Inspect DRA devices

Kubernetes Dynamic Resource Allocation (DRA) describes every device a driver finds on a node in a ResourceSlice. This page shows how to find those devices in Lens K8S IDE and how to read what a driver says about them. For the views themselves, see Device Resources.

Whatever the hardware, a device is described by the same three fields: a name, attributes, and capacity.

Find the devices on a node

To see what a node advertises:

1. From your cluster, navigate to Nodes.
2. Click the node to open its details panel.
3. Scroll to the Devices section.

The section lists every device published for that node, with the driver that published it, its pool, the ResourceClaim holding it, and its attributes. A device with no claim in the Allocated By column is free. Click a row to open the ResourceSlice the device came from.

Read a ResourceSlice

A slice belongs to one driver, lives in a pool, targets a node, and carries a list of devices:

apiVersion: resource.k8s.io/v1
kind: ResourceSlice
spec:
  driver: <DRIVER-NAME>
  pool:
    name: <POOL-NAME>
    generation: 1
    resourceSliceCount: 1
  nodeName: <NODE-NAME>     # or nodeSelector, or allNodes: true
  devices:
    - name: <DEVICE-NAME>
      attributes: {}         # int, bool, string, or version values
      capacity: {}           # quantities such as "40Gi"
      taints: []

Read a device description

Attributes name the hardware. The gpu.nvidia.com driver names full GPUs gpu-0, gpu-1, and so on, and describes each one like this:

- name: gpu-0
  attributes:
    productName:           { string: "NVIDIA A100-PCIE-40GB" }
    architecture:          { string: "Ampere" }
    cudaComputeCapability: { version: "8.0.0" }
    driverVersion:         { version: "580.126.20" }
    uuid:                  { string: "GPU-2fa81118-5a5f-aa66-7660-471eed407181" }
    resource.kubernetes.io/pciBusID: { string: "0000:65:00.0" }
    resource.kubernetes.io/numaNode: { int: 0 }
  capacity:
    memory:          { value: "40Gi" }
    multiprocessors: { value: "108" }

Most attributes are namespaced to the driver that publishes them, so their names differ between a GPU driver and a network driver. Attributes prefixed resource.kubernetes.io/ are standardized, and every driver reports them the same way. The pcieRoot and numaNode attributes are what let one claim request a GPU and a network interface on the same PCIe root.

For the attributes and capacities a specific driver publishes, read that driver's documentation. Device classes select on these values, so to see which devices a class covers, open it in the Device Classes view and read its Selectors.