feat(operator): DynamoGraphDeployment rollout restart mechanism (#5118)

deploy/helm/charts/crds/templates/nvidia.com_dynamocomponentdeployments.yaml

@@ -10335,8 +10335,6 @@ spec:
               properties:
                 conditions:
                   description: |-
-                    INSERT ADDITIONAL STATUS FIELD - define observed state of cluster
-                    Important: Run "make" to regenerate code after modifying this file
                     Conditions captures the latest observed state of the component (including
                     availability and readiness) using standard Kubernetes condition types.
                   items:
@@ -10393,6 +10391,10 @@ spec:
                       - type
                     type: object
                   type: array
+                observedGeneration:
+                  description: ObservedGeneration is the most recent generation observed by the controller.
+                  format: int64
+                  type: integer
                 podSelector:
                   additionalProperties:
                     type: string

deploy/helm/charts/crds/templates/nvidia.com_dynamographdeployments.yaml

@@ -208,6 +208,34 @@ spec:
                         rule: '!has(self.create) || self.create == false || (has(self.size) && has(self.storageClass) && has(self.volumeAccessMode))'
                   maxItems: 100
                   type: array
+                restart:
+                  description: Restart specifies the restart policy for the graph deployment.
+                  properties:
+                    id:
+                      description: |-
+                        ID is an arbitrary string that triggers a restart when changed.
+                        Any modification to this value will initiate a restart of the graph deployment according to the strategy.
+                      minLength: 1
+                      type: string
+                    strategy:
+                      description: Strategy specifies the restart strategy for the graph deployment.
+                      properties:
+                        order:
+                          description: Order specifies the order in which the services should be restarted.
+                          items:
+                            type: string
+                          type: array
+                        type:
+                          default: Sequential
+                          description: Type specifies the restart strategy type.
+                          enum:
+                            - Sequential
+                            - Parallel
+                          type: string
+                      type: object
+                  required:
+                    - id
+                  type: object
                 services:
                   additionalProperties:
                     properties:
@@ -10530,6 +10558,23 @@ spec:
                       - type
                     type: object
                   type: array
+                restart:
+                  description: Restart contains the status of the restart of the graph deployment.
+                  properties:
+                    inProgress:
+                      description: InProgress contains the names of the services that are currently being restarted.
+                      items:
+                        type: string
+                      type: array
+                    observedID:
+                      description: |-
+                        ObservedID is the restart ID that has been observed and is being processed.
+                        Matches the Restart.ID field in the spec.
+                      type: string
+                    phase:
+                      description: Phase is the phase of the restart.
+                      type: string
+                  type: object
                 services:
                   additionalProperties:
                     description: ServiceReplicaStatus contains replica information for a single service.

deploy/operator/api/v1alpha1/dynamocomponentdeployment_types.go

@@ -172,6 +172,11 @@ func (i *IngressSpec) IsVirtualServiceEnabled() bool {
 type DynamoComponentDeploymentStatus struct {
 	// INSERT ADDITIONAL STATUS FIELD - define observed state of cluster
 	// Important: Run "make" to regenerate code after modifying this file
+
+	// ObservedGeneration is the most recent generation observed by the controller.
+	// +optional
+	ObservedGeneration int64 `json:"observedGeneration,omitempty"`
+
 	// Conditions captures the latest observed state of the component (including
 	// availability and readiness) using standard Kubernetes condition types.
 	Conditions []metav1.Condition `json:"conditions"`

deploy/operator/api/v1alpha1/dynamographdeployment_types.go

@@ -60,8 +60,42 @@ type DynamoGraphDeploymentSpec struct {
 	// BackendFramework specifies the backend framework (e.g., "sglang", "vllm", "trtllm").
 	// +kubebuilder:validation:Enum=sglang;vllm;trtllm
 	BackendFramework string `json:"backendFramework,omitempty"`
+
+	// Restart specifies the restart policy for the graph deployment.
+	// +kubebuilder:validation:Optional
+	Restart *Restart `json:"restart,omitempty"`
+}
+
+type Restart struct {
+	// ID is an arbitrary string that triggers a restart when changed.
+	// Any modification to this value will initiate a restart of the graph deployment according to the strategy.
+	// +kubebuilder:validation:Required
+	// +kubebuilder:validation:MinLength=1
+	ID string `json:"id"`
+
+	// Strategy specifies the restart strategy for the graph deployment.
+	// +kubebuilder:validation:Optional
+	Strategy *RestartStrategy `json:"strategy,omitempty"`
 }
 
+type RestartStrategy struct {
+	// Type specifies the restart strategy type.
+	// +kubebuilder:validation:Enum=Sequential;Parallel
+	// +kubebuilder:default=Sequential
+	Type RestartStrategyType `json:"type,omitempty"`
+
+	// Order specifies the order in which the services should be restarted.
+	// +kubebuilder:validation:Optional
+	Order []string `json:"order,omitempty"`
+}
+
+type RestartStrategyType string
+
+const (
+	RestartStrategyTypeSequential RestartStrategyType = "Sequential"
+	RestartStrategyTypeParallel   RestartStrategyType = "Parallel"
+)
+
 // DynamoGraphDeploymentStatus defines the observed state of DynamoGraphDeployment.
 type DynamoGraphDeploymentStatus struct {
 	// State is a high-level textual status of the graph deployment lifecycle.
@@ -73,8 +107,33 @@ type DynamoGraphDeploymentStatus struct {
 	// The map key is the service name from spec.services.
 	// +optional
 	Services map[string]ServiceReplicaStatus `json:"services,omitempty"`
+
+	// Restart contains the status of the restart of the graph deployment.
+	// +optional
+	Restart *RestartStatus `json:"restart,omitempty"`
 }
 
+// RestartStatus contains the status of the restart of the graph deployment.
+type RestartStatus struct {
+	// ObservedID is the restart ID that has been observed and is being processed.
+	// Matches the Restart.ID field in the spec.
+	ObservedID string `json:"observedID,omitempty"`
+	// Phase is the phase of the restart.
+	Phase RestartPhase `json:"phase,omitempty"`
+	// InProgress contains the names of the services that are currently being restarted.
+	// +optional
+	InProgress []string `json:"inProgress,omitempty"`
+}
+
+type RestartPhase string
+
+const (
+	RestartPhasePending    RestartPhase = "Pending"
+	RestartPhaseRestarting RestartPhase = "Restarting"
+	RestartPhaseCompleted  RestartPhase = "Completed"
+	RestartPhaseFailed     RestartPhase = "Failed"
+)
+
 // ServiceReplicaStatus contains replica information for a single service.
 type ServiceReplicaStatus struct {
 	// ComponentKind is the underlying resource kind (e.g., "PodClique", "PodCliqueScalingGroup", "Deployment", "LeaderWorkerSet").

deploy/operator/api/v1alpha1/zz_generated.deepcopy.go

@@ -751,6 +751,11 @@ func (in *DynamoGraphDeploymentSpec) DeepCopyInto(out *DynamoGraphDeploymentSpec
 			(*in)[i].DeepCopyInto(&(*out)[i])
 		}
 	}
+	if in.Restart != nil {
+		in, out := &in.Restart, &out.Restart
+		*out = new(Restart)
+		(*in).DeepCopyInto(*out)
+	}
 }
 
 // DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new DynamoGraphDeploymentSpec.
@@ -780,6 +785,11 @@ func (in *DynamoGraphDeploymentStatus) DeepCopyInto(out *DynamoGraphDeploymentSt
 			(*out)[key] = *val.DeepCopy()
 		}
 	}
+	if in.Restart != nil {
+		in, out := &in.Restart, &out.Restart
+		*out = new(RestartStatus)
+		(*in).DeepCopyInto(*out)
+	}
 }
 
 // DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new DynamoGraphDeploymentStatus.
@@ -1211,6 +1221,66 @@ func (in *Resources) DeepCopy() *Resources {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *Restart) DeepCopyInto(out *Restart) {
+	*out = *in
+	if in.Strategy != nil {
+		in, out := &in.Strategy, &out.Strategy
+		*out = new(RestartStrategy)
+		(*in).DeepCopyInto(*out)
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Restart.
+func (in *Restart) DeepCopy() *Restart {
+	if in == nil {
+		return nil
+	}
+	out := new(Restart)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *RestartStatus) DeepCopyInto(out *RestartStatus) {
+	*out = *in
+	if in.InProgress != nil {
+		in, out := &in.InProgress, &out.InProgress
+		*out = make([]string, len(*in))
+		copy(*out, *in)
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new RestartStatus.
+func (in *RestartStatus) DeepCopy() *RestartStatus {
+	if in == nil {
+		return nil
+	}
+	out := new(RestartStatus)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *RestartStrategy) DeepCopyInto(out *RestartStrategy) {
+	*out = *in
+	if in.Order != nil {
+		in, out := &in.Order, &out.Order
+		*out = make([]string, len(*in))
+		copy(*out, *in)
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new RestartStrategy.
+func (in *RestartStrategy) DeepCopy() *RestartStrategy {
+	if in == nil {
+		return nil
+	}
+	out := new(RestartStrategy)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *ScalingAdapter) DeepCopyInto(out *ScalingAdapter) {
 	*out = *in

