News › linux
By Zayden R., September 4, 2026
Kubernetes 1.37 introduces Dynamic Resource Allocation (DRA) Extended Resource support to GA, allowing streamlined resource requests without device plugins. Engineers should explore gang scheduling beta to optimize GPU idle costs.
Kubernetes 1.37 has officially launched, bringing with it a significant milestone for Dynamic Resource Allocation (DRA): Extended Resource support has graduated to General Availability (GA). This advancement simplifies the process for DRA drivers to fulfill resource requests, such as example.com/gpu in a Pod spec, without the need for a separate device plugin. The transition to GA signifies a stable feature, following its path from Alpha in 1.35 to Beta in 1.36. For engineers, this means existing workloads utilizing extended resources can continue seamlessly while the underlying allocation shifts to DRA.
In addition to DRA Extended Resources, Kubernetes 1.37 introduces several other noteworthy updates. ResourceClaims now include a devices field in their status, providing visibility into device states like interface names, MAC addresses, and IPs. This enhancement is particularly useful for building network services dependent on a device's reported IPs.
The release also sees DRA device taints and tolerations becoming stable, enabling devices to be marked as tainted, preventing new Pod scheduling. Admins can enforce these taints cluster-wide without needing driver reconfiguration, allowing individual devices to be taken offline for maintenance without disrupting the cluster.
Another highlight is the standardization of the NUMA node attribute, resource.kubernetes.io/numaNode, which allows devices from different drivers to be compared on the same NUMA node. This feature landed directly as stable in this release.
Furthermore, Kubernetes 1.37 introduces gang scheduling in beta, a feature aimed at reducing GPU idle costs by ensuring that related tasks are scheduled together, optimizing resource utilization. Engineers managing workloads with significant GPU demands should consider experimenting with this feature to enhance efficiency.
Overall, Kubernetes 1.37 delivers a suite of improvements that enhance resource management and scheduling, offering engineers robust tools to optimize their cluster operations.