DRA plugins can be installed, but do not really work because the required scheduler plugin DynamicResources isn't enabled.

4.14.3

Always

1. Install an OpenShift cluster and enable the TechPreviewNoUpgrade feature set either during installation or post-install. The feature set includes the DynamicResourceAllocation feature gate.
2. Install a DRA plugin by any vendor, e.g. by NVIDIA (requires at least one GPU worker with NVIDIA GPU drivers installed on the node, and a few tweaks to allow the plugin to run on OpenShift).
3. Create a resource claim.
4. Create a pod that consumes the resource claim.

The pod remains in ContainerCreating state, the claim in WaitingForFirstConsumer state forever, without any meaningful event or error message.

A resource is allocated according to the resource claim, and assigned to the pod.

The problem is caused by the DynamicResources scheduler plugin not being automatically enabled when the feature flag is turned on. This makes DRA plugins run without issues (the right APIs are available), but do nothing.