Amazon EKS — Cluster Autoscaler

1 minute read

Cluster Autoscaler

When HPA adds more pods but there’s no room on existing nodes, Cluster Autoscaler adds more nodes to the cluster.

How It Works

Cluster Autoscaler Flow
┌─────────────────────────────────────────────────────┐
│                                                     │
│  1. Pod is Pending (insufficient resources)          │
│  2. Cluster Autoscaler detects pending pods          │
│  3. Evaluates which node group can fit the pod       │
│  4. Increases ASG desired count                      │
│  5. New node registers with cluster                  │
│  6. Pod scheduled on new node                        │
│                                                     │
│  Scale Down:                                         │
│  1. Node utilization < 50% for 10 minutes            │
│  2. All pods can be moved to other nodes             │
│  3. Node cordoned, pods drained, node terminated     │
│                                                     │
└─────────────────────────────────────────────────────┘

Install Cluster Autoscaler

# Step 1: Create IRSA
$ eksctl create iamserviceaccount \
    --cluster demo \
    --name cluster-autoscaler \
    --namespace kube-system \
    --attach-policy-arn arn:aws:iam::123456789012:policy/ClusterAutoscalerPolicy \
    --approve

# Step 2: Deploy
$ kubectl apply -f https://raw.githubusercontent.com/kubernetes/autoscaler/master/cluster-autoscaler/cloudprovider/aws/examples/cluster-autoscaler-autodiscover.yaml

# Step 3: Add cluster name annotation
$ kubectl -n kube-system annotate deployment.apps/cluster-autoscaler \
    cluster-autoscaler.kubernetes.io/safe-to-evict="false"

Configuration

# Key flags for Cluster Autoscaler
--balance-similar-node-groups=true      # Balance across AZs
--skip-nodes-with-local-storage=false   # Allow scale down of nodes with emptyDir
--expander=least-waste                  # Efficient node selection
--scale-down-utilization-threshold=0.5  # Scale down if < 50% utilized
--scale-down-delay-after-add=10m        # Wait 10 min after adding a node

Common Mistakes

Mistake Impact
Missing ASG tags Autoscaler can’t find node groups
maxSize too low Can’t scale enough during peaks
Not using --balance-similar-node-groups Uneven AZ distribution
PodDisruptionBudgets blocking scale-down Nodes can’t be drained

What’s Next?

The modern alternative: Karpenter.

You may also enjoy