How does Karpenter handle node unresponsiveness in cases like OOMs?

[The9thLime]

Hey. I have a nodepool spec like this

spec:
  disruption:
    budgets:
      - nodes: 10%
      - nodes: 0%
        reasons:
          - Drifted
    consolidateAfter: 30m
    consolidationPolicy: WhenEmptyOrUnderutilized
  limits:
    cpu: '60'
  template:
    spec:
      expireAfter: 720h
      nodeClassRef:
        group: karpenter.k8s.aws
        kind: EC2NodeClass
        name: mimir
      requirements:
        - key: topology.kubernetes.io/zone
          operator: In
          values:
            - us-east-1b
        - key: karpenter.sh/capacity-type
          operator: In
          values:
            - on-demand
        - key: node.kubernetes.io/instance-type
          operator: In
          values:
            - r6g.xlarge
            - r6a.xlarge
            - r5a.xlarge
      taints:
        - effect: NoSchedule
          key: kind
          value: mimir-ingesters

Once, all the three nodes in this NodePool became unready and unresponsive because of OOM. They stopped responding and the pods on the node got stuck in a terminating stage. Karpenter was not able to terminate those nodes (and the NodeClaims) and was not unable to provision new replacements. On the AWS console, I could see that those instances were in the terminating stage.

I was wondering how Karpenter handles these node-level OOMs since when something like this happens, the nodes become unresponsive and get into a zombie like state.