deploy/operator/config/crd/bases/nvidia.com_dynamocomponentdeployments.yaml

@@ -10335,8 +10335,6 @@ spec:
               properties:
                 conditions:
                   description: |-
-                    INSERT ADDITIONAL STATUS FIELD - define observed state of cluster
-                    Important: Run "make" to regenerate code after modifying this file
                     Conditions captures the latest observed state of the component (including
                     availability and readiness) using standard Kubernetes condition types.
                   items:
@@ -10393,6 +10391,10 @@ spec:
                       - type
                     type: object
                   type: array
+                observedGeneration:
+                  description: ObservedGeneration is the most recent generation observed by the controller.
+                  format: int64
+                  type: integer
                 podSelector:
                   additionalProperties:
                     type: string

deploy/operator/config/crd/bases/nvidia.com_dynamographdeployments.yaml

@@ -208,6 +208,34 @@ spec:
                         rule: '!has(self.create) || self.create == false || (has(self.size) && has(self.storageClass) && has(self.volumeAccessMode))'
                   maxItems: 100
                   type: array
+                restart:
+                  description: Restart specifies the restart policy for the graph deployment.
+                  properties:
+                    id:
+                      description: |-
+                        ID is an arbitrary string that triggers a restart when changed.
+                        Any modification to this value will initiate a restart of the graph deployment according to the strategy.
+                      minLength: 1
+                      type: string
+                    strategy:
+                      description: Strategy specifies the restart strategy for the graph deployment.
+                      properties:
+                        order:
+                          description: Order specifies the order in which the services should be restarted.
+                          items:
+                            type: string
+                          type: array
+                        type:
+                          default: Sequential
+                          description: Type specifies the restart strategy type.
+                          enum:
+                            - Sequential
+                            - Parallel
+                          type: string
+                      type: object
+                  required:
+                    - id
+                  type: object
                 services:
                   additionalProperties:
                     properties:
@@ -10530,6 +10558,23 @@ spec:
                       - type
                     type: object
                   type: array
+                restart:
+                  description: Restart contains the status of the restart of the graph deployment.
+                  properties:
+                    inProgress:
+                      description: InProgress contains the names of the services that are currently being restarted.
+                      items:
+                        type: string
+                      type: array
+                    observedID:
+                      description: |-
+                        ObservedID is the restart ID that has been observed and is being processed.
+                        Matches the Restart.ID field in the spec.
+                      type: string
+                    phase:
+                      description: Phase is the phase of the restart.
+                      type: string
+                  type: object
                 services:
                   additionalProperties:
                     description: ServiceReplicaStatus contains replica information for a single service.

deploy/operator/internal/consts/consts.go

@@ -95,6 +95,8 @@ const (
 	GroveRoleSuffixWorker = "wkr"
 
 	MainContainerName = "main"
+
+	RestartAnnotation = "nvidia.com/restartAt"
 )
 
 type MultinodeDeploymentType string

deploy/operator/internal/controller/dynamocomponentdeployment_controller.go

@@ -307,6 +307,16 @@ func (r *DynamoComponentDeploymentReconciler) reconcileDeploymentResources(ctx c
 		return ComponentReconcileResult{}, fmt.Errorf("failed to create or update the deployment: %w", err)
 	}
 
+	logger.V(1).Info("Deployment sync completed",
+		"deploymentModified", deploymentModified,
+		"deploymentName", deployment.Name,
+		"deploymentGeneration", deployment.Generation,
+		"deploymentObservedGeneration", deployment.Status.ObservedGeneration,
+		"deploymentReplicas", deployment.Status.Replicas,
+		"deploymentUpdatedReplicas", deployment.Status.UpdatedReplicas,
+		"deploymentAvailableReplicas", deployment.Status.AvailableReplicas,
+		"deploymentReadyReplicas", deployment.Status.ReadyReplicas)
+
 	serviceReplicaStatus := &v1alpha1.ServiceReplicaStatus{
 		ComponentKind:     v1alpha1.ComponentKindDeployment,
 		ComponentName:     deployment.Name,
@@ -317,7 +327,6 @@ func (r *DynamoComponentDeploymentReconciler) reconcileDeploymentResources(ctx c
 	}
 
 	if IsDeploymentReady(deployment) {
-		logger.Info("Deployment is ready. Setting available status condition to true.")
 		return ComponentReconcileResult{
 			modified:             deploymentModified,
 			status:               metav1.ConditionTrue,
@@ -458,6 +467,7 @@ func (r *DynamoComponentDeploymentReconciler) setStatusConditionAndServiceReplic
 
 	meta.SetStatusCondition(&dynamoComponentDeployment.Status.Conditions, condition)
 	dynamoComponentDeployment.Status.Service = componentReconcileResult.serviceReplicaStatus
+	dynamoComponentDeployment.Status.ObservedGeneration = dynamoComponentDeployment.Generation
 
 	err := r.Status().Update(ctx, dynamoComponentDeployment)
 	if err != nil {
@@ -763,9 +773,11 @@ func IsDeploymentReady(deployment *appsv1.Deployment) bool {
 	// 1. ObservedGeneration: Deployment controller has observed the latest configuration
 	// 2. UpdatedReplicas: All replicas have been updated to the latest version
 	// 3. AvailableReplicas: All desired replicas are available (schedulable and healthy)
+	// 4. Replicas: Total replicas equals desired (no surge pods remaining from rolling update)
 	if status.ObservedGeneration < deployment.Generation ||
 		status.UpdatedReplicas < desiredReplicas ||
-		status.AvailableReplicas < desiredReplicas {
+		status.AvailableReplicas < desiredReplicas ||
+		status.Replicas != desiredReplicas {
 		return false
 	}
 	// Finally, check for the DeploymentAvailable condition
@@ -1234,6 +1246,12 @@ func (r *DynamoComponentDeploymentReconciler) generatePodTemplateSpec(ctx contex
 		maps.Copy(podLabels, extraPodMetadata.Labels)
 	}
 
+	// Propagate restart annotation to pod template to trigger rolling restart
+	// This is the same mechanism used by kubectl rollout restart
+	if restartAt, exists := resourceAnnotations[commonconsts.RestartAnnotation]; exists {
+		podAnnotations[commonconsts.RestartAnnotation] = restartAt
+	}
+
 	if podSpec.ServiceAccountName == "" {
 		serviceAccounts := &corev1.ServiceAccountList{}
 		err = r.List(ctx, serviceAccounts, client.InNamespace(opt.dynamoComponentDeployment.Namespace), client.MatchingLabels{

deploy/operator/internal/controller/dynamocomponentdeployment_controller_test.go

@@ -93,6 +93,26 @@ func TestIsDeploymentReady(t *testing.T) {
 			},
 			want: false,
 		},
+		{
+			name: "not ready (surging)",
+			args: args{
+				deployment: &appsv1.Deployment{
+					ObjectMeta: metav1.ObjectMeta{
+						Generation: 1,
+					},
+					Spec: appsv1.DeploymentSpec{
+						Replicas: &[]int32{2}[0],
+					},
+					Status: appsv1.DeploymentStatus{
+						ObservedGeneration: 1,
+						UpdatedReplicas:    1,
+						AvailableReplicas:  1,
+						Replicas:           2,
+					},
+				},
+			},
+			want: false,
+		},
 		{
 			name: "ready",
 			args: args{
@@ -107,6 +127,7 @@ func TestIsDeploymentReady(t *testing.T) {
 						ObservedGeneration: 1,
 						UpdatedReplicas:    1,
 						AvailableReplicas:  1,
+						Replicas:           1,
 						Conditions: []appsv1.DeploymentCondition{
 							{
 								Type:   appsv1.DeploymentAvailable,
@@ -1855,6 +1876,7 @@ func Test_setStatusConditionAndServiceReplicaStatus(t *testing.T) {
 		wantConditionReason      string
 		wantConditionMessage     string
 		wantServiceReplicaStatus *v1alpha1.ServiceReplicaStatus
+		wantObservedGeneration   int64
 	}{
 		{
 			name: "deployment backed DCD that is unready",
@@ -1976,10 +1998,12 @@ func Test_setStatusConditionAndServiceReplicaStatus(t *testing.T) {
 			g.Expect(err).NotTo(gomega.HaveOccurred())
 
 			// Create DynamoComponentDeployment
+			generation := int64(5)
 			dcd := &v1alpha1.DynamoComponentDeployment{
 				ObjectMeta: metav1.ObjectMeta{
-					Name:      "test-component",
-					Namespace: "default",
+					Name:       "test-component",
+					Namespace:  "default",
+					Generation: generation,
 				},
 				Spec: v1alpha1.DynamoComponentDeploymentSpec{
 					BackendFramework: string(dynamo.BackendFrameworkVLLM),
@@ -2031,6 +2055,9 @@ func Test_setStatusConditionAndServiceReplicaStatus(t *testing.T) {
 
 			// Assert the service replica status
 			g.Expect(updatedDCD.Status.Service).To(gomega.Equal(tt.wantServiceReplicaStatus))
+
+			// Assert the observed generation
+			g.Expect(updatedDCD.Status.ObservedGeneration).To(gomega.Equal(generation))
 		})
 	}
 }

deploy/operator/internal/dynamo/graph.go

@@ -44,6 +44,142 @@ import (
 	networkingv1 "k8s.io/api/networking/v1"
 )
 
