Vertical Pod Autoscaler and Cluster Autoscaler

Most teams set resource requests once during initial deployment and never touch them again. The values are either copied from a template, borrowed from a colleague's manifest, or guessed based on a vague memory of load testing three quarters ago. The result: some pods hoard CPU they never use while others OOM-kill every few days because nobody bumped the memory request.

The Vertical Pod Autoscaler (VPA) and Cluster Autoscaler (CA) close this gap. VPA adjusts pod resource requests based on actual consumption. Cluster Autoscaler adds and removes nodes to match what those workloads actually need. Together with the HPA covered in Horizontal Pod Autoscaler, they form a complete autoscaling stack: HPA scales replicas, VPA rightsizes each replica, and CA provides the underlying compute.