Add validation rules for nvidia

kubernetes · Jul 17, 2021 · 73aec4c · 73aec4c
1 parent 2f7f7fe
commit 73aec4c
Show file tree

Hide file tree

Showing 2 changed files with 60 additions and 0 deletions.
diff --git a/pkg/apis/kops/validation/validation.go b/pkg/apis/kops/validation/validation.go
@@ -208,6 +208,10 @@ func validateClusterSpec(spec *kops.ClusterSpec, c *kops.Cluster, fieldPath *fie
 		allErrs = append(allErrs, validateDockerConfig(spec.Docker, fieldPath.Child("docker"))...)
 	}
 
+	if spec.Nvidia != nil {
+		allErrs = append(allErrs, validateNvidiaConfig(spec, fieldPath.Child("nvidia"))...)
+	}
+
 	if spec.Assets != nil {
 		if spec.Assets.ContainerProxy != nil && spec.Assets.ContainerRegistry != nil {
 			allErrs = append(allErrs, field.Forbidden(fieldPath.Child("assets", "containerProxy"), "containerProxy cannot be used in conjunction with containerRegistry"))
@@ -1366,6 +1370,20 @@ func validateDockerConfig(config *kops.DockerConfig, fldPath *field.Path) field.
 	return allErrs
 }
 
+func validateNvidiaConfig(spec *kops.ClusterSpec, fldPath *field.Path) (allErrs field.ErrorList) {
+	nvidia := spec.Nvidia
+	if !fi.BoolValue(nvidia.Enabled) {
+		return allErrs
+	}
+	if kops.CloudProviderID(spec.CloudProvider) != kops.CloudProviderAWS {
+		allErrs = append(allErrs, field.Forbidden(fldPath, "Nvidia is only supported on AWS"))
+	}
+	if spec.ContainerRuntime != "containerd" {
+		allErrs = append(allErrs, field.Forbidden(fldPath, "Nvidia is only supported using containerd"))
+	}
+	return allErrs
+}
+
 func validateRollingUpdate(rollingUpdate *kops.RollingUpdate, fldpath *field.Path, onMasterInstanceGroup bool) field.ErrorList {
 	allErrs := field.ErrorList{}
 	var err error

diff --git a/pkg/apis/kops/validation/validation_test.go b/pkg/apis/kops/validation/validation_test.go
@@ -1311,3 +1311,45 @@ func TestValidateSAExternalPermissions(t *testing.T) {
 	}
 
 }
+
+func Test_Validate_Nvdia(t *testing.T) {
+
+	grid := []struct {
+		Input          kops.ClusterSpec
+		ExpectedErrors []string
+	}{
+		{
+			Input: kops.ClusterSpec{
+				Nvidia: &kops.NvidiaConfig{
+					Enabled: fi.Bool(true),
+				},
+				CloudProvider:    "aws",
+				ContainerRuntime: "containerd",
+			},
+		},
+		{
+			Input: kops.ClusterSpec{
+				Nvidia: &kops.NvidiaConfig{
+					Enabled: fi.Bool(true),
+				},
+				CloudProvider:    "gce",
+				ContainerRuntime: "containerd",
+			},
+			ExpectedErrors: []string{"Forbidden::nvidia"},
+		},
+		{
+			Input: kops.ClusterSpec{
+				Nvidia: &kops.NvidiaConfig{
+					Enabled: fi.Bool(true),
+				},
+				CloudProvider:    "aws",
+				ContainerRuntime: "docker",
+			},
+			ExpectedErrors: []string{"Forbidden::nvidia"},
+		},
+	}
+	for _, g := range grid {
+		errs := validateNvidiaConfig(&g.Input, field.NewPath("nvidia"))
+		testErrors(t, g.Input, errs, g.ExpectedErrors)
+	}
+}