+// RestartState holds the restart state for DGD services.
+type RestartState struct {
+	// Timestamp is the restart timestamp to apply as the annotation value.
+	// Format: RFC3339
+	Timestamp string
+	// ServicesToAnnotate is the set of service names that should have the restart annotation.
+	ServicesToAnnotate map[string]bool
+}
+
+// ShouldAnnotateService returns true if the given service should have a restart annotation.
+func (s *RestartState) ShouldAnnotateService(serviceName string) bool {
+	if s == nil || s.ServicesToAnnotate == nil {
+		return false
+	}
+	return s.ServicesToAnnotate[serviceName]
+}
+
+// DetermineRestartState computes the restart state for DGD services.
+func DetermineRestartState(dgd *v1alpha1.DynamoGraphDeployment, restartStatus *v1alpha1.RestartStatus) *RestartState {
+	if restartStatus == nil {
+		return nil
+	}
+
+	if dgd.Spec.Restart == nil || dgd.Spec.Restart.ID == "" {
+		// Check if there's a completed restart we need to preserve
+		if restartStatus.ObservedID != "" {
+			return &RestartState{
+				Timestamp:          restartStatus.ObservedID,
+				ServicesToAnnotate: getAllServiceNames(dgd),
+			}
+		}
+		return nil
+	}
+
+	specID := dgd.Spec.Restart.ID
+
+	isNewRestart := restartStatus.ObservedID == "" ||
+		dgd.Spec.Restart.ID != restartStatus.ObservedID
+
+	if !isNewRestart && restartStatus.Phase == v1alpha1.RestartPhaseCompleted {
+		return &RestartState{
+			Timestamp:          specID,
+			ServicesToAnnotate: getAllServiceNames(dgd),
+		}
+	}
+
+	if IsParallelRestart(dgd) {
+		return &RestartState{
+			Timestamp:          specID,
+			ServicesToAnnotate: getAllServiceNames(dgd),
+		}
+	}
+
+	// Sequential restart (default or specified)
+	return &RestartState{
+		Timestamp:          specID,
+		ServicesToAnnotate: getServicesToAnnotateForSequentialRestart(dgd, restartStatus),
+	}
+}
+
+// getAllServiceNames returns a map of all service names in the DGD.
+func getAllServiceNames(dgd *v1alpha1.DynamoGraphDeployment) map[string]bool {
+	services := make(map[string]bool, len(dgd.Spec.Services))
+	for serviceName := range dgd.Spec.Services {
+		services[serviceName] = true
+	}
+	return services
+}
+
+// IsParallelRestart returns true if the restart strategy is parallel.
+func IsParallelRestart(dgd *v1alpha1.DynamoGraphDeployment) bool {
+	if dgd.Spec.Restart == nil || dgd.Spec.Restart.Strategy == nil {
+		return false // Default is sequential
+	}
+	return dgd.Spec.Restart.Strategy.Type == v1alpha1.RestartStrategyTypeParallel
+}
+
+// getServicesToAnnotateForSequentialRestart determines which services should be annotated
+// for a sequential restart in progress.
+func getServicesToAnnotateForSequentialRestart(dgd *v1alpha1.DynamoGraphDeployment, status *v1alpha1.RestartStatus) map[string]bool {
+	services := make(map[string]bool)
+
+	order := GetRestartOrder(dgd)
+	if len(order) == 0 {
+		return services
+	}
+
+	// New restart or Pending phase - only first service needs to be annotated
+	if status == nil ||
+		status.Phase == v1alpha1.RestartPhasePending ||
+		len(status.InProgress) == 0 {
+		services[order[0]] = true
+		return services
+	}
+
+	// Find the max index among in-progress services
+	inProgress := make(map[string]bool)
+	for _, svc := range status.InProgress {
+		inProgress[svc] = true
+	}
+
+	maxIndex := -1
+	for i, svc := range order {
+		if inProgress[svc] {
+			if i > maxIndex {
+				maxIndex = i
+			}
+		}
+	}
+
+	// Add all services up to and including maxIndex
+	// Services before the in-progress one have completed and need their annotation preserved
+	if maxIndex >= 0 {
+		for i := 0; i <= maxIndex; i++ {
+			services[order[i]] = true
+		}
+	}
+
+	return services
+}
+
+// GetRestartOrder returns the order of services for sequential restart.
+// If not specified, returns a deterministic alphabetical order.
+func GetRestartOrder(dgd *v1alpha1.DynamoGraphDeployment) []string {
+	if dgd.Spec.Restart != nil && dgd.Spec.Restart.Strategy != nil && len(dgd.Spec.Restart.Strategy.Order) > 0 {
+		return dgd.Spec.Restart.Strategy.Order
+	}
+
+	order := make([]string, 0, len(dgd.Spec.Services))
+	for serviceName := range dgd.Spec.Services {
+		order = append(order, serviceName)
+	}
+	sort.Strings(order)
+	return order
+}
+
 // ServiceConfig represents the YAML configuration structure for a service
 type DynamoConfig struct {
 	Enabled       bool   `yaml:"enabled"`
@@ -113,7 +249,7 @@ func ParseDynDeploymentConfig(ctx context.Context, jsonContent []byte) (DynDeplo
 }
 
 // GenerateDynamoComponentsDeployments generates a map of DynamoComponentDeployments from a DynamoGraphConfig
-func GenerateDynamoComponentsDeployments(ctx context.Context, parentDynamoGraphDeployment *v1alpha1.DynamoGraphDeployment, defaultIngressSpec *v1alpha1.IngressSpec) (map[string]*v1alpha1.DynamoComponentDeployment, error) {
+func GenerateDynamoComponentsDeployments(ctx context.Context, parentDynamoGraphDeployment *v1alpha1.DynamoGraphDeployment, defaultIngressSpec *v1alpha1.IngressSpec, restartState *RestartState, existingRestartAnnotations map[string]string) (map[string]*v1alpha1.DynamoComponentDeployment, error) {
 	deployments := make(map[string]*v1alpha1.DynamoComponentDeployment)
 	for componentName, component := range parentDynamoGraphDeployment.Spec.Services {
 		dynamoNamespace := getDynamoNamespace(parentDynamoGraphDeployment, component)
@@ -146,6 +282,23 @@ func GenerateDynamoComponentsDeployments(ctx context.Context, parentDynamoGraphD
 			}
 		}
 
+		// Apply restart annotation if this service should be restarted.
+		// For services not in the current restart order, preserve their existing annotation
+		// to avoid triggering unwanted rollouts when a new restart begins.
+		if restartState.ShouldAnnotateService(componentName) {
+			if deployment.Spec.Annotations == nil {
+				deployment.Spec.Annotations = make(map[string]string)
+			}
+			deployment.Spec.Annotations[commonconsts.RestartAnnotation] = restartState.Timestamp
+		} else if existingRestartAnnotations != nil {
+			if existingRestartAt, ok := existingRestartAnnotations[componentName]; ok && existingRestartAt != "" {
+				if deployment.Spec.Annotations == nil {
+					deployment.Spec.Annotations = make(map[string]string)
+				}
+				deployment.Spec.Annotations[commonconsts.RestartAnnotation] = existingRestartAt
+			}
+		}
+
 		if component.ComponentType == commonconsts.ComponentTypePlanner {
 			// ensure that the extraPodSpec is not nil
 			if deployment.Spec.ExtraPodSpec == nil {
@@ -959,6 +1112,8 @@ func GenerateGrovePodCliqueSet(
 	dynamoDeployment *v1alpha1.DynamoGraphDeployment,
 	controllerConfig controller_common.Config,
 	secretsRetriever SecretsRetriever,
+	restartState *RestartState,
+	existingRestartAnnotations map[string]string,
 ) (*grovev1alpha1.PodCliqueSet, error) {
 	gangSet := &grovev1alpha1.PodCliqueSet{}
 	gangSet.Name = dynamoDeployment.Name
@@ -1040,6 +1195,23 @@ func GenerateGrovePodCliqueSet(
 			if err != nil {
 				return nil, fmt.Errorf("failed to generate annotations: %w", err)
 			}
+
+			// Apply restart annotation if this service should be restarted.
+			// For services not in the current restart order, preserve their existing annotation
+			// to avoid triggering unwanted rollouts when a new restart begins.
+			if restartState.ShouldAnnotateService(serviceName) {
+				if annotations == nil {
+					annotations = make(map[string]string)
+				}
+				annotations[commonconsts.RestartAnnotation] = restartState.Timestamp
+			} else if existingRestartAnnotations != nil {
+				if existingTimestamp, ok := existingRestartAnnotations[serviceName]; ok {
+					if annotations == nil {
+						annotations = make(map[string]string)
+					}
+					annotations[commonconsts.RestartAnnotation] = existingTimestamp
+				}
+			}
 			clique.Annotations = annotations
 
 			// Inject kai-scheduler settings if enabled

deploy/operator/internal/dynamo/graph_test.go

@@ -722,7 +722,7 @@ func TestGenerateDynamoComponentsDeployments(t *testing.T) {
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := GenerateDynamoComponentsDeployments(context.Background(), tt.args.parentDynamoGraphDeployment, tt.args.ingressSpec)
+			got, err := GenerateDynamoComponentsDeployments(context.Background(), tt.args.parentDynamoGraphDeployment, tt.args.ingressSpec, nil, nil)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("GenerateDynamoComponentsDeployments() error = %v, wantErr %v", err, tt.wantErr)
 				return
@@ -756,7 +756,7 @@ func Test_GetDynamoComponentDeploymentsGlobalNamespace(t *testing.T) {
 		},
 	}
 
