Can My AWS EKS Cluster Handle 100,000 Users? Load Testing + Autoscaling, Proven Live
Part 2 of a 4-part series — this article covers the repo at tag v2.0-load-testing-autoscaling. Part 1 (EKS deployment, tag v1.0-eks-deployment) is here . The Setup Last build: Google's Online Boutique running on a fixed 2-node AWS EKS cluster — 11 services, one Helm release, no autoscaling. The question this time: what actually happens under load? Not "the YAML looks right" — actually driven,…
The article details the results of load testing Amazon Web Services Elastic Kubernetes Service (EKS) clusters to determine if they can handle 100,000 users. The setup used a fixed 2-node EKS cluster running Google's Online Boutique with 11 services and no autoscaling. The new build introduced HorizontalPodAutoscaler (HPA) and metrics-server add-on, which were verified to be running before scaling. Karpenter was also introduced to provision new nodes when needed, eliminating the fixed node-group ceiling.
During the load test, k6 was run in-cluster with 3,000 virtual users (VUs) rather than the full 100,000, to avoid exceeding the free-tier-eligible node ceiling. The HPA replicas climbed live as CPU/memory percentages exceeded target thresholds, while Karpenter watched for pending pods and provisioned new nodes as needed. Once the load test finished, Karpenter consolidated idle nodes back to the original 2 nodes.
The key takeaways from the article are:
1. Verify metrics-server before trusting HPA, as an HPA without a metrics source doesn't scale and only appears as "unknown" targets.
2. Karpenter removes the node-group ceiling, but EC2NodeClass constraints still need to match what the account is allowed to provision.
3. Watching the load test end-to-end is crucial, as HPA config and node pool definition alone don't prove autoscaling works.
4. Be honest about scale-down results; the test used 3,000 VUs rather than the full 100,000.
The article concludes by asking whether to use Karpenter or Cluster Autoscaler for EKS in 2026.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
