test(operator): cover GMS discovery annotation pruning

deploy/operator/internal/dynamo/failover_test.go

@@ -493,6 +493,8 @@ func intraPodFailoverPodSpec() corev1.PodSpec {
 					{Name: "DYN_SYSTEM_USE_ENDPOINT_HEALTH_STATUS", Value: "true"},
 					{Name: "DYN_HEALTH_CHECK_ENABLED", Value: "true"},
 					{Name: commonconsts.DynamoDiscoveryBackendEnvVar, Value: "kubernetes"},
+					{Name: "CONTAINER_NAME", Value: commonconsts.MainContainerName},
+					{Name: "DYN_KUBE_DISCOVERY_MODE", Value: "container"},
 					{Name: "TMPDIR", Value: gms.SharedMountPath},
 				},
 				Ports: []corev1.ContainerPort{
@@ -555,15 +557,18 @@ func TestBuildFailoverPod_RejectsNonVLLM(t *testing.T) {
 }
 
 func TestBuildFailoverPod_EngineEnvVars(t *testing.T) {
+	t.Log("Build the active-passive engine containers from a container-discovery base")
 	ps := intraPodFailoverPodSpec()
 	err := buildFailoverPod(&ps, 1, BackendFrameworkVLLM)
 	require.NoError(t, err)
 
+	t.Log("Verify each engine keeps container discovery and receives its own container identity")
 	for i := range 2 {
 		engine := ps.Containers[i]
 		env := envToMap(engine.Env)
 		assert.Equal(t, strconv.Itoa(i), env["ENGINE_ID"], "engine-%d ENGINE_ID", i)
 		assert.Equal(t, fmt.Sprintf("engine-%d", i), env["CONTAINER_NAME"], "engine-%d CONTAINER_NAME", i)
+		assert.Equal(t, "container", env["DYN_KUBE_DISCOVERY_MODE"], "engine-%d discovery mode", i)
 		assert.Equal(t, intraPodFailoverLockFile, env["FAILOVER_LOCK_PATH"], "engine-%d FAILOVER_LOCK_PATH", i)
 		assert.Equal(t, "notready", env["DYN_SYSTEM_STARTING_HEALTH_STATUS"], "engine-%d starting health", i)
 		assert.Equal(t, "true", env["DYN_SYSTEM_ENABLED"], "engine-%d system enabled", i)

deploy/operator/internal/webhook/validation/dynamocomponentdeployment_validation_envtest_test.go

@@ -18,6 +18,7 @@
 package validation_test
 
 import (
+	"maps"
 	"testing"
 
 	nvidiacomv1alpha1 "github.com/ai-dynamo/dynamo/deploy/operator/api/v1alpha1"
@@ -59,6 +60,7 @@ func TestDynamoComponentDeploymentValidator_Validate(t *testing.T) {
 		wantCELErr         string
 		wantWebhookErrs    []string
 		wantWarnings       []string
+		wantPodAnnotations map[string]string
 	}{
 		// Baseline schema and webhook behavior.
 		{
@@ -871,15 +873,21 @@ func TestDynamoComponentDeploymentValidator_Validate(t *testing.T) {
 			wantCELErr: "spec.podTemplate.metadata.annotations: Invalid value: podTemplate backend annotation must be mp or ray, case-insensitively",
 		},
 		{
-			name: "v1beta1 valid pod template backend annotation reaches the webhook",
+			// Generated DCD pod metadata must survive structural pruning.
+			name: "v1beta1 discovery annotation survives the generated DCD API server round trip",
 			deployment: betaDCDForAdmission(func(dcd *nvidiacomv1beta1.DynamoComponentDeployment) {
 				dcd.Spec.PodTemplate = &corev1.PodTemplateSpec{
 					ObjectMeta: metav1.ObjectMeta{Annotations: map[string]string{
 						consts.KubeAnnotationVLLMDistributedExecutorBackend: "RaY",
+						consts.KubeAnnotationDynamoKubeDiscoveryMode:        "container",
 					}},
 					Spec: corev1.PodSpec{Containers: []corev1.Container{{Name: consts.MainContainerName}}},
 				}
 			}),
+			wantPodAnnotations: map[string]string{
+				consts.KubeAnnotationVLLMDistributedExecutorBackend: "RaY",
+				consts.KubeAnnotationDynamoKubeDiscoveryMode:        "container",
+			},
 		},
 		{
 			name: "v1alpha1 invalid extra pod metadata annotation reaches the webhook",
@@ -1016,7 +1024,20 @@ func TestDynamoComponentDeploymentValidator_Validate(t *testing.T) {
 				seedGates.Checkpoint = true
 				test.seedGates = &seedGates
 			}
-			runAdmissionTest(t, test)
+			actual := runAdmissionTest(t, test)
+			if tt.wantPodAnnotations != nil {
+				t.Log("Verify the API server preserved embedded pod-template annotations")
+				var actualDCD nvidiacomv1beta1.DynamoComponentDeployment
+				if err := runtime.DefaultUnstructuredConverter.FromUnstructured(actual.Object, &actualDCD); err != nil {
+					t.Fatalf("convert admitted DCD: %v", err)
+				}
+				if actualDCD.Spec.PodTemplate == nil {
+					t.Fatal("admitted DCD has no spec.podTemplate")
+				}
+				if got := actualDCD.Spec.PodTemplate.Annotations; !maps.Equal(got, tt.wantPodAnnotations) {
+					t.Fatalf("spec.podTemplate.metadata.annotations = %v, want %v", got, tt.wantPodAnnotations)
+				}
+			}
 		})
 	}
 }

