How do I right-size Kubernetes resource requests without over-provisioning?

Use actual historical usage data — not guesses — via a tool like the Vertical Pod Autoscaler in recommendation-only mode, or Kubecost’s request-sizing recommendations, run over at least a full weekly traffic cycle.

A common mistake

Sizing requests off peak usage rather than typical usage inflates cluster costs significantly; size requests to typical (e.g., p50) usage and rely on limits plus horizontal pod autoscaling to absorb peaks.