News › performance
By Zayden R., September 16, 2026
Kubernetes v1.37 has advanced Pod-Level Resource Managers to Beta, enhancing resource allocation flexibility for cluster operators. Engineers can now better manage NUMA-aligned resources and sidecars, improving performance for latency-critical applications.
In a move that will interest engineers managing complex Kubernetes deployments, version 1.37 has promoted the Pod-Level Resource Managers feature to Beta status. This shift, while subtle, represents a significant leap in how Kubernetes handles resource allocation, particularly for those running latency-critical applications with sidecars.
Initially introduced as an Alpha feature in Kubernetes v1.36, Pod-Level Resource Managers aim to resolve a persistent dilemma for cluster operators: the need to allocate exclusive NUMA-aligned CPU cores and memory without inefficiently dedicating resources to every container in a pod. Before this feature, operators faced an all-or-nothing scenario, which often resulted in wasted resources or suboptimal performance.
With the Beta graduation, Kubernetes equips its Topology Manager, CPU Manager, and Memory Manager to utilize pod-level resource declarations directly. This means the Kubelet can now reserve exclusive NUMA-aligned resources for primary application containers while allowing non-Guaranteed sidecars to operate within a pod-isolated shared pool. Such an allocation model ensures that primary workloads benefit from unthrottled, NUMA-local performance, while sidecars enjoy local alignment without consuming dedicated cores.
The Beta release brings operational and API enhancements, particularly through the PodResources API. The v1 PodResources gRPC service now includes top-level cpu_ids and memory fields in its responses, allowing monitoring tools and device plugins to query pod-level exclusive assignments without the risk of double-counting container allocations.
While the feature is disabled by default, operators eager to leverage these improvements can enable it via the PodLevelResourceManagers feature gate. For those looking to get started, the Kubernetes blog provides detailed documentation and tutorials on configuring and deploying workloads with these new capabilities. This update is poised to provide a more efficient and flexible resource management experience for Kubernetes users, especially those dealing with modern workloads that include lightweight sidecars.