The nodeclaim being unknown is a known bug that are planning on pushing to fix. The issue there is the status of how karpenter is marking the status rather then any issue with the nodeclaim #7494

[fe80]

Hello,

With 1.1.1 I've still have Unknown if my node still have the statupTaint

╰─➤ k get nodeclaims.karpenter.sh std-linux-core-dpxmc                      
NAME                   TYPE         CAPACITY   ZONE         NODE                                          READY     AGE
std-linux-core-dpxmc   t3a.medium   spot       eu-west-3b   ip-10-157-24-133.eu-west-3.compute.internal   Unknown   51m

╰─➤ k get node ip-10-157-24-133.eu-west-3.compute.internal
NAME                                          STATUS   ROLES    AGE   VERSION
ip-10-157-24-133.eu-west-3.compute.internal   Ready    <none>   50m   v1.30.7-eks-59bf375

╰─➤ k get nodes ip-10-157-24-133.eu-west-3.compute.internal -o json | jq '.spec.taints | length'      
1

The node repair feature works by looking at the node readiness

So they have no possibility to automatically destroy the nodeclaim if it's not at True ?