feat(nvca): extend control-plane validator with HA checks, DaemonSet n2n, and route CR type check

src/compute-plane-services/nvca/cmd/cluster-validator/BUILD.bazel

@@ -13,6 +13,7 @@ go_library(
         "//src/compute-plane-services/nvca/cmd/internal",
         "//src/compute-plane-services/nvca/internal/clustervalidator",
         "//src/compute-plane-services/nvca/vendor/github.com/NVIDIA/nvcf/src/libraries/go/lib/pkg/core",
+        "//src/compute-plane-services/nvca/vendor/k8s.io/client-go/dynamic",
     ],
 )
 
@@ -42,4 +43,7 @@ go_test(
     name = "cluster-validator_test",
     srcs = ["main_test.go"],
     embed = [":cluster-validator_lib"],
+    deps = [
+        "//src/compute-plane-services/nvca/internal/clustervalidator",
+    ],
 )

src/compute-plane-services/nvca/cmd/cluster-validator/main.go

@@ -72,11 +72,34 @@ func main() {
 			clustervalidator.SummaryConfigMapNamespaceEnv)
 	}
 
-	if err := clustervalidator.Run(ctx, client, configNS, configName, summaryNS, emitMetrics); err != nil {
+	// VALIDATOR_ROLE selects which check set runs: "control-plane" enables
+	// gateway and StorageClass checks and skips GPU/SMB; anything else (including
+	// unset) runs the compute-plane check set (backward-compatible default).
+	roleEnv := os.Getenv("VALIDATOR_ROLE")
+	role := parseRole(roleEnv)
+	if roleEnv != "" && role == "" {
+		log.Warnf("VALIDATOR_ROLE=%q is not recognized; defaulting to compute-plane", roleEnv)
+	}
+
+	if err := clustervalidator.Run(ctx, client, configNS, configName, summaryNS, emitMetrics, role); err != nil {
 		log.WithError(err).Fatal("Cluster validation failed")
 	}
 }
 
+// parseRole normalizes the VALIDATOR_ROLE env value. Returns the matching
+// clustervalidator constant for "control-plane" or "compute-plane"; returns ""
+// (compute-plane default) for any other value so unknown inputs are safe.
+func parseRole(v string) string {
+	switch strings.ToLower(strings.TrimSpace(v)) {
+	case clustervalidator.RoleControlPlane:
+		return clustervalidator.RoleControlPlane
+	case clustervalidator.RoleComputePlane:
+		return clustervalidator.RoleComputePlane
+	default:
+		return ""
+	}
+}
+
 // preflightMode reports whether this is a one-shot preflight run (e.g. nvcf-cli,
 // before NVCA is installed), which skips the summary write. Read from an env
 // (not a flag) so an unknown value is ignored rather than crashing arg parsing.

src/compute-plane-services/nvca/cmd/cluster-validator/main_test.go

@@ -17,7 +17,35 @@ limitations under the License.
 
 package main
 
-import "testing"
+import (
+	"testing"
+
+	"github.com/NVIDIA/nvcf/src/compute-plane-services/nvca/internal/clustervalidator"
+)
+
+func TestParseRole(t *testing.T) {
+	tests := []struct {
+		in   string
+		want string
+	}{
+		// Known roles are normalized.
+		{"control-plane", clustervalidator.RoleControlPlane},
+		{"CONTROL-PLANE", clustervalidator.RoleControlPlane},
+		{" control-plane ", clustervalidator.RoleControlPlane},
+		{"compute-plane", clustervalidator.RoleComputePlane},
+		{"COMPUTE-PLANE", clustervalidator.RoleComputePlane},
+		// Unknown values (including unset) fall back to "" = compute-plane default.
+		{"", ""},
+		{"gpu", ""},
+		{"both", ""},
+		{"control_plane", ""}, // underscore, not hyphen
+	}
+	for _, tt := range tests {
+		if got := parseRole(tt.in); got != tt.want {
+			t.Errorf("parseRole(%q) = %q, want %q", tt.in, got, tt.want)
+		}
+	}
+}
 
 func TestPreflightMode(t *testing.T) {
 	tests := []struct {

src/compute-plane-services/nvca/internal/clustervalidator/BUILD.bazel

@@ -20,11 +20,13 @@ go_library(
     deps = [
         "//src/compute-plane-services/nvca/vendor/github.com/NVIDIA/nvcf/src/libraries/go/lib/pkg/core",
         "//src/compute-plane-services/nvca/vendor/github.com/sirupsen/logrus",
+        "//src/compute-plane-services/nvca/vendor/k8s.io/api/apps/v1:apps",
         "//src/compute-plane-services/nvca/vendor/k8s.io/api/core/v1:core",
         "//src/compute-plane-services/nvca/vendor/k8s.io/api/networking/v1:networking",
         "//src/compute-plane-services/nvca/vendor/k8s.io/apimachinery/pkg/api/errors",
         "//src/compute-plane-services/nvca/vendor/k8s.io/apimachinery/pkg/api/resource",
         "//src/compute-plane-services/nvca/vendor/k8s.io/apimachinery/pkg/apis/meta/v1:meta",
+        "//src/compute-plane-services/nvca/vendor/k8s.io/apimachinery/pkg/util/rand",
         "//src/compute-plane-services/nvca/vendor/k8s.io/apimachinery/pkg/util/intstr",
         "//src/compute-plane-services/nvca/vendor/k8s.io/client-go/discovery",
         "//src/compute-plane-services/nvca/vendor/k8s.io/client-go/kubernetes",
@@ -41,6 +43,7 @@ alias(
 go_test(
     name = "clustervalidator_test",
     srcs = [
+        "checks_controlplane_test.go",
         "checks_test.go",
         "config_test.go",
         "enforcement_test.go",