-	got, err := GenerateDynamoComponentsDeployments(context.Background(), dgd, nil)
+	got, err := GenerateDynamoComponentsDeployments(context.Background(), dgd, nil, nil, nil)
 	if !assert.NoError(t, err) {
 		return
 	}
@@ -3548,7 +3548,7 @@ func TestGenerateGrovePodCliqueSet(t *testing.T) {
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := GenerateGrovePodCliqueSet(tt.args.ctx, tt.args.dynamoDeployment, tt.args.controllerConfig, nil)
+			got, err := GenerateGrovePodCliqueSet(tt.args.ctx, tt.args.dynamoDeployment, tt.args.controllerConfig, nil, nil, nil)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("GenerateGrovePodCliqueSet() error = %v, wantErr %v", err, tt.wantErr)
 				return
@@ -3600,7 +3600,7 @@ func Test_GeneratePodCliqueSetGlobalDynamoNamespace(t *testing.T) {
 		},
 	}
 
-	got, err := GenerateGrovePodCliqueSet(context.Background(), dynamoDeployment, controller_common.Config{}, nil)
+	got, err := GenerateGrovePodCliqueSet(context.Background(), dynamoDeployment, controller_common.Config{}, nil, nil, nil)
 	if !assert.NoError(t, err) {
 		return
 	}
@@ -4683,7 +4683,7 @@ func TestGenerateGrovePodCliqueSet_StartsAfterDependencies(t *testing.T) {
 				NatsAddress: "nats-address",
 			}
 
-			got, err := GenerateGrovePodCliqueSet(context.Background(), dynamoDeployment, controllerConfig, secretsRetriever)
+			got, err := GenerateGrovePodCliqueSet(context.Background(), dynamoDeployment, controllerConfig, secretsRetriever, nil, nil)
 			if err != nil {
 				t.Errorf("GenerateGrovePodCliqueSet() error = %v", err)
 				return
@@ -5984,3 +5984,552 @@ func TestGenerateBasePodSpec_SecurityContext(t *testing.T) {
 		})
 	}
 }
+
+func TestDetermineGroveRestartState(t *testing.T) {
+	restartID := "restart-1"
+	oldRestartID := "restart-0"
+
+	tests := []struct {
+		name          string
+		dgd           *v1alpha1.DynamoGraphDeployment
+		restartStatus *v1alpha1.RestartStatus
+		want          *RestartState
+		wantNil       bool
+		wantSvcs      []string // expected services to annotate (sorted)
+		wantTimestamp *string
+	}{
+		{
+			name: "restartStatus nil returns nil",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+				},
+			},
+			wantNil: true,
+		},
+		{
+			name: "spec.restart.at nil and restartStatus.observedAt nil returns nil",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+					Restart: &v1alpha1.Restart{},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: "",
+			},
+			wantNil: true,
+		},
+		{
+			name: "new parallel restart annotates all services",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID,
+						Strategy: &v1alpha1.RestartStrategy{
+							Type: v1alpha1.RestartStrategyTypeParallel,
+						},
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: restartID,
+				Phase:      v1alpha1.RestartPhaseRestarting,
+				InProgress: []string{"Frontend", "Worker"},
+			},
+			wantSvcs:      []string{"Frontend", "Worker"},
+			wantTimestamp: ptr.To(restartID),
+		},
+		{
+			name: "new sequential restart annotates only first service",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID,
+						Strategy: &v1alpha1.RestartStrategy{
+							Type:  v1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"Worker", "Frontend"},
+						},
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: restartID,
+				Phase:      v1alpha1.RestartPhaseRestarting,
+				InProgress: []string{"Worker"},
+			},
+			wantSvcs:      []string{"Worker"},
+			wantTimestamp: ptr.To(restartID),
+		},
+		{
+			name: "sequential restart in progress annotates completed + in-progress",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+						"Backend":  {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID,
+						Strategy: &v1alpha1.RestartStrategy{
+							Type:  v1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"Frontend", "Worker", "Backend"},
+						},
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: restartID,
+				Phase:      v1alpha1.RestartPhaseRestarting,
+				InProgress: []string{"Worker"},
+			},
+			wantSvcs:      []string{"Frontend", "Worker"}, // Frontend completed, Worker in progress
+			wantTimestamp: ptr.To(restartID),
+		},
+		{
+			name: "default restart in progress annotates completed + in-progress",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+						"Backend":  {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID,
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: restartID,
+				Phase:      v1alpha1.RestartPhaseRestarting,
+				InProgress: []string{"Worker"},
+			},
+			wantSvcs:      []string{"Frontend", "Worker", "Backend"},
+			wantTimestamp: ptr.To(restartID),
+		},
+		{
+			name: "completed restart with empty spec restart preserves all annotations",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: oldRestartID,
+				Phase:      v1alpha1.RestartPhaseCompleted,
+			},
+			wantSvcs:      []string{"Frontend", "Worker"},
+			wantTimestamp: ptr.To(oldRestartID),
+		},
+		{
+			name: "completed restart",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID,
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: restartID,
+				Phase:      v1alpha1.RestartPhaseCompleted,
+			},
+			wantSvcs:      []string{"Frontend", "Worker"},
+			wantTimestamp: ptr.To(restartID),
+		},
+		{
+			name: "new restart after completed restart",
+			dgd: &v1alpha1.DynamoGraphDeployment{
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+						"Frontend": {},
+						"Worker":   {},
+					},
+					Restart: &v1alpha1.Restart{
+						ID: restartID, // new time
+						Strategy: &v1alpha1.RestartStrategy{
+							Type: v1alpha1.RestartStrategyTypeParallel,
+						},
+					},
+				},
+			},
+			restartStatus: &v1alpha1.RestartStatus{
+				ObservedID: oldRestartID,
+				Phase:      v1alpha1.RestartPhaseCompleted,
+			},
+			wantSvcs:      []string{"Frontend", "Worker"},
+			wantTimestamp: ptr.To(restartID),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := DetermineRestartState(tt.dgd, tt.restartStatus)
+
+			if tt.wantNil {
+				if got != nil {
+					t.Errorf("DetermineGroveRestartState() = %v, want nil", got)
+				}
+				return
+			}
+
+			if got == nil {
+				t.Errorf("DetermineGroveRestartState() = nil, want non-nil")
+				return
+			}
+
+			var gotSvcs []string
+			for svc, shouldAnnotate := range got.ServicesToAnnotate {
+				if shouldAnnotate {
+					gotSvcs = append(gotSvcs, svc)
+				}
+			}
+			sort.Strings(gotSvcs)
+			sort.Strings(tt.wantSvcs)
+
+			if !reflect.DeepEqual(gotSvcs, tt.wantSvcs) {
+				t.Errorf("DetermineGroveRestartState() services = %v, want %v", gotSvcs, tt.wantSvcs)
+			}
+			if tt.wantTimestamp != nil && (got.Timestamp != *tt.wantTimestamp) {
+				t.Errorf("DetermineGroveRestartState() timestamp = %v, want %v", got.Timestamp, *tt.wantTimestamp)
+			}
+		})
+	}
+}
+
+func TestGroveRestartStateShouldAnnotateService(t *testing.T) {
+	tests := []struct {
+		name        string
+		state       *RestartState
+		serviceName string
+		want        bool
+	}{
+		{
+			name:        "nil state returns false",
+			state:       nil,
+			serviceName: "Frontend",
+			want:        false,
+		},
+		{
+			name: "nil services map returns false",
+			state: &RestartState{
+				Timestamp:          "2024-01-01T00:00:00Z",
+				ServicesToAnnotate: nil,
+			},
+			serviceName: "Frontend",
+			want:        false,
+		},
+		{
+			name: "service in map returns true",
+			state: &RestartState{
+				Timestamp:          "2024-01-01T00:00:00Z",
+				ServicesToAnnotate: map[string]bool{"Frontend": true, "Worker": true},
+			},
+			serviceName: "Frontend",
+			want:        true,
+		},
+		{
+			name: "service not in map returns false",
+			state: &RestartState{
+				Timestamp:          "2024-01-01T00:00:00Z",
+				ServicesToAnnotate: map[string]bool{"Frontend": true},
+			},
+			serviceName: "Worker",
+			want:        false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := tt.state.ShouldAnnotateService(tt.serviceName); got != tt.want {
+				t.Errorf("ShouldAnnotateService() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
+
+func TestGenerateGrovePodCliqueSet_RestartAnnotations(t *testing.T) {
+	restartTimestamp := "2024-01-05T10:00:00Z"
+
+	tests := []struct {
+		name                     string
+		restartState             *RestartState
+		services                 map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec
+		wantAnnotationsPerClique map[string]bool              // clique name -> should have restart annotation
+		wantPreservedAnnotations map[string]map[string]string // clique name -> preserved annotations to verify
+	}{
+		{
+			name:         "nil restartState - no annotations",
+			restartState: nil,
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+				"Worker": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": false,
+				"worker":   false,
+			},
+		},
+		{
+			name: "nil ServicesToAnnotate - no annotations",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: nil,
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": false,
+			},
+		},
+		{
+			name: "all services annotated - parallel restart",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Frontend": true, "Worker": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+				"Worker": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": true,
+				"worker":   true,
+			},
+		},
+		{
+			name: "only first service annotated - sequential restart start",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Frontend": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+				"Worker": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": true,
+				"worker":   false,
+			},
+		},
+		{
+			name: "completed services keep annotation - sequential restart in progress",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Frontend": true, "Worker": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+				"Worker": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(1)),
+				},
+				"Backend": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": true,
+				"worker":   true,
+				"backend":  false,
+			},
+		},
+		{
+			name: "service not in DGD spec - annotation still applied if in ServicesToAnnotate",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Frontend": true, "NonExistent": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": true,
+			},
+		},
+		{
+			name: "multinode service - all cliques get restart annotation",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Worker": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Worker": {
+					ComponentType: commonconsts.ComponentTypeWorker,
+					Replicas:      ptr.To(int32(2)),
+					Multinode: &v1alpha1.MultinodeSpec{
+						NodeCount: 2,
+					},
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"worker-ldr": true,
+				"worker-wkr": true,
+			},
+		},
+		{
+			name: "preserves existing annotations when adding restart annotation",
+			restartState: &RestartState{
+				Timestamp:          restartTimestamp,
+				ServicesToAnnotate: map[string]bool{"Frontend": true},
+			},
+			services: map[string]*v1alpha1.DynamoComponentDeploymentSharedSpec{
+				"Frontend": {
+					ComponentType: commonconsts.ComponentTypeFrontend,
+					Replicas:      ptr.To(int32(1)),
+					Annotations: map[string]string{
+						"custom-annotation": "custom-value",
+						"another-key":       "another-value",
+					},
+				},
+			},
+			wantAnnotationsPerClique: map[string]bool{
+				"frontend": true,
+			},
+			wantPreservedAnnotations: map[string]map[string]string{
+				"frontend": {
+					"custom-annotation": "custom-value",
+					"another-key":       "another-value",
+				},
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			dgd := &v1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-dgd",
+					Namespace: "default",
+				},
+				Spec: v1alpha1.DynamoGraphDeploymentSpec{
+					Services: tt.services,
+				},
+			}
+
+			controllerConfig := controller_common.Config{
+				EtcdAddress: "etcd-address",
+				NatsAddress: "nats-address",
+			}
+
+			got, err := GenerateGrovePodCliqueSet(context.Background(), dgd, controllerConfig, nil, tt.restartState, nil)
+			if err != nil {
+				t.Fatalf("GenerateGrovePodCliqueSet() error = %v", err)
+			}
+
+			// Build a map of clique annotations
+			cliqueAnnotations := make(map[string]map[string]string)
+			for _, clique := range got.Spec.Template.Cliques {
+				cliqueAnnotations[clique.Name] = clique.Annotations
+			}
+
+			// Verify restart annotations per clique
+			for cliqueName, shouldHaveAnnotation := range tt.wantAnnotationsPerClique {
+				annotations := cliqueAnnotations[cliqueName]
+
+				if shouldHaveAnnotation {
+					if annotations == nil {
+						t.Errorf("Clique %q: expected restart annotation, but annotations is nil", cliqueName)
+						continue
+					}
+					restartValue, exists := annotations[commonconsts.RestartAnnotation]
+					if !exists {
+						t.Errorf("Clique %q: expected restart annotation %q, but not found. Annotations: %v",
+							cliqueName, commonconsts.RestartAnnotation, annotations)
+						continue
+					}
+					if restartValue != restartTimestamp {
+						t.Errorf("Clique %q: restart annotation value = %q, want %q",
+							cliqueName, restartValue, restartTimestamp)
+					}
+				} else {
+					if annotations != nil {
+						if _, exists := annotations[commonconsts.RestartAnnotation]; exists {
+							t.Errorf("Clique %q: unexpected restart annotation found", cliqueName)
+						}
+					}
+				}
+			}
+
+			// Verify no unexpected restart annotations on cliques not in wantAnnotationsPerClique
+			for cliqueName, annotations := range cliqueAnnotations {
+				if _, specified := tt.wantAnnotationsPerClique[cliqueName]; !specified {
+					if annotations != nil {
+						if _, exists := annotations[commonconsts.RestartAnnotation]; exists {
+							t.Errorf("Clique %q: unexpected restart annotation found (clique not in wantAnnotationsPerClique)", cliqueName)
+						}
+					}
+				}
+			}
+
+			// Verify preserved annotations
+			for cliqueName, expectedAnnotations := range tt.wantPreservedAnnotations {
+				annotations := cliqueAnnotations[cliqueName]
+				if annotations == nil {
+					t.Errorf("Clique %q: expected preserved annotations, but annotations is nil", cliqueName)
+					continue
+				}
+				for key, expectedValue := range expectedAnnotations {
+					if actualValue, exists := annotations[key]; !exists {
+						t.Errorf("Clique %q: expected preserved annotation %q, but not found", cliqueName, key)
+					} else if actualValue != expectedValue {
+						t.Errorf("Clique %q: preserved annotation %q = %q, want %q",
+							cliqueName, key, actualValue, expectedValue)
+					}
+				}
+			}
+		})
+	}
+}

