kubernetes-sigs/dra-driver-nvidia-gpu v0.5.1 on GitHub

This is a patch release of the DRA Driver for NVIDIA GPUs. It includes reliability fixes for ComputeDomain and Kubelet plugin.

Bug Fixes

Fixed a race in the ComputeDomain controller that could incorrectly remove live nodes from ComputeDomain status, reassign IMEX node indices, and disrupt GPU memory export/import operations. This was most reproducible with ComputeDomainCliques disabled (#1440, #1418).
Fixed an IMEX DNS search-domain flood caused by unresolved static node names falling through to cluster DNS. #1466
Add a resources.computeDomains.imex.config Helm value to override or add arbitrary nvidia-imex daemon config file settings (e.g. IMEX_NODE_DISCONNECTED_GRACE_TIME). Defaults to {}, no behavior change unless set. #1497
Compute Domain daemon pods now reuse the controller's own priorityClassName. CD daemon pods now default to system-node-critical priority, matching the controller's own default. #1498
Fixed bug where certain MIG profiles do not get the CI profile with the most multiprocessors when created. #1454
Disabled registration of the unimplemented DRA resource health service in both kubelet plugins, preventing repeated device health reporting is not supported by this driver errors on Kubernetes v1.36 (#1425, #1423).
Standardized log messages and error formatting across the driver to improve troubleshooting (#1376, #1369).

Installation

Helm install

helm install dra-driver-nvidia-gpu oci://registry.k8s.io/dra-driver-nvidia/charts/dra-driver-nvidia-gpu \
  --version 0.5.1 \
  --namespace dra-driver-nvidia-gpu \
  --create-namespace \
  --set gpuResourcesEnabledOverride=true

Upgrade

Upgrade from v0.5.0, using your existing Helm release name, namespace, and configured values:

helm upgrade dra-driver-nvidia-gpu oci://registry.k8s.io/dra-driver-nvidia/charts/dra-driver-nvidia-gpu \
  --version 0.5.1 \
  --namespace dra-driver-nvidia-gpu \
  -f values.yaml

Replace values.yaml with the values used for your existing installation.

For installations using the earlier release name nvidia-dra-driver-gpu, retain that release name and namespace, along with nameOverride=nvidia-dra-driver-gpu in your configured values. Refer to the upgrade documentation for details.

Container images

registry.k8s.io/dra-driver-nvidia/dra-driver-nvidia-gpu:v0.5.1

Full Changelog: v0.5.0...v0.5.1