Kubernetes autoscaling works teams match the metric to the workload: queue latency for synchronous web traffic, queue depth for background jobs.