Karpenter Node drift towards smaller instances with Version 1.14

Author: hvoigtCreated Aug 21, 2026Updated Sep 14, 2026

Labelstriage/needs-information

Description

Observed Behavior:

After upgrading from Karpenter 1.13 to Karpenter 1.14 the cluster drifted towards smaller instances. Since we have a fixed daemonset configuration this lead to higher costs. We were able to mitigate this behaviour by excluding the smaller instance sizes for the NodePool. In our case the large and xlarge types on AWS. After applying the mitigation the cluster costs/usage returned to pre Karpenter 1.14 values.

Expected Behavior:

Selection of nodes does not prefer less efficient size of node.

Reproduction Steps (Please include YAML):

I am not sure how to reproduce this. Here is the (cleaned) NodePool configuration. This config includes the mitigation. Usually it would have large and xlarge as karpenter.k8s.aws/instance-size:

yaml

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
  name: default
spec:
  disruption:
    budgets:
    - duration: 1m0s
      nodes: "1"
      schedule: 0 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 1 * * * *
    - duration: 1m0s
      nodes: "1"
      schedule: 10 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 11 * * * *
    - duration: 1m0s
      nodes: "1"
      schedule: 20 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 21 * * * *
    - duration: 1m0s
      nodes: "1"
      schedule: 30 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 31 * * * *
    - duration: 1m0s
      nodes: "1"
      schedule: 40 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 41 * * * *
    - duration: 1m0s
      nodes: "1"
      schedule: 50 * * * *
    - duration: 9m0s
      nodes: "0"
      schedule: 51 * * * *
    consolidateAfter: 1m
    consolidationPolicy: WhenEmptyOrUnderutilized
  limits:
    cpu: 6000
    memory: 12000Gi
  template:
    spec:
      expireAfter: 720h
      nodeClassRef:
        group: karpenter.k8s.aws
        kind: EC2NodeClass
        name: default
      requirements:
      - key: kubernetes.io/arch
        operator: In
        values:
        - amd64
      - key: kubernetes.io/os
        operator: In
        values:
        - linux
      - key: karpenter.sh/capacity-type
        operator: In
        values:
        - on-demand
      - key: karpenter.k8s.aws/instance-category
        operator: In
        values:
        - c
        - m
        - r
      - key: karpenter.k8s.aws/instance-generation
        operator: Gt
        values:
        - "5"
      - key: karpenter.k8s.aws/instance-size
        operator: In
        values:
        - 2xlarge
        - 4xlarge
      terminationGracePeriod: 1h

We saw this on a cluster with around 800 cores. Node count was increased by 60% and through the higher ratio of daemonset resources the total core count was increased by a value about 20%.

The cluster (with larger nodes, thus less daemonsets) is running ~1600 pods.

Versions:

Chart Version: 1.14.0
Kubernetes Version (kubectl version): v1.35.4

Let me know if anything is missing or additional information that would be helpful to diagnose this, thanks

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

Source: aws/karpenter-provider-aws

View original on GitHub View discussion on GitHub