tests/ai-conformance: create simple smoketest for CUDA via DRA

Author: justinsb

tests/e2e/scenarios/ai-conformance/validators/accelerators/dra_support/dra_cuda_test.go

@@ -18,10 +18,8 @@ package dra_support
 
 import (
 	"fmt"
-	"strings"
 	"testing"
 
-	"k8s.io/apimachinery/pkg/runtime/schema"
 	"k8s.io/kops/tests/e2e/scenarios/ai-conformance/validators"
 )
 
@@ -33,6 +31,22 @@ func TestDRAWorks(t *testing.T) {
 	h.Logf("## Listing device classes")
 	deviceClasses := h.ListDeviceClasses()
 	for _, deviceClass := range deviceClasses {
-		h.Logf("* %s", deviceClass.Name)
+		h.Logf("* %s", deviceClass.Name())
 	}
+
+	h.Logf("## Listing resource slices")
+	resourceSlices := h.ListResourceSlices()
+	for _, resourceSlice := range resourceSlices {
+		h.Logf("* %s", resourceSlice.Name())
+	}
+
+	if !h.HasDeviceClass("gpu.nvidia.com") {
+		t.Skipf("gpu.nvidia.com device class not found; skipping")
+	}
+
+	h.Logf("## Run cuda-smoketest")
+	ns := "default"
+	h.ShellExec(fmt.Sprintf("kubectl apply --namespace %s -f testdata/cuda-smoketest.yaml", ns))
+	h.ShellExec(fmt.Sprintf("kubectl wait --for=condition=complete --namespace %s job/cuda-smoketest --timeout=5m", ns))
+	h.ShellExec(fmt.Sprintf("kubectl logs --namespace %s job/cuda-smoketest", ns))
 }

tests/e2e/scenarios/ai-conformance/validators/accelerators/dra_support/testdata/cuda-smoketest.yaml

@@ -0,0 +1,39 @@
+# Create a ResourceClaim and Job to test DRA
+apiVersion: resource.k8s.io/v1
+kind: ResourceClaim
+metadata:
+  name: cuda-smoketest
+spec:
+  devices:
+    requests:
+    - name: single-gpu
+      exactly:
+        deviceClassName: gpu.nvidia.com
+        allocationMode: ExactCount
+        count: 1
+
+---
+
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: cuda-smoketest
+spec:
+  template:
+    spec:
+      restartPolicy: Never
+      tolerations:
+      - key: "nvidia.com/gpu"
+        operator: "Exists"
+        effect: "NoSchedule"
+      containers:
+      - name: test
+        image: nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0-ubuntu22.04
+        command: ["/bin/sh", "-c"]
+        args: ["/cuda-samples/vectorAdd"]
+        resources:
+          claims:
+          - name: gpu
+      resourceClaims:
+      - name: gpu
+        resourceClaimName: cuda-smoketest

tests/e2e/scenarios/ai-conformance/validators/kube.go

@@ -24,6 +24,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
+// DynamicClient returns a dynamic client for the kubernetes cluster.
 func (h *ValidatorHarness) DynamicClient() dynamic.Interface {
 	if h.dynamicClient == nil {
 		dynamicClient, err := dynamic.NewForConfig(h.restConfig)
@@ -35,6 +36,7 @@ func (h *ValidatorHarness) DynamicClient() dynamic.Interface {
 	return h.dynamicClient
 }
 
+// DeviceClass is a wrapper around the DRA DeviceClass type.
 type DeviceClass struct {
 	u *unstructured.Unstructured
 }
@@ -49,6 +51,7 @@ var deviceClassGVR = schema.GroupVersionResource{
 	Resource: "deviceclasses",
 }
 
+// ListDeviceClasses lists all device classes in the cluster.
 func (h *ValidatorHarness) ListDeviceClasses() []*DeviceClass {
 	objectList, err := h.DynamicClient().Resource(deviceClassGVR).List(h.Context(), metav1.ListOptions{})
 	if err != nil {
@@ -60,3 +63,41 @@ func (h *ValidatorHarness) ListDeviceClasses() []*DeviceClass {
 	}
 	return out
 }
+
+// HasDeviceClass returns true if a device class with the given name exists.
+func (h *ValidatorHarness) HasDeviceClass(name string) bool {
+	for _, deviceClass := range h.ListDeviceClasses() {
+		if deviceClass.Name() == name {
+			return true
+		}
+	}
+	return false
+}
+
+// ResourceSlice is a wrapper around the DRA ResourceSlice type.
+type ResourceSlice struct {
+	u *unstructured.Unstructured
+}
+
+func (d *ResourceSlice) Name() string {
+	return d.u.GetName()
+}
+
+var resourceSliceGVR = schema.GroupVersionResource{
+	Group:    "resource.k8s.io",
+	Version:  "v1",
+	Resource: "resourceslices",
+}
+
+// ListResourceSlices lists all resource slices in the cluster.
+func (h *ValidatorHarness) ListResourceSlices() []*ResourceSlice {
+	objectList, err := h.DynamicClient().Resource(resourceSliceGVR).List(h.Context(), metav1.ListOptions{})
+	if err != nil {
+		h.Fatalf("failed to list resource slices: %v", err)
+	}
+	var out []*ResourceSlice
+	for i := range objectList.Items {
+		out = append(out, &ResourceSlice{u: &objectList.Items[i]})
+	}
+	return out
+}