KEDA 3.0 Scale-to-Zero Slashes Kubernetes Idle Workload Costs

KEDA 3.0 introduces scale-to-zero autoscaling across 80+ event sources, including Kafka, RabbitMQ, and AWS SQS, allowing Kubernetes pods to drop to zero replicas when no events are queued. Unlike the standard Horizontal Pod Autoscaler, which enforces a minimum of one running replica, KEDA scales entirely based on event presence rather than CPU metrics. Workloads that are idle for more than half the day and can tolerate brief cold-start delays stand to benefit most, including queue consumers, batch jobs, and dev/staging environments. However, the approach carries tradeoffs: cold starts can delay first-event processing, and cost savings are only fully realized if the cluster's node autoscaler removes the underlying empty nodes. Tuning the cooldown period is also critical, as setting it too short causes rapid scaling oscillation while setting it too long leaves idle resources running unnecessarily.