Add support for NO_RUNTIME_CONFIG to disable runtime config in the nvidia-container-toolkit

controllers/object_controls.go

@@ -1366,14 +1366,29 @@ func TransformToolkit(obj *appsv1.DaemonSet, config *gpuv1.ClusterPolicySpec, n
 		}
 	}
 
-	if len(config.Toolkit.Env) > 0 {
-		for _, env := range config.Toolkit.Env {
-			setContainerEnv(toolkitMainContainer, env.Name, env.Value)
-		}
-	}
-
 	// configure runtime
 	runtime := n.runtime.String()
+	var noRuntimeConfig bool
+	// Update the main container environment from the user-specified values.
+	for _, env := range config.Toolkit.Env {
+		switch env.Name {
+		case "RUNTIME":
+			// If the user has specified the runtime, we overide the detected
+			// value.
+			// TODO: Add logging.
+			runtime = env.Value
+		case "NO_RUNTIME_CONFIG":
+			noRuntimeConfig = func() bool {
+				v, _ := strconv.ParseBool(env.Value)
+				return v
+			}()
+		}
+		setContainerEnv(toolkitMainContainer, env.Name, env.Value)
+	}
+
+	if noRuntimeConfig {
+		return nil
+	}
 	err = transformForRuntime(obj, config, runtime, toolkitMainContainer)
 	if err != nil {
 		return fmt.Errorf("error transforming toolkit daemonset : %w", err)

controllers/transforms_test.go

@@ -17,6 +17,7 @@
 package controllers
 
 import (
+	"errors"
 	"path/filepath"
 	"testing"
 
@@ -777,11 +778,12 @@ func TestApplyCommonDaemonsetMetadata(t *testing.T) {
 
 func TestTransformToolkit(t *testing.T) {
 	testCases := []struct {
-		description string
-		ds          Daemonset                // Input DaemonSet
-		cpSpec      *gpuv1.ClusterPolicySpec // Input configuration
-		runtime     gpuv1.Runtime
-		expectedDs  Daemonset // Expected output DaemonSet
+		description   string
+		ds            Daemonset                // Input DaemonSet
+		cpSpec        *gpuv1.ClusterPolicySpec // Input configuration
+		runtime       gpuv1.Runtime
+		expectedError error
+		expectedDs    Daemonset // Expected output DaemonSet
 	}{
 		{
 			description: "transform nvidia-container-toolkit-ctr container",
@@ -1002,6 +1004,42 @@ func TestTransformToolkit(t *testing.T) {
 				WithHostPathVolume("crio-config", "/etc/crio", ptr.To(corev1.HostPathDirectoryOrCreate)).
 				WithHostPathVolume("crio-drop-in-config", "/etc/crio/crio.conf.d", ptr.To(corev1.HostPathDirectoryOrCreate)),
 		},
+		{
+			description:   "no nvidia-container-toolkit-ctr container",
+			ds:            NewDaemonset(),
+			expectedError: errors.New(`failed to find toolkit container "nvidia-container-toolkit-ctr"`),
+			expectedDs:    NewDaemonset(),
+		},
+		{
+			description: "transform nvidia-container-toolkit-ctr container with no-runtime-config",
+			ds: NewDaemonset().
+				WithContainer(corev1.Container{Name: "nvidia-container-toolkit-ctr"}),
+			runtime: gpuv1.Containerd,
+			cpSpec: &gpuv1.ClusterPolicySpec{
+				Toolkit: gpuv1.ToolkitSpec{
+					Repository: "nvcr.io/nvidia/cloud-native",
+					Image:      "nvidia-container-toolkit",
+					Version:    "v1.0.0",
+					Env: []gpuv1.EnvVar{
+						{Name: "NO_RUNTIME_CONFIG", Value: "true"},
+					},
+				},
+			},
+			expectedDs: NewDaemonset().
+				WithContainer(corev1.Container{
+					Name:            "nvidia-container-toolkit-ctr",
+					Image:           "nvcr.io/nvidia/cloud-native/nvidia-container-toolkit:v1.0.0",
+					ImagePullPolicy: corev1.PullIfNotPresent,
+					Env: []corev1.EnvVar{
+						{Name: "CDI_ENABLED", Value: "true"},
+						{Name: "NVIDIA_RUNTIME_SET_AS_DEFAULT", Value: "false"},
+						{Name: "NVIDIA_CONTAINER_RUNTIME_MODE", Value: "cdi"},
+						{Name: "CRIO_CONFIG_MODE", Value: "config"},
+						{Name: "NO_RUNTIME_CONFIG", Value: "true"},
+					},
+					VolumeMounts: nil,
+				}),
+		},
 	}
 
 	for _, tc := range testCases {
@@ -1012,7 +1050,7 @@ func TestTransformToolkit(t *testing.T) {
 			}
 
 			err := TransformToolkit(tc.ds.DaemonSet, tc.cpSpec, controller)
-			require.NoError(t, err)
+			require.EqualValues(t, tc.expectedError, err)
 			require.EqualValues(t, tc.expectedDs, tc.ds)
 		})
 	}