deploy/operator/internal/dynamo/grove.go

@@ -68,14 +68,14 @@ func GetComponentReadinessAndServiceReplicaStatuses(ctx context.Context, client
 
 		if isMultinode {
 			// Check PodCliqueScalingGroup: spec.replicas == status.availableReplicas
-			ok, reason, serviceStatus := checkPCSGReady(ctx, client, resourceName, dgd.Namespace, logger)
+			ok, reason, serviceStatus := CheckPCSGReady(ctx, client, resourceName, dgd.Namespace, logger)
 			serviceStatuses[serviceName] = serviceStatus
 			if !ok {
 				notReadyComponents = append(notReadyComponents, fmt.Sprintf("pcsg/%s: %s", resourceName, reason))
 			}
 		} else {
 			// Check PodClique: spec.replicas == status.readyReplicas
-			ok, reason, serviceStatus := checkPodCliqueReady(ctx, client, resourceName, dgd.Namespace, logger)
+			ok, reason, serviceStatus := CheckPodCliqueReady(ctx, client, resourceName, dgd.Namespace, logger)
 			serviceStatuses[serviceName] = serviceStatus
 			if !ok {
 				notReadyComponents = append(notReadyComponents, fmt.Sprintf("podclique/%s: %s", resourceName, reason))
@@ -90,8 +90,11 @@ func GetComponentReadinessAndServiceReplicaStatuses(ctx context.Context, client
 	return true, "", serviceStatuses
 }
 
-// checkPodCliqueReady checks if a PodClique has spec.replicas == status.readyReplicas
-func checkPodCliqueReady(ctx context.Context, client client.Client, resourceName, namespace string, logger logr.Logger) (bool, string, v1alpha1.ServiceReplicaStatus) {
+// CheckPodCliqueReady determines if a Grove PodClique is fully ready and available.
+// It checks various status fields to ensure all replicas are available and the PodClique
+// configuration has been fully applied. This is the PodClique equivalent of IsDeploymentReady
+// for standard Kubernetes Deployments.
+func CheckPodCliqueReady(ctx context.Context, client client.Client, resourceName, namespace string, logger logr.Logger) (bool, string, v1alpha1.ServiceReplicaStatus) {
 	podClique := &grovev1alpha1.PodClique{}
 	err := client.Get(ctx, types.NamespacedName{Name: resourceName, Namespace: namespace}, podClique)
 	if err != nil {
@@ -105,6 +108,20 @@ func checkPodCliqueReady(ctx context.Context, client client.Client, resourceName
 
 	desiredReplicas := podClique.Spec.Replicas
 	readyReplicas := podClique.Status.ReadyReplicas
+	updatedReplicas := podClique.Status.UpdatedReplicas
+	replicas := podClique.Status.Replicas
+	observedGeneration := podClique.Status.ObservedGeneration
+	generation := podClique.Generation
+
+	logger.V(1).Info("CheckPodCliqueFullyUpdated",
+		"resourceName", resourceName,
+		"generation", podClique.Generation,
+		"observedGeneration", podClique.Status.ObservedGeneration,
+		"desiredReplicas", desiredReplicas,
+		"readyReplicas", readyReplicas,
+		"updatedReplicas", updatedReplicas,
+		"replicas", replicas,
+	)
 
 	serviceStatus := v1alpha1.ServiceReplicaStatus{
 		ComponentKind:   v1alpha1.ComponentKindPodClique,
@@ -114,8 +131,17 @@ func checkPodCliqueReady(ctx context.Context, client client.Client, resourceName
 		ReadyReplicas:   &readyReplicas,
 	}
 
+	if observedGeneration == nil {
+		logger.V(1).Info("PodClique observedGeneration is nil", "resourceName", resourceName)
+		return false, "observedGeneration is nil", serviceStatus
+	}
+
+	if observedGeneration != nil && *observedGeneration < generation {
+		logger.V(1).Info("PodClique spec not yet processed", "resourceName", resourceName, "generation", generation, "observedGeneration", observedGeneration)
+		return false, fmt.Sprintf("spec not yet processed: generation=%d, observedGeneration=%d", generation, *observedGeneration), serviceStatus
+	}
+
 	if desiredReplicas == 0 {
-		// No replicas desired, so it's ready
 		return true, "", serviceStatus
 	}
 
@@ -124,11 +150,24 @@ func checkPodCliqueReady(ctx context.Context, client client.Client, resourceName
 		return false, fmt.Sprintf("desired=%d, ready=%d", desiredReplicas, readyReplicas), serviceStatus
 	}
 
+	if desiredReplicas != updatedReplicas {
+		logger.V(1).Info("PodClique not fully updated", "resourceName", resourceName, "desired", desiredReplicas, "updated", updatedReplicas)
+		return false, fmt.Sprintf("desired=%d, updated=%d", desiredReplicas, updatedReplicas), serviceStatus
+	}
+
+	if replicas != desiredReplicas {
+		logger.V(1).Info("PodClique performing rolling update", "resourceName", resourceName, "desired", desiredReplicas, "replicas", replicas)
+		return false, fmt.Sprintf("performing rolling update: desired=%d, replicas=%d", desiredReplicas, replicas), serviceStatus
+	}
+
 	return true, "", serviceStatus
 }
 
