What is a Cluster Autoscaler?

What is a Cluster Autoscaler?

Cluster Autoscaler is a Kubernetes component that automatically adds or removes nodes in a cluster based on pod scheduling needs.

• It scales the number of nodes (in contrast to Horizontal Pod Autoscaler which scales pods).
• Helps when you run out of resources for pending pods.
• Saves cost by removing underutilized nodes.

---

AWS EKS vs GKE: Built-in Autoscaler

• GKE

  • Built-in Autoscaler: Yes (Native support)
  • Manual Setup Needed?: Not required (Just enable it)

• EKS

  • Built-in Autoscaler: No (Not enabled by default)
  • Manual Setup Needed?: Yes, you must install & configure it

---

How to Use It on GKE (Google Kubernetes Engine)

In GKE, you simply enable autoscaling when creating a node pool.

Step-by-step (Console):

1. Go to Kubernetes Engine → Clusters.
2. Click on your cluster name.
3. Click Edit for the node pool.
4. Enable Autoscaling.
5. Set min/max node count (e.g., 1–5).
6. Save.

That’s it! GKE manages it from there.

---

How to Use It on EKS (AWS Kubernetes)

You need to install Cluster Autoscaler manually on EKS. Here's how:

Step 1: Enable IAM permissions

Attach the required IAM policy to your node group role. Example:

aws iam attach-role-policy \
  --policy-arn arn:aws:iam::aws:policy/AutoScalingFullAccess \
  --role-name <NodeInstanceRole>

Step 2: Deploy Cluster Autoscaler

You can deploy it with Helm or use the official YAML.

Here’s the official YAML method:

kubectl apply -f https://github.com/kubernetes/autoscaler/releases/download/cluster-autoscaler-<version>/cluster-autoscaler-autodiscover.yaml

Replace <version> with your Kubernetes version (e.g. v1.27.0)

Step 3: Modify the deployment

Edit the cluster-autoscaler deployment:

kubectl -n kube-system edit deployment cluster-autoscaler

And in the args, specify:

- --cloud-provider=aws
- --nodes=1:5:<your-node-group-name>
- --balance-similar-node-groups
- --skip-nodes-with-system-pods=false
- --skip-nodes-with-local-storage=false

Also, set:

containers:
  - name: cluster-autoscaler
    image: k8s.gcr.io/autoscaling/cluster-autoscaler:<version>
    ...
    env:
      - name: AWS_REGION
        value: us-east-1

---

Cluster Autoscaler Works With:

• HPA (Horizontal Pod Autoscaler): You scale pods and when new pods can't be scheduled due to resource limits, Cluster Autoscaler adds nodes.
• VPA (Vertical Pod Autoscaler): Adjusts pod resources, which may trigger node scaling.

---

How to Test

1. Deploy a pod with more resources than your node can handle.
2. You’ll see it pending.
3. Cluster Autoscaler will detect the unschedulable pod and add a new node.
4. When nodes are underutilized, it removes them automatically.

---

Tip: Labels Matter

Cluster Autoscaler needs node groups or pools to be properly labeled:

For GKE:

--node-group-autoprovisioning

For EKS: Ensure the node group name is referenced in the autoscaler args.