Autoscaling in Kubernetes: HPA, VPA, and Cluster Autoscaler

Load spikes. HPA is configured. New pods get created. Then they sit in Pending for three minutes while your users hammer a service that can't keep up. Not because the autoscaler failed. It worked exactly as designed. There were simply no nodes to schedule on when the pods arrived.

That's the most common production autoscaling failure, and it happens because autoscaling in Kubernetes is a three-layer system that breaks when any layer is missing or misconfigured. This lesson maps the three layers, where they interact, and where they conflict, so the deep dives that follow have a clear foundation.