This is a patch release of the DRA Driver for NVIDIA GPUs. It includes bug fixes and updates to supported platforms.
New Features
Added support for Red Hat OpenShift Container Platform 4.22.
Bug Fixes
- Bump github.com/NVIDIA/nvidia-container-toolkit from 1.19.0 to 1.19.1 in #1161
- Fix nil pointer panic in verifyDisabledVFs for non-SR-IOV GPUs in #1194
- Update go-nvlib to v0.11.0 in #1198
- Fixed an issue causing MPS failure on Red Hat OpenShift by using kubelet-plugin service account for MPS control daemon in #1214
- The compute-domain kubelet plugin now rejects stale prepare retries for ResourceClaims that have already completed unprepare, preventing stale prepared-device checkpoint state from being recreated for the same claim version (#1188, #1162).
- Fixed an issue where the default timeslicing settings were applied on unsupported GPUs (#1183, #363).
- Resolved an issue where redundant component labels caused errors when Helm templates were rendered with strict parsing in GitOps tools such as ArgoCD (#1187, #1184).
Installation
Helm install
helm install dra-driver-nvidia-gpu oci://registry.k8s.io/dra-driver-nvidia/charts/dra-driver-nvidia-gpu \
--version 0.4.1 \
--namespace dra-driver-nvidia-gpu \
--create-namespace \
--set gpuResourcesEnabledOverride=true
Upgrade
Upgrade from v0.4.0:
helm upgrade -i nvidia-dra-driver-gpu oci://registry.k8s.io/dra-driver-nvidia/charts/dra-driver-nvidia-gpu \
--version 0.4.1 \
--namespace nvidia-dra-driver-gpu \
--set gpuResourcesEnabledOverride=true \
--set nameOverride=nvidia-dra-driver-gpu
--set nameOverride=nvidia-dra-driver-gpu is only required on the first upgrade to v0.4.0 or later (to keep existing Kubernetes resource names stable). If you already upgraded to v0.4.0 with that flag, subsequent helm upgrade calls do not need it. Refer to the upgrade documentation for more details.
Container images
registry.k8s.io/dra-driver-nvidia/dra-driver-nvidia-gpu:v0.4.1
Full Changelog: v0.4.0...v0.4.1