-// checkPCSGReady checks if a PodCliqueScalingGroup has spec.replicas == status.availableReplicas
-func checkPCSGReady(ctx context.Context, client client.Client, resourceName, namespace string, logger logr.Logger) (bool, string, v1alpha1.ServiceReplicaStatus) {
+// CheckPCSGReady determines if a Grove PodCliqueScalingGroup is fully ready and available.
+// It checks various status fields to ensure all replicas are available and the PodClique
+// configuration has been fully applied. This is the PodCliqueScalingGroup equivalent of IsDeploymentReady
+// for standard Kubernetes Deployments.
+func CheckPCSGReady(ctx context.Context, client client.Client, resourceName, namespace string, logger logr.Logger) (bool, string, v1alpha1.ServiceReplicaStatus) {
 	pcsg := &grovev1alpha1.PodCliqueScalingGroup{}
 	err := client.Get(ctx, types.NamespacedName{Name: resourceName, Namespace: namespace}, pcsg)
 	if err != nil {
@@ -142,6 +181,20 @@ func checkPCSGReady(ctx context.Context, client client.Client, resourceName, nam
 
 	desiredReplicas := pcsg.Spec.Replicas
 	availableReplicas := pcsg.Status.AvailableReplicas
+	updatedReplicas := pcsg.Status.UpdatedReplicas
+	replicas := pcsg.Status.Replicas
+	observedGeneration := pcsg.Status.ObservedGeneration
+	generation := pcsg.Generation
+
+	logger.V(1).Info("CheckPCSGFullyUpdated",
+		"resourceName", resourceName,
+		"generation", pcsg.Generation,
+		"observedGeneration", pcsg.Status.ObservedGeneration,
+		"desiredReplicas", desiredReplicas,
+		"availableReplicas", availableReplicas,
+		"updatedReplicas", updatedReplicas,
+		"replicas", replicas,
+	)
 
 	serviceStatus := v1alpha1.ServiceReplicaStatus{
 		ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
@@ -151,6 +204,16 @@ func checkPCSGReady(ctx context.Context, client client.Client, resourceName, nam
 		AvailableReplicas: &availableReplicas,
 	}
 
+	if observedGeneration == nil {
+		logger.V(1).Info("PodCliqueScalingGroup observedGeneration is nil", "resourceName", resourceName)
+		return false, "observedGeneration is nil", serviceStatus
+	}
+
+	if observedGeneration != nil && *observedGeneration < generation {
+		logger.V(1).Info("PodCliqueScalingGroup spec not yet processed", "resourceName", resourceName, "generation", generation, "observedGeneration", observedGeneration)
+		return false, fmt.Sprintf("spec not yet processed: generation=%d, observedGeneration=%d", generation, *observedGeneration), serviceStatus
+	}
+
 	if desiredReplicas == 0 {
 		// No replicas desired, so it's ready
 		return true, "", serviceStatus
@@ -161,6 +224,16 @@ func checkPCSGReady(ctx context.Context, client client.Client, resourceName, nam
 		return false, fmt.Sprintf("desired=%d, available=%d", desiredReplicas, availableReplicas), serviceStatus
 	}
 
+	if desiredReplicas != updatedReplicas {
+		logger.V(1).Info("PodCliqueScalingGroup not fully updated", "resourceName", resourceName, "desired", desiredReplicas, "updated", updatedReplicas)
+		return false, fmt.Sprintf("desired=%d, updated=%d", desiredReplicas, updatedReplicas), serviceStatus
+	}
+
+	if replicas != desiredReplicas {
+		logger.V(1).Info("PodCliqueScalingGroup performing rolling update", "resourceName", resourceName, "desired", desiredReplicas, "replicas", replicas)
+		return false, fmt.Sprintf("performing rolling update: desired=%d, replicas=%d", desiredReplicas, replicas), serviceStatus
+	}
+
 	return true, "", serviceStatus
 }
 

deploy/operator/internal/dynamo/grove_test.go

@@ -20,6 +20,7 @@ import (
 	"k8s.io/utils/ptr"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+	"sigs.k8s.io/controller-runtime/pkg/log"
 )
 
 func TestResolveKaiSchedulerQueueName(t *testing.T) {
@@ -318,6 +319,536 @@ func TestEnsureQueueExists(t *testing.T) {
 	}
 }
 
+func TestCheckPodCliqueReady(t *testing.T) {
+	ctx := context.Background()
+
+	tests := []struct {
+		name               string
+		resourceName       string
+		namespace          string
+		existingPodClique  *grovev1alpha1.PodClique
+		wantReady          bool
+		wantReasonContains string
+		wantServiceStatus  v1alpha1.ServiceReplicaStatus
+	}{
+		{
+			name:               "PodClique not found",
+			resourceName:       "missing-podclique",
+			namespace:          "default",
+			wantReady:          false,
+			wantReasonContains: "resource not found",
+			wantServiceStatus:  v1alpha1.ServiceReplicaStatus{},
+		},
+		{
+			name:         "PodClique fully ready",
+			resourceName: "ready-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "ready-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           3,
+					ReadyReplicas:      3,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady: true,
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "ready-podclique",
+				Replicas:        3,
+				UpdatedReplicas: 3,
+				ReadyReplicas:   ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PodClique with zero replicas desired",
+			resourceName: "zero-replicas-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "zero-replicas-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 0,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           0,
+					ReadyReplicas:      0,
+					UpdatedReplicas:    0,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady: true,
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "zero-replicas-podclique",
+				Replicas:        0,
+				UpdatedReplicas: 0,
+				ReadyReplicas:   ptr.To(int32(0)),
+			},
+		},
+		{
+			name:         "PodClique spec not yet processed - observedGeneration < generation",
+			resourceName: "stale-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "stale-podclique",
+					Namespace:  "default",
+					Generation: 3,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 2,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           2,
+					ReadyReplicas:      2,
+					UpdatedReplicas:    2,
+					ObservedGeneration: ptr.To(int64(2)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "spec not yet processed",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "stale-podclique",
+				Replicas:        2,
+				UpdatedReplicas: 2,
+				ReadyReplicas:   ptr.To(int32(2)),
+			},
+		},
+		{
+			name:         "PodClique not ready - ready replicas less than desired",
+			resourceName: "not-ready-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "not-ready-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           3,
+					ReadyReplicas:      1,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "desired=3, ready=1",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "not-ready-podclique",
+				Replicas:        3,
+				UpdatedReplicas: 3,
+				ReadyReplicas:   ptr.To(int32(1)),
+			},
+		},
+		{
+			name:         "PodClique not fully updated - updated replicas less than desired",
+			resourceName: "not-updated-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "not-updated-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           3,
+					ReadyReplicas:      3,
+					UpdatedReplicas:    2,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "desired=3, updated=2",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "not-updated-podclique",
+				Replicas:        3,
+				UpdatedReplicas: 2,
+				ReadyReplicas:   ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PodClique performing rolling update - replicas != desired",
+			resourceName: "rolling-update-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "rolling-update-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           4,
+					ReadyReplicas:      3,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "performing rolling update",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "rolling-update-podclique",
+				Replicas:        4,
+				UpdatedReplicas: 3,
+				ReadyReplicas:   ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PodClique with nil observedGeneration",
+			resourceName: "nil-observed-gen-podclique",
+			namespace:    "default",
+			existingPodClique: &grovev1alpha1.PodClique{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "nil-observed-gen-podclique",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueSpec{
+					Replicas: 2,
+				},
+				Status: grovev1alpha1.PodCliqueStatus{
+					Replicas:           2,
+					ReadyReplicas:      2,
+					UpdatedReplicas:    2,
+					ObservedGeneration: nil,
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "observedGeneration is nil",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:   v1alpha1.ComponentKindPodClique,
+				ComponentName:   "nil-observed-gen-podclique",
+				Replicas:        2,
+				UpdatedReplicas: 2,
+				ReadyReplicas:   ptr.To(int32(2)),
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			g := gomega.NewGomegaWithT(t)
+
+			s := scheme.Scheme
+			err := v1alpha1.AddToScheme(s)
+			g.Expect(err).NotTo(gomega.HaveOccurred())
+			err = grovev1alpha1.AddToScheme(s)
+			g.Expect(err).NotTo(gomega.HaveOccurred())
+
+			var objects []client.Object
+			if tt.existingPodClique != nil {
+				objects = append(objects, tt.existingPodClique)
+			}
+
+			fakeKubeClient := fake.NewClientBuilder().
+				WithScheme(s).
+				WithObjects(objects...).
+				WithStatusSubresource(objects...).
+				Build()
+
+			logger := log.FromContext(ctx)
+			ready, reason, serviceStatus := CheckPodCliqueReady(ctx, fakeKubeClient, tt.resourceName, tt.namespace, logger)
+
+			g.Expect(ready).To(gomega.Equal(tt.wantReady))
+			if tt.wantReasonContains != "" {
+				g.Expect(reason).To(gomega.ContainSubstring(tt.wantReasonContains))
+			} else {
+				g.Expect(reason).To(gomega.Equal(""))
+			}
+			g.Expect(serviceStatus).To(gomega.Equal(tt.wantServiceStatus))
+		})
+	}
+}
+
+func TestCheckPCSGReady(t *testing.T) {
+	ctx := context.Background()
+
+	tests := []struct {
+		name               string
+		resourceName       string
+		namespace          string
+		existingPCSG       *grovev1alpha1.PodCliqueScalingGroup
+		wantReady          bool
+		wantReasonContains string
+		wantServiceStatus  v1alpha1.ServiceReplicaStatus
+	}{
+		{
+			name:               "PCSG not found",
+			resourceName:       "missing-pcsg",
+			namespace:          "default",
+			wantReady:          false,
+			wantReasonContains: "resource not found",
+			wantServiceStatus:  v1alpha1.ServiceReplicaStatus{},
+		},
+		{
+			name:         "PCSG fully ready",
+			resourceName: "ready-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "ready-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           3,
+					AvailableReplicas:  3,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady: true,
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "ready-pcsg",
+				Replicas:          3,
+				UpdatedReplicas:   3,
+				AvailableReplicas: ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PCSG with zero replicas desired",
+			resourceName: "zero-replicas-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "zero-replicas-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 0,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           0,
+					AvailableReplicas:  0,
+					UpdatedReplicas:    0,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady: true,
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "zero-replicas-pcsg",
+				Replicas:          0,
+				UpdatedReplicas:   0,
+				AvailableReplicas: ptr.To(int32(0)),
+			},
+		},
+		{
+			name:         "PCSG spec not yet processed - observedGeneration < generation",
+			resourceName: "stale-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "stale-pcsg",
+					Namespace:  "default",
+					Generation: 3,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 2,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           2,
+					AvailableReplicas:  2,
+					UpdatedReplicas:    2,
+					ObservedGeneration: ptr.To(int64(2)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "spec not yet processed",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "stale-pcsg",
+				Replicas:          2,
+				UpdatedReplicas:   2,
+				AvailableReplicas: ptr.To(int32(2)),
+			},
+		},
+		{
+			name:         "PCSG not ready - available replicas less than desired",
+			resourceName: "not-ready-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "not-ready-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           3,
+					AvailableReplicas:  1,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "desired=3, available=1",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "not-ready-pcsg",
+				Replicas:          3,
+				UpdatedReplicas:   3,
+				AvailableReplicas: ptr.To(int32(1)),
+			},
+		},
+		{
+			name:         "PCSG not fully updated - updated replicas less than desired",
+			resourceName: "not-updated-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "not-updated-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           3,
+					AvailableReplicas:  3,
+					UpdatedReplicas:    2,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "desired=3, updated=2",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "not-updated-pcsg",
+				Replicas:          3,
+				UpdatedReplicas:   2,
+				AvailableReplicas: ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PCSG performing rolling update - replicas != desired",
+			resourceName: "rolling-update-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "rolling-update-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 3,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           4,
+					AvailableReplicas:  3,
+					UpdatedReplicas:    3,
+					ObservedGeneration: ptr.To(int64(1)),
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "performing rolling update",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "rolling-update-pcsg",
+				Replicas:          4,
+				UpdatedReplicas:   3,
+				AvailableReplicas: ptr.To(int32(3)),
+			},
+		},
+		{
+			name:         "PCSG with nil observedGeneration",
+			resourceName: "nil-observed-gen-pcsg",
+			namespace:    "default",
+			existingPCSG: &grovev1alpha1.PodCliqueScalingGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "nil-observed-gen-pcsg",
+					Namespace:  "default",
+					Generation: 1,
+				},
+				Spec: grovev1alpha1.PodCliqueScalingGroupSpec{
+					Replicas: 2,
+				},
+				Status: grovev1alpha1.PodCliqueScalingGroupStatus{
+					Replicas:           2,
+					AvailableReplicas:  2,
+					UpdatedReplicas:    2,
+					ObservedGeneration: nil,
+				},
+			},
+			wantReady:          false,
+			wantReasonContains: "observedGeneration is nil",
+			wantServiceStatus: v1alpha1.ServiceReplicaStatus{
+				ComponentKind:     v1alpha1.ComponentKindPodCliqueScalingGroup,
+				ComponentName:     "nil-observed-gen-pcsg",
+				Replicas:          2,
+				UpdatedReplicas:   2,
+				AvailableReplicas: ptr.To(int32(2)),
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			g := gomega.NewGomegaWithT(t)
+
+			s := scheme.Scheme
+			err := v1alpha1.AddToScheme(s)
+			g.Expect(err).NotTo(gomega.HaveOccurred())
+			err = grovev1alpha1.AddToScheme(s)
+			g.Expect(err).NotTo(gomega.HaveOccurred())
+
+			var objects []client.Object
+			if tt.existingPCSG != nil {
+				objects = append(objects, tt.existingPCSG)
+			}
+
+			fakeKubeClient := fake.NewClientBuilder().
+				WithScheme(s).
+				WithObjects(objects...).
+				WithStatusSubresource(objects...).
+				Build()
+
+			logger := log.FromContext(ctx)
+			ready, reason, serviceStatus := CheckPCSGReady(ctx, fakeKubeClient, tt.resourceName, tt.namespace, logger)
+
+			g.Expect(ready).To(gomega.Equal(tt.wantReady))
+			if tt.wantReasonContains != "" {
+				g.Expect(reason).To(gomega.ContainSubstring(tt.wantReasonContains))
+			} else {
+				g.Expect(reason).To(gomega.Equal(""))
+			}
+			g.Expect(serviceStatus).To(gomega.Equal(tt.wantServiceStatus))
+		})
+	}
+}
+
 func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 	ctx := context.Background()
 
