fix: handle groveTerminationDelay and auto-detect grove installation (#2190)

deploy/cloud/helm/deploy.sh

@@ -49,7 +49,6 @@ export ISTIO_GATEWAY="${ISTIO_GATEWAY:=istio-system/istio-ingressgateway}"
 export INGRESS_CLASS="${INGRESS_CLASS:=nginx}"
 export VIRTUAL_SERVICE_SUPPORTS_HTTPS="${VIRTUAL_SERVICE_SUPPORTS_HTTPS:=false}"
 export ENABLE_LWS="${ENABLE_LWS:=false}"
-export ENABLE_GROVE="${ENABLE_GROVE:=false}"
 
 # Add command line options
 INTERACTIVE=false
@@ -165,7 +164,7 @@ echo "DYNAMO_INGRESS_SUFFIX: $DYNAMO_INGRESS_SUFFIX"
 echo "VIRTUAL_SERVICE_SUPPORTS_HTTPS: $VIRTUAL_SERVICE_SUPPORTS_HTTPS"
 echo "INSTALL_CRDS: $INSTALL_CRDS"
 
-envsubst '${NAMESPACE} ${RELEASE_NAME} ${DOCKER_USERNAME} ${DOCKER_PASSWORD} ${DOCKER_SERVER} ${IMAGE_TAG} ${DYNAMO_INGRESS_SUFFIX} ${PIPELINES_DOCKER_SERVER} ${PIPELINES_DOCKER_USERNAME} ${PIPELINES_DOCKER_PASSWORD} ${DOCKER_SECRET_NAME} ${INGRESS_ENABLED} ${ISTIO_ENABLED} ${INGRESS_CLASS} ${ISTIO_GATEWAY} ${VIRTUAL_SERVICE_SUPPORTS_HTTPS} ${ENABLE_LWS} ${ENABLE_GROVE}' < dynamo-platform-values.yaml > generated-values.yaml
+envsubst '${NAMESPACE} ${RELEASE_NAME} ${DOCKER_USERNAME} ${DOCKER_PASSWORD} ${DOCKER_SERVER} ${IMAGE_TAG} ${DYNAMO_INGRESS_SUFFIX} ${PIPELINES_DOCKER_SERVER} ${PIPELINES_DOCKER_USERNAME} ${PIPELINES_DOCKER_PASSWORD} ${DOCKER_SECRET_NAME} ${INGRESS_ENABLED} ${ISTIO_ENABLED} ${INGRESS_CLASS} ${ISTIO_GATEWAY} ${VIRTUAL_SERVICE_SUPPORTS_HTTPS} ${ENABLE_LWS}' < dynamo-platform-values.yaml > generated-values.yaml
 echo "generated file contents:"
 cat generated-values.yaml
 

deploy/cloud/helm/dynamo-platform-values.yaml

@@ -24,7 +24,6 @@ dynamo-operator:
 
   dynamo:
     enableLWS: ${ENABLE_LWS}
-    enableGrove: ${ENABLE_GROVE}
     ingress:
       enabled: ${INGRESS_ENABLED}
       className: ${INGRESS_CLASS}

deploy/cloud/helm/platform/components/operator/templates/deployment.yaml

@@ -100,8 +100,8 @@ spec:
         {{- if .Values.dynamo.enableLWS }}
           - --enable-lws
         {{- end }}
-        {{- if .Values.dynamo.enableGrove }}
-          - --enable-grove
+        {{- if .Values.dynamo.groveTerminationDelay }}
+          - --grove-termination-delay={{ .Values.dynamo.groveTerminationDelay }}
         {{- end }}
         command:
         - /manager

deploy/cloud/helm/platform/components/operator/templates/manager-rbac.yaml

@@ -116,7 +116,6 @@ rules:
   - patch
   - update
   - watch
-{{- if .Values.dynamo.enableGrove }}
 - apiGroups:
   - grove.io
   resources:
@@ -129,7 +128,6 @@ rules:
   - patch
   - update
   - watch
-{{- end }}
 - apiGroups:
   - apps
   resources:

deploy/cloud/helm/platform/components/operator/values.yaml

@@ -82,7 +82,7 @@ dynamo:
       annotations: {}
 
   enableLWS: false
-  enableGrove: false
+  groveTerminationDelay: 15m
 
   internalImages:
     debugger: python:3.12-slim

deploy/cloud/helm/platform/values.yaml

@@ -34,7 +34,7 @@ dynamo-operator:
   imagePullSecrets: []
   dynamo:
     enableLWS: false
-    enableGrove: false
+    groveTerminationDelay: 15m
     internalImages:
       debugger: python:3.12-slim
     enableRestrictedSecurityContext: false

deploy/cloud/operator/cmd/main.go

@@ -30,6 +30,7 @@ import (
 	// to ensure that exec-entrypoint and run can make use of them.
 	clientv3 "go.etcd.io/etcd/client/v3"
 	corev1 "k8s.io/api/core/v1"
+	apiextensionsv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
 	"k8s.io/client-go/informers"
 	"k8s.io/client-go/kubernetes"
 	_ "k8s.io/client-go/plugin/pkg/client/auth"
@@ -50,6 +51,7 @@ import (
 
