Kubernetes v1.37 introduces Beta support for scaling workloads to zero replicas using object or external metrics. This feature eliminates the need for custom add-ons to terminate idle pods, though it requires an external metrics provider like Prometheus to trigger scaling operations when no pods are active.
Key points
HorizontalPodAutoscaler can now scale workloads to zero replicas natively in Kubernetes v1.37.
Scaling to zero requires external or object metrics, as CPU and memory metrics rely on active pods.
Scaling to zero is most effective for resource-heavy background tasks like queue processors.
HTTP workloads require a separate buffering layer because Kubernetes services do not buffer requests for zero-replica services.
A metrics adapter is necessary to expose external system metrics to the Kubernetes API for HPA configuration.