@@ -351,9 +882,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 2,
 					},
 					Status: grovev1alpha1.PodCliqueStatus{
-						Replicas:        2,
-						UpdatedReplicas: 2,
-						ReadyReplicas:   1,
+						Replicas:           2,
+						UpdatedReplicas:    2,
+						ReadyReplicas:      1,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 			},
@@ -403,9 +935,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 2,
 					},
 					Status: grovev1alpha1.PodCliqueScalingGroupStatus{
-						Replicas:          2,
-						UpdatedReplicas:   2,
-						AvailableReplicas: 2,
+						Replicas:           2,
+						UpdatedReplicas:    2,
+						AvailableReplicas:  2,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 				&grovev1alpha1.PodCliqueScalingGroup{
@@ -417,9 +950,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 3,
 					},
 					Status: grovev1alpha1.PodCliqueScalingGroupStatus{
-						Replicas:          3,
-						UpdatedReplicas:   3,
-						AvailableReplicas: 3,
+						Replicas:           3,
+						UpdatedReplicas:    3,
+						AvailableReplicas:  3,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 			},
@@ -467,9 +1001,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 2,
 					},
 					Status: grovev1alpha1.PodCliqueScalingGroupStatus{
-						Replicas:          2,
-						UpdatedReplicas:   2,
-						AvailableReplicas: 1,
+						Replicas:           2,
+						UpdatedReplicas:    2,
+						AvailableReplicas:  1,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 			},
@@ -525,9 +1060,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 1,
 					},
 					Status: grovev1alpha1.PodCliqueStatus{
-						Replicas:        1,
-						UpdatedReplicas: 1,
-						ReadyReplicas:   1,
+						Replicas:           1,
+						UpdatedReplicas:    1,
+						ReadyReplicas:      1,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 				&grovev1alpha1.PodCliqueScalingGroup{
@@ -539,9 +1075,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 2,
 					},
 					Status: grovev1alpha1.PodCliqueScalingGroupStatus{
-						Replicas:          2,
-						UpdatedReplicas:   2,
-						AvailableReplicas: 1,
+						Replicas:           2,
+						UpdatedReplicas:    2,
+						AvailableReplicas:  1,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 				&grovev1alpha1.PodCliqueScalingGroup{
@@ -553,9 +1090,10 @@ func Test_GetComponentReadinessAndServiceReplicaStatuses(t *testing.T) {
 						Replicas: 2,
 					},
 					Status: grovev1alpha1.PodCliqueScalingGroupStatus{
-						Replicas:          2,
-						UpdatedReplicas:   2,
-						AvailableReplicas: 2,
+						Replicas:           2,
+						UpdatedReplicas:    2,
+						AvailableReplicas:  2,
+						ObservedGeneration: ptr.To(int64(1)),
 					},
 				},
 			},

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

@@ -54,6 +54,11 @@ func (v *DynamoGraphDeploymentValidator) Validate() (admission.Warnings, error)
 		return nil, err
 	}
 
+	// Validate restart
+	if err := v.validateRestart(); err != nil {
+		return nil, err
+	}
+
 	var allWarnings admission.Warnings
 
 	// Validate each service
@@ -263,6 +268,64 @@ func (v *DynamoGraphDeploymentValidator) validatePVC(index int, pvc *nvidiacomv1
 	return err
 }
 
+func (v *DynamoGraphDeploymentValidator) validateRestart() error {
+	if v.deployment.Spec.Restart == nil {
+		return nil
+	}
+
+	restart := v.deployment.Spec.Restart
+
+	var err error
+	if restart.ID == "" {
+		err = errors.Join(err, fmt.Errorf("spec.restart.id is required"))
+	}
+
+	return errors.Join(err, v.validateRestartStrategyOrder())
+}
+
+func (v *DynamoGraphDeploymentValidator) validateRestartStrategyOrder() error {
+	restart := v.deployment.Spec.Restart
+	if restart.Strategy == nil || len(restart.Strategy.Order) == 0 {
+		return nil
+	}
+
+	if restart.Strategy.Type == nvidiacomv1alpha1.RestartStrategyTypeParallel {
+		return errors.New("spec.restart.strategy.order cannot be specified when strategy is parallel")
+	}
+
+	var err error
+
+	uniqueOrder := getUnique(restart.Strategy.Order)
+
+	if len(uniqueOrder) != len(restart.Strategy.Order) {
+		err = errors.Join(err, fmt.Errorf("spec.restart.strategy.order must be unique"))
+	}
+
+	if len(uniqueOrder) != len(v.deployment.Spec.Services) {
+		err = errors.Join(err, fmt.Errorf("spec.restart.strategy.order must have the same number of unique services as the deployment"))
+	}
+
+	for _, serviceName := range uniqueOrder {
+		if _, exists := v.deployment.Spec.Services[serviceName]; !exists {
+			err = errors.Join(err, fmt.Errorf("spec.restart.strategy.order contains unknown service: %s", serviceName))
+		}
+	}
+
+	return err
+}
+
+func getUnique[T comparable](slice []T) []T {
+	seen := make(map[T]struct{}, len(slice))
+	uniqueSlice := make([]T, 0, len(slice))
+	for _, element := range slice {
+		if _, exists := seen[element]; !exists {
+			seen[element] = struct{}{}
+			uniqueSlice = append(uniqueSlice, element)
+		}
+	}
+	return uniqueSlice
+}
+
 // getServiceNames extracts service names from a services map.
 // Returns a set-like map for efficient lookup and comparison.
 func getServiceNames(services map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec) map[string]struct{} {

