Kubernetes v1.37 Releases Dynamic Resource Allocation (DRA) Updates

DRA Extended Resource Support Reaches GA

Kubernetes v1.37 marks the General Availability of DRA Extended Resource support. This feature enables DRA drivers to fulfill requests made through the traditional extended resource API, such as example.com/gpu in a Pod specification, without requiring a separate device plugin. An extended resource name can be directly set on a DeviceClass, allowing Pods requesting it to be matched to a device via DRA without needing a ResourceClaim from the workload.

Enhanced Device Status Reporting

The ResourceClaims status now includes a 'devices' field, allowing DRA drivers to report per-device status. For network devices, this includes interface name, MAC address, and IP addresses. This provides users and controllers with visibility into device state that was previously unavailable once a device was configured in a Pod, enabling the development of network services reliant on reported device IPs.

Stable Device Taints and Tolerations

DRA device taints and tolerations have graduated to stable. This allows DRA drivers to mark devices as tainted, preventing them from being scheduled for new Pods. Cluster administrators can also apply these taints cluster-wide using a DeviceTaintRule. Pods already using a tainted device can be evicted automatically unless their ResourceClaim explicitly tolerates the taint, mirroring node taints for device maintenance.

Standardized NUMA Node Attribute

Kubernetes v1.37 introduces the standardized resource.kubernetes.io/numaNode as a shared attribute name. This allows devices from different drivers to be compared on the same NUMA node consistently, eliminating the need for each driver to invent its own naming convention for NUMA node identification. This attribute was introduced directly as stable in this release.