 	grovev1alpha1 "github.com/NVIDIA/grove/operator/api/core/v1alpha1"
 	nvidiacomv1alpha1 "github.com/ai-dynamo/dynamo/deploy/cloud/operator/api/v1alpha1"
+	"github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/consts"
 	"github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/controller"
 	commonController "github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/controller_common"
 	"github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/etcd"
@@ -73,6 +75,10 @@ func init() {
 	utilruntime.Must(volcanoscheme.AddToScheme(scheme))
 
 	utilruntime.Must(grovev1alpha1.AddToScheme(scheme))
+
+	utilruntime.Must(apiextensionsv1.AddToScheme(scheme))
+
+	utilruntime.Must(istioclientsetscheme.AddToScheme(scheme))
 	//+kubebuilder:scaffold:scheme
 }
 
@@ -92,7 +98,7 @@ func main() {
 	var ingressControllerTLSSecretName string
 	var ingressHostSuffix string
 	var enableLWS bool
-	var enableGrove bool
+	var groveTerminationDelay time.Duration
 	flag.StringVar(&metricsAddr, "metrics-bind-address", ":8080", "The address the metric endpoint binds to.")
 	flag.StringVar(&probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.")
 	flag.BoolVar(&enableLeaderElection, "leader-elect", false,
@@ -120,22 +126,23 @@ func main() {
 		"The suffix to use for the ingress host")
 	flag.BoolVar(&enableLWS, "enable-lws", false,
 		"If set, enable leader worker set")
-	flag.BoolVar(&enableGrove, "enable-grove", false,
-		"If set, enable grove")
+	flag.DurationVar(&groveTerminationDelay, "grove-termination-delay", consts.DefaultGroveTerminationDelay,
+		"The termination delay for Grove PodGangSets")
 	opts := zap.Options{
 		Development: true,
 	}
 	opts.BindFlags(flag.CommandLine)
 	flag.Parse()
 
-	utilruntime.Must(istioclientsetscheme.AddToScheme(scheme))
-
 	ctrlConfig := commonController.Config{
 		RestrictedNamespace: restrictedNamespace,
 		EnableLWS:           enableLWS,
-		EnableGrove:         enableGrove,
-		EtcdAddress:         etcdAddr,
-		NatsAddress:         natsAddr,
+		Grove: commonController.GroveConfig{
+			Enabled:          false, // Will be set after Grove discovery
+			TerminationDelay: groveTerminationDelay,
+		},
+		EtcdAddress: etcdAddr,
+		NatsAddress: natsAddr,
 		IngressConfig: commonController.IngressConfig{
 			VirtualServiceGateway:      istioVirtualServiceGateway,
 			IngressControllerClassName: ingressControllerClassName,
@@ -201,6 +208,11 @@ func main() {
 		os.Exit(1)
 	}
 
+	// Detect Grove availability using discovery client
+	setupLog.Info("Detecting Grove availability...")
+	groveEnabled := commonController.DetectGroveAvailability(mainCtx, mgr)
+	ctrlConfig.Grove.Enabled = groveEnabled
+
 	// Create etcd client
 	cli, err := clientv3.New(clientv3.Config{
 		Endpoints:            []string{etcdAddr},

deploy/cloud/operator/internal/consts/consts.go

@@ -1,5 +1,7 @@
 package consts
 
+import "time"
+
 const (
 	HPACPUDefaultAverageUtilization = 80
 
@@ -37,4 +39,6 @@ const (
 	PlannerServiceAccountName = "planner-serviceaccount"
 
 	DefaultIngressSuffix = "local"
+
+	DefaultGroveTerminationDelay = 15 * time.Minute
 )

deploy/cloud/operator/internal/controller/dynamographdeployment_controller.go

@@ -144,7 +144,7 @@ type Resource interface {
 
 func (r *DynamoGraphDeploymentReconciler) reconcileResources(ctx context.Context, dynamoDeployment *nvidiacomv1alpha1.DynamoGraphDeployment) (State, Reason, Message, error) {
 	logger := log.FromContext(ctx)
-	if r.Config.EnableGrove {
+	if r.Config.Grove.Enabled {
 		// check if explicit opt out of grove
 		if dynamoDeployment.Annotations[consts.KubeAnnotationEnableGrove] == consts.KubeLabelValueFalse {
 			logger.Info("Grove is explicitly disabled for this deployment, skipping grove resources reconciliation")
@@ -308,7 +308,7 @@ func (r *DynamoGraphDeploymentReconciler) SetupWithManager(mgr ctrl.Manager) err
 			GenericFunc: func(ge event.GenericEvent) bool { return true },
 		})).
 		WithEventFilter(commonController.EphemeralDeploymentEventFilter(r.Config))
-	if r.Config.EnableGrove {
+	if r.Config.Grove.Enabled {
 		ctrlBuilder = ctrlBuilder.Owns(&grovev1alpha1.PodGangSet{}, builder.WithPredicates(predicate.Funcs{
 			// ignore creation cause we don't want to be called again after we create the pod gang set
 			CreateFunc:  func(ce event.CreateEvent) bool { return false },

deploy/cloud/operator/internal/controller_common/predicate.go

@@ -20,18 +20,28 @@ package controller_common
 import (
 	"context"
 	"strings"
+	"time"
 
