Kubernetes 1.37 has promoted Pod-Level Resource Managers to beta status. The feature lets the kubelet’s Topology Manager, CPU Manager and Memory Manager use resource declarations made at the pod level when making hardware-placement decisions.
The capability is designed for workloads that need predictable, NUMA-aware performance while also running auxiliary containers such as logging agents or telemetry exporters. Instead of assigning exclusive resources to every container, operators can reserve dedicated resources for primary application containers and place sidecars in a pod-isolated shared pool.
The beta feature is disabled by default and is controlled by the PodLevelResourceManagers feature gate. Kubernetes 1.37 also adds top-level CPU and memory fields to the v1 PodResources gRPC service, enabling monitoring tools and device plugins to inspect pod-level exclusive assignments.
The change is aimed at latency-sensitive and resource-intensive workloads where container-level allocation can be wasteful or insufficient. Operators evaluating the feature will need to test scheduling behavior, monitoring integrations and workload isolation before enabling it in production.
