v1: how to expire nodes only during certain times?

[mikkoc]

In 0.37 this config allows us to expire nodes only during weekdays/working hours:

 disruption:
  consolidationPolicy: WhenUnderutilized
  expireAfter: 48h
  # Number of nodes: disrupt 1 node at a time, only during weekdays, from 10 to 18 UTC.
  # Schedule: zero disruptions (0) are allowed starting at 8am UTC on Friday, lasting 3 days + 2hrs.
  budgets:
  - nodes: "1"
   schedule: "0 10 * * mon-thu"
   duration: 8h
  - nodes: "0"
   schedule: "0 8 * * fri"
   duration: 74h

After migrating to v1 the config looks more like this:

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
 name: generic
spec:
 template:
  spec:
   nodeClassRef:
    apiVersion: karpenter.k8s.aws/v1
    kind: EC2NodeClass
    name: generic
   expireAfter: 48h   ### I don't want this to take effect during nights and weekends.

 disruption:
  consolidationPolicy: WhenEmptyOrUnderutilized
  consolidateAfter: 30m
  budgets:
  - nodes: "1"
   schedule: "0 10 * * mon-thu"
   duration: 8h
  - nodes: "0"
   schedule: "0 8 * * fri"
   duration: 74h

And nodes are expired also during weekends. How can I keep the behaviour of karpenter 0.37 in v1? From the documentation, it is unclear to me how to achieve this.

Thanks

[jmdeal]

This isn't currently possible to achieve with expiration on v1. Expiration was changed from a graceful to a forceful expiration method at v1 (RFC: kubernetes-sigs/karpenter#1303) which means it can no longer be restricted by disruption budgets. This enabled enforcing strict maximum node lifetimes through expiration and termination grace period. Graceful disruption methods are still subject to disruption budgets (i.e. drift and consolidation). I'm wondering if your goals can be better met by drift, what are you trying to achieve by rolling nodes roughly every 48 hours?

[mikkoc]

From the Karpenter 0.37 NodePool example config: https://karpenter.sh/v0.37/concepts/nodepools/

# Avoiding long-running Nodes helps to reduce security vulnerabilities as well as to reduce the chance of issues that can plague Nodes with long uptimes such as file fragmentation or memory leaks from system processes

Since I ever started working with Kubernetes I was always taught to avoid long-running nodes. So even after all these years I try to keep my nodes "fresh" by rotating them.

If you have an alternative mechanism to achieve this with Karpenter v1, please share. I would like to avoid having my production nodes reaching 2 weeks of life (that's how often we deploy). But at the same time, I'd like to retain more control over when nodes rotate.