News › performance
By Zayden R., September 3, 2026
Kubernetes v1.37 now supports scaling workloads to zero replicas using HorizontalPodAutoscaler, helping engineers optimize resource use and cost. This feature is in Beta and enabled by default, simplifying previous complexities.
Kubernetes version 1.37 has officially introduced a feature that many system administrators and engineers have been eagerly anticipating: the ability to scale workloads to zero replicas using the HorizontalPodAutoscaler (HPA). This update, now in Beta and enabled by default, allows Kubernetes to scale workloads down to zero, and then bring them back up as needed based on specific object or external metrics.
In prior versions, achieving a zero-replica state required either external components or enabling an Alpha feature gate. The inclusion of this capability in the core Kubernetes API simplifies the process considerably. This is a solid improvement for those managing workloads like queue consumers and batch processors, where removing the last idle Pod can lead to significant cost savings, especially when those Pods are reserving expensive resources such as dedicated CPUs or GPUs.
However, there's a trade-off to consider: cold-start time. When scaling from zero, the HPA needs to detect a change in the metric, schedule a Pod, and start the application. This latency is acceptable if the workload can tolerate waiting, such as when tasks are queued in a durable system. On the other hand, request-driven workloads like HTTP services will need a buffering layer, as Kubernetes Services do not buffer requests when no Pods are ready.
To effectively scale to zero, engineers must use object or external metrics since traditional CPU or memory metrics become unavailable once there are no running Pods. For instance, a metric like queue length remains accessible even when no workers are active. An example configuration might use a Prometheus metric named queue_consumer_lag to monitor queue length and trigger scaling actions. Engineers should ensure their setup includes a metrics adapter like the Prometheus Adapter to expose these metrics through the External Metrics API.
This advancement in Kubernetes v1.37 is an important step towards more efficient resource management, allowing engineers to optimize their systems better while minimizing unnecessary overhead.