Karpenter Node Autoscaling

Karpenter replaces Cluster Autoscaler for workload nodes. It provisions right-sized EC2 instances directly via the AWS API in under 60 seconds, supports bin-packing, handles Spot interruptions gracefully, and consolidates underutilized nodes automatically.

Karpenter vs Cluster Autoscaler

Feature	Karpenter	Cluster Autoscaler
Provisioning speed	~30–60s	2–5 minutes
Instance selection	Any instance type, dynamic	Fixed ASG instance type
Bin-packing	Yes — right-sizes per pending pod	No — fixed instance type
Spot interruption handling	Native, graceful drain	Manual setup required
Node consolidation	Automatic (WhenUnderutilized)	Limited
Multi-architecture	Yes (arm64/amd64)	Via separate ASGs

NodePool Configuration

kubernetes/karpenter/nodepool-default.yaml

apiVersion: karpenter.sh/v1beta1
kind: NodePool
metadata:
  name: default
spec:
  template:
    metadata:
      labels:
        role: workloads
    spec:
      nodeClassRef:
        name: default
      requirements:
        # Prefer On-Demand, fall back to Spot
        - key: karpenter.sh/capacity-type
          operator: In
          values: ["on-demand", "spot"]
        # Allow a broad set of instance families for flexibility
        - key: node.kubernetes.io/instance-type
          operator: In
          values:
            - m6i.xlarge
            - m6i.2xlarge
            - m6i.4xlarge
            - m6a.xlarge
            - m6a.2xlarge
            - m6a.4xlarge
            - m7i.xlarge
            - m7i.2xlarge
            - c6i.2xlarge
            - c6i.4xlarge
        - key: topology.kubernetes.io/zone
          operator: In
          values: ["us-east-1a", "us-east-1b", "us-east-1c"]
        - key: kubernetes.io/arch
          operator: In
          values: ["amd64"]
  limits:
    cpu: "200"
    memory: 400Gi
  disruption:
    consolidationPolicy: WhenUnderutilized
    consolidateAfter: 30s
    # Rotate nodes every 30 days (security best practice)
    expireAfter: 720h

EC2NodeClass Configuration

kubernetes/karpenter/nodeclass-default.yaml

apiVersion: karpenter.k8s.aws/v1beta1
kind: EC2NodeClass
metadata:
  name: default
spec:
  amiFamily: AL2
  role: "KarpenterNodeRole-devopsgenie-production"

  subnetSelectorTerms:
    - tags:
        karpenter.sh/discovery: devopsgenie-production

  securityGroupSelectorTerms:
    - tags:
        karpenter.sh/discovery: devopsgenie-production

  blockDeviceMappings:
    - deviceName: /dev/xvda
      ebs:
        volumeSize: 100Gi
        volumeType: gp3
        iops: 3000
        throughput: 125
        encrypted: true
        deleteOnTermination: true

  # User data runs after node initialization
  userData: |
    #!/bin/bash
    /etc/eks/bootstrap.sh devopsgenie-production \
      --container-runtime containerd \
      --kubelet-extra-args '--max-pods=110'

  tags:
    Environment: production
    ManagedBy: karpenter

Spot Interruption Handling

Karpenter handles Spot interruptions automatically using SQS + EventBridge:

terraform/environments/production/karpenter-spot.tf

resource "aws_sqs_queue" "karpenter_interruption" {
  name                      = "karpenter-interruption-devopsgenie-production"
  message_retention_seconds = 300
  sqs_managed_sse_enabled   = true
}

resource "aws_cloudwatch_event_rule" "spot_interruption" {
  name        = "karpenter-spot-interruption"
  description = "Spot instance interruption notification"
  event_pattern = jsonencode({
    source      = ["aws.ec2"]
    detail-type = ["EC2 Spot Instance Interruption Warning"]
  })
}

resource "aws_cloudwatch_event_target" "spot_interruption" {
  rule = aws_cloudwatch_event_rule.spot_interruption.name
  arn  = aws_sqs_queue.karpenter_interruption.arn
}

Verifying Karpenter

# Watch Karpenter provision a node
kubectl apply -f - <<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: inflate
spec:
  replicas: 10
  selector:
    matchLabels:
      app: inflate
  template:
    metadata:
      labels:
        app: inflate
    spec:
      containers:
        - name: inflate
          image: public.ecr.aws/eks-distro/kubernetes/pause:3.7
          resources:
            requests:
              cpu: "1"
              memory: 1Gi
EOF

# Watch nodes being provisioned
kubectl get nodes --watch

# Check Karpenter logs
kubectl logs -n karpenter -l app.kubernetes.io/name=karpenter -f --tail=50

# Clean up
kubectl delete deployment inflate

Useful Karpenter Commands

# List NodePools
kubectl get nodepools

# List Karpenter-managed nodes
kubectl get nodes -l karpenter.sh/nodepool

# Manually trigger consolidation
kubectl annotate nodepool default karpenter.sh/do-not-consolidate-

# View node utilization
kubectl top nodes -l karpenter.sh/nodepool=default