fix: (GH-#340) Add RuntimeClass for GPU sharing (#360)

Signed-off-by: Saurabh Kumar Singh <[email protected]>

Author: singh1203

docs/gpu-sharing/README.md

@@ -18,6 +18,11 @@ GPU sharing is disabled by default. To enable it, add the following flag to the
 --set "global.gpuSharing=true"
 ```
 
+### RuntimeClass Requirement
+KAI Scheduler requires the use of a specific RuntimeClass for GPU sharing. The recommended RuntimeClass is `nvidia`.
+
+KAI explicitly sets `runtimeClassName: "nvidia"` in the resource reservation pod spec. Ensure that your cluster has the `nvidia` RuntimeClass configured — this is typically provided by the NVIDIA device plugin.
+
 ### GPU Sharing Pod
 To submit a pod that can share a GPU device, run this command:
 ```

pkg/binder/binding/resourcereservation/resource_reservation.go

@@ -42,6 +42,8 @@ const (
 	unknownGpuIndicator            = "-1"
 )
 
+var runtimeClassName = "nvidia"
+
 type service struct {
 	fakeGPuNodes        bool
 	kubeClient          client.WithWatch
@@ -449,6 +451,7 @@ func (rsc *service) createResourceReservationPod(
 		},
 		Spec: v1.PodSpec{
 			NodeName:           nodeName,
+			RuntimeClassName:   &runtimeClassName,
 			ServiceAccountName: rsc.serviceAccountName,
 			Containers: []v1.Container{
 				{