Kubernetes HPA Generator

The HorizontalPodAutoscaler is the Kubernetes object that raises and lowers the replica count of a Deployment as load changes. It watches a metric, compares it against the target you set and adjusts the count to bring the average back to that target. Without it, sizing means choosing between paying for idle capacity or having no headroom at peak.

Fill in the resource name, the target Deployment, the minimum and maximum replica bounds and the desired CPU percentage. The page returns the manifest on autoscaling/v2, the current version and the one that accepts multiple metrics. Save it to a file and apply with kubectl apply -f. A 70% target is a reasonable starting point: it leaves room for the time a new pod takes to come up and become ready.

One requirement sinks a lot of configurations: the CPU percentage is computed against each container's requests.cpu, not against the limit and not against node capacity. If the Deployment declares no CPU requests, the HPA cannot work out the ratio and sits there with no metric, showing unknown under kubectl get hpa. metrics-server also has to be installed in the cluster — without it, there is nowhere to read consumption from.