deploy/operator/internal/webhook/validation/dynamographdeployment_validation_envtest_test.go

@@ -19,6 +19,7 @@ package validation_test
 
 import (
 	"fmt"
+	"maps"
 	"strings"
 	"testing"
 
@@ -56,12 +57,13 @@ func TestDynamoGraphDeploymentValidator_Validate(t *testing.T) {
 		checkpointOff   bool                             // disables checkpoint creation and restore
 		username        string                           // supplies the admission request identity
 
-		wantSchemaErr     string
-		wantCELErr        string
-		wantAdmissionErrs []string
-		wantWebhookErrs   []string
-		wantWarnings      []string
-		notWantErr        string
+		wantSchemaErr      string
+		wantCELErr         string
+		wantAdmissionErrs  []string
+		wantWebhookErrs    []string
+		wantWarnings       []string
+		notWantErr         string
+		wantPodAnnotations map[string]string
 	}{
 		// Baseline create-path rules.
 		{
@@ -295,15 +297,21 @@ func TestDynamoGraphDeploymentValidator_Validate(t *testing.T) {
 			wantCELErr: "spec.components[1].podTemplate.metadata.annotations: Invalid value: podTemplate backend annotation must be mp or ray, case-insensitively",
 		},
 		{
-			name: "v1beta1 valid pod template backend annotation reaches the webhook",
+			// Component pod metadata must survive structural pruning.
+			name: "v1beta1 discovery annotation survives the component pod template API server round trip",
 			deployment: betaDGDForAdmission(func(dgd *nvidiacomv1beta1.DynamoGraphDeployment) {
 				betaWorkerComponent(dgd).PodTemplate = &corev1.PodTemplateSpec{
 					ObjectMeta: metav1.ObjectMeta{Annotations: map[string]string{
 						consts.KubeAnnotationVLLMDistributedExecutorBackend: "RaY",
+						consts.KubeAnnotationDynamoKubeDiscoveryMode:        "container",
 					}},
 					Spec: corev1.PodSpec{Containers: []corev1.Container{{Name: consts.MainContainerName}}},
 				}
 			}),
+			wantPodAnnotations: map[string]string{
+				consts.KubeAnnotationVLLMDistributedExecutorBackend: "RaY",
+				consts.KubeAnnotationDynamoKubeDiscoveryMode:        "container",
+			},
 		},
 		{
 			name: "v1alpha1 converted extraPodMetadata annotation does not receive v1beta1 CEL validation",
@@ -1658,7 +1666,26 @@ func TestDynamoGraphDeploymentValidator_Validate(t *testing.T) {
 					test.seedGates = &seedGates
 				}
 			}
-			runAdmissionTest(t, test)
+			actual := runAdmissionTest(t, test)
+			if tt.wantPodAnnotations != nil {
+				t.Log("Verify the API server preserved embedded component pod-template annotations")
+				var actualDGD nvidiacomv1beta1.DynamoGraphDeployment
+				if err := runtime.DefaultUnstructuredConverter.FromUnstructured(actual.Object, &actualDGD); err != nil {
+					t.Fatalf("convert admitted DGD: %v", err)
+				}
+				component := actualDGD.GetComponentByName(dgdAdmissionWorkerName)
+				if component == nil || component.PodTemplate == nil {
+					t.Fatalf("admitted DGD has no pod template for component %q", dgdAdmissionWorkerName)
+				}
+				if got := component.PodTemplate.Annotations; !maps.Equal(got, tt.wantPodAnnotations) {
+					t.Fatalf(
+						"spec.components[%q].podTemplate.metadata.annotations = %v, want %v",
+						dgdAdmissionWorkerName,
+						got,
+						tt.wantPodAnnotations,
+					)
+				}
+			}
 		})
 	}
 }