 	"k8s.io/apimachinery/pkg/api/meta"
+	"k8s.io/client-go/discovery"
+	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/log"
 	"sigs.k8s.io/controller-runtime/pkg/predicate"
 )
 
+type GroveConfig struct {
+	// Enabled is automatically determined by checking if Grove CRDs are installed in the cluster
+	Enabled bool
+	// TerminationDelay configures the termination delay for Grove PodGangSets
+	TerminationDelay time.Duration
+}
+
 type Config struct {
 	// Enable resources filtering, only the resources belonging to the given namespace will be handled.
 	RestrictedNamespace string
 	EnableLWS           bool
-	EnableGrove         bool
+	Grove               GroveConfig
 	EtcdAddress         string
 	NatsAddress         string
 	IngressConfig       IngressConfig
@@ -48,6 +58,43 @@ func (i *IngressConfig) UseVirtualService() bool {
 	return i.VirtualServiceGateway != ""
 }
 
+// DetectGroveAvailability checks if Grove is available by checking if the Grove API group is registered
+// This approach uses the discovery client which is simpler and more reliable
+func DetectGroveAvailability(ctx context.Context, mgr ctrl.Manager) bool {
+	logger := log.FromContext(ctx)
+
+	// Use the discovery client to check if Grove API groups are available
+	cfg := mgr.GetConfig()
+	if cfg == nil {
+		logger.Info("Grove detection failed, no discovery client available")
+		return false
+	}
+
+	// Try to create a discovery client
+	discoveryClient, err := discovery.NewDiscoveryClientForConfig(cfg)
+	if err != nil {
+		logger.Error(err, "Grove detection failed, could not create discovery client")
+		return false
+	}
+
+	// Check if grove.io API group is available
+	apiGroups, err := discoveryClient.ServerGroups()
+	if err != nil {
+		logger.Error(err, "Grove detection failed, could not list server groups")
+		return false
+	}
+
+	for _, group := range apiGroups.Groups {
+		if group.Name == "grove.io" {
+			logger.Info("Grove is available, grove.io API group found")
+			return true
+		}
+	}
+
+	logger.Info("Grove not available, grove.io API group not found")
+	return false
+}
+
 func EphemeralDeploymentEventFilter(config Config) predicate.Predicate {
 	return predicate.NewPredicateFuncs(func(o client.Object) bool {
 		l := log.FromContext(context.Background())

deploy/cloud/operator/internal/dynamo/graph.go

@@ -316,6 +316,9 @@ func GenerateGrovePodGangSet(ctx context.Context, dynamoDeployment *v1alpha1.Dyn
 	gangSet.Name = dynamoDeployment.Name
 	gangSet.Namespace = dynamoDeployment.Namespace
 	gangSet.Spec.Replicas = 1
+	if controllerConfig.Grove.TerminationDelay > 0 {
+		gangSet.Spec.Template.TerminationDelay = &metav1.Duration{Duration: controllerConfig.Grove.TerminationDelay}
+	}
 	for componentName, component := range dynamoDeployment.Spec.Services {
 		container := corev1.Container{
 			Name:           "main",

deploy/cloud/operator/internal/dynamo/graph_test.go

@@ -23,6 +23,7 @@ import (
 	"reflect"
 	"sort"
 	"testing"
+	"time"
 
 	grovev1alpha1 "github.com/NVIDIA/grove/operator/api/core/v1alpha1"
 	"github.com/ai-dynamo/dynamo/deploy/cloud/operator/api/dynamo/common"
@@ -1136,6 +1137,9 @@ func TestGenerateGrovePodGangSet(t *testing.T) {
 				controllerConfig: controller_common.Config{
 					EtcdAddress: "etcd-address",
 					NatsAddress: "nats-address",
+					Grove: controller_common.GroveConfig{
+						TerminationDelay: 15 * time.Minute,
+					},
 				},
 				dynamoDeployment: &v1alpha1.DynamoGraphDeployment{
 					ObjectMeta: metav1.ObjectMeta{
@@ -1272,6 +1276,7 @@ func TestGenerateGrovePodGangSet(t *testing.T) {
 				Spec: grovev1alpha1.PodGangSetSpec{
 					Replicas: 1,
 					Template: grovev1alpha1.PodGangSetTemplateSpec{
+						TerminationDelay: &metav1.Duration{Duration: 15 * time.Minute},
 						Cliques: []*grovev1alpha1.PodCliqueTemplateSpec{
 							{
 								Name: "frontend",