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

@@ -322,6 +322,190 @@ func TestDynamoGraphDeploymentValidator_Validate(t *testing.T) {
 			wantErr: true,
 			errMsg:  "spec.services[main].sharedMemory.size is required when disabled is false",
 		},
+		// Restart validation test cases
+		{
+			name: "restart with nil at",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "",
+					},
+				},
+			},
+			wantErr: true,
+			errMsg:  "spec.restart.id is required",
+		},
+		{
+			name: "restart with valid id and no strategy",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+					},
+				},
+			},
+			wantErr: false,
+		},
+		{
+			name: "restart with parallel strategy and order specified",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main":    {},
+						"prefill": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeParallel,
+							Order: []string{"main", "prefill"},
+						},
+					},
+				},
+			},
+			wantErr: true,
+			errMsg:  "spec.restart.strategy.order cannot be specified when strategy is parallel",
+		},
+		{
+			name: "restart with sequential strategy and duplicate services in order",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main":    {},
+						"prefill": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"main", "main", "prefill"},
+						},
+					},
+				},
+			},
+			wantErr:     true,
+			errMsg:      "spec.restart.strategy.order must be unique",
+			errContains: true,
+		},
+		{
+			name: "restart with sequential strategy and unknown service in order",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main":    {},
+						"prefill": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"main", "unknown"},
+						},
+					},
+				},
+			},
+			wantErr:     true,
+			errMsg:      "spec.restart.strategy.order contains unknown service: unknown",
+			errContains: true,
+		},
+		{
+			name: "restart with sequential strategy and missing service in order",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main":    {},
+						"prefill": {},
+						"decode":  {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"main", "prefill"},
+						},
+					},
+				},
+			},
+			wantErr:     true,
+			errMsg:      "spec.restart.strategy.order must have the same number of unique services as the deployment",
+			errContains: true,
+		},
+		{
+			name: "restart with valid sequential strategy and order",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main":    {},
+						"prefill": {},
+						"decode":  {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeSequential,
+							Order: []string{"prefill", "decode", "main"},
+						},
+					},
+				},
+			},
+			wantErr: false,
+		},
+		{
+			name: "restart with sequential strategy and empty order is valid",
+			deployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-graph",
+					Namespace: "default",
+				},
+				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
+					Services: map[string]*nvidiacomv1alpha1.DynamoComponentDeploymentSharedSpec{
+						"main": {},
+					},
+					Restart: &nvidiacomv1alpha1.Restart{
+						ID: "restart-id",
+						Strategy: &nvidiacomv1alpha1.RestartStrategy{
+							Type:  nvidiacomv1alpha1.RestartStrategyTypeSequential,
+							Order: []string{},
+						},
+					},
+				},
+			},
+			wantErr: false,
+		},
 	}
 
 	for _, tt := range tests {
@@ -791,7 +975,7 @@ func TestDynamoGraphDeploymentValidator_ValidateUpdate(t *testing.T) {
 			errMsg:  "spec.services[main] cannot change node topology (between single-node and multi-node) after creation",
 		},
 		{
-			name: "adding new service with multinode is allowed",
+			name: "adding new service with multinode is not allowed", // service topology is immutable
 			oldDeployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
 				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
 					BackendFramework: "sglang",
@@ -810,7 +994,6 @@ func TestDynamoGraphDeploymentValidator_ValidateUpdate(t *testing.T) {
 							Multinode: nil,
 						},
 						"decode": {
-							// New service with multinode - should be allowed
 							Multinode: &nvidiacomv1alpha1.MultinodeSpec{
 								NodeCount: 4,
 							},
@@ -818,10 +1001,11 @@ func TestDynamoGraphDeploymentValidator_ValidateUpdate(t *testing.T) {
 					},
 				},
 			},
-			wantErr: false,
+			wantErr: true,
+			errMsg:  "service topology is immutable and cannot be modified after creation: services added: [decode]",
 		},
 		{
-			name: "adding new service without multinode is allowed",
+			name: "adding new service without multinode is not allowed", // service topology is immutable
 			oldDeployment: &nvidiacomv1alpha1.DynamoGraphDeployment{
 				Spec: nvidiacomv1alpha1.DynamoGraphDeploymentSpec{
 					BackendFramework: "sglang",
@@ -850,7 +1034,8 @@ func TestDynamoGraphDeploymentValidator_ValidateUpdate(t *testing.T) {
 					},
 				},
 			},
-			wantErr: false,
+			wantErr: true,
+			errMsg:  "service topology is immutable and cannot be modified after creation: services added: [gateway]",
 		},
 	}
 

docs/kubernetes/api_reference.md

@@ -434,6 +434,7 @@ _Appears in:_
 | `services` _object (keys:string, values:[DynamoComponentDeploymentSharedSpec](#dynamocomponentdeploymentsharedspec))_ | Services are the services to deploy as part of this deployment. |  | MaxProperties: 25 <br />Optional: \{\} <br /> |
 | `envs` _[EnvVar](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.28/#envvar-v1-core) array_ | Envs are environment variables applied to all services in the deployment unless<br />overridden by service-specific configuration. |  | Optional: \{\} <br /> |
 | `backendFramework` _string_ | BackendFramework specifies the backend framework (e.g., "sglang", "vllm", "trtllm"). |  | Enum: [sglang vllm trtllm] <br /> |
+| `restart` _[Restart](#restart)_ | Restart specifies the restart policy for the graph deployment. |  | Optional: \{\} <br /> |
 
 
 #### DynamoGraphDeploymentStatus
@@ -452,6 +453,7 @@ _Appears in:_
 | `state` _string_ | State is a high-level textual status of the graph deployment lifecycle. |  |  |
 | `conditions` _[Condition](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.28/#condition-v1-meta) array_ | Conditions contains the latest observed conditions of the graph deployment.<br />The slice is merged by type on patch updates. |  |  |
 | `services` _object (keys:string, values:[ServiceReplicaStatus](#servicereplicastatus))_ | Services contains per-service replica status information.<br />The map key is the service name from spec.services. |  |  |
+| `restart` _[RestartStatus](#restartstatus)_ | Restart contains the status of the restart of the graph deployment. |  |  |
 
 
 #### DynamoModel
@@ -742,6 +744,94 @@ _Appears in:_
 | `claims` _[ResourceClaim](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.28/#resourceclaim-v1-core) array_ | Claims specifies resource claims for dynamic resource allocation |  |  |
 
 
+#### Restart
+
+
+
+
+
+
+
+_Appears in:_
+- [DynamoGraphDeploymentSpec](#dynamographdeploymentspec)
+
+| Field | Description | Default | Validation |
+| --- | --- | --- | --- |
+| `id` _string_ | ID is an arbitrary string that triggers a restart when changed.<br />Any modification to this value will initiate a restart of the graph deployment according to the strategy. |  | MinLength: 1 <br />Required: \{\} <br /> |
+| `strategy` _[RestartStrategy](#restartstrategy)_ | Strategy specifies the restart strategy for the graph deployment. |  | Optional: \{\} <br /> |
+
+
+#### RestartPhase
+
+_Underlying type:_ _string_
+
+
+
+
+
+_Appears in:_
+- [RestartStatus](#restartstatus)
+
+| Field | Description |
+| --- | --- |
+| `Pending` |  |
+| `Restarting` |  |
+| `Completed` |  |
+| `Failed` |  |
+
+
+#### RestartStatus
+
+
+
+RestartStatus contains the status of the restart of the graph deployment.
+
+
+
+_Appears in:_
+- [DynamoGraphDeploymentStatus](#dynamographdeploymentstatus)
+
+| Field | Description | Default | Validation |
+| --- | --- | --- | --- |
+| `observedID` _string_ | ObservedID is the restart ID that has been observed and is being processed.<br />Matches the Restart.ID field in the spec. |  |  |
+| `phase` _[RestartPhase](#restartphase)_ | Phase is the phase of the restart. |  |  |
+| `inProgress` _string array_ | InProgress contains the names of the services that are currently being restarted. |  |  |
+
+
+#### RestartStrategy
+
+
+
+
+
+
+
+_Appears in:_
+- [Restart](#restart)
+
+| Field | Description | Default | Validation |
+| --- | --- | --- | --- |
+| `type` _[RestartStrategyType](#restartstrategytype)_ | Type specifies the restart strategy type. | Sequential | Enum: [Sequential Parallel] <br /> |
+| `order` _string array_ | Order specifies the order in which the services should be restarted. |  | Optional: \{\} <br /> |
+
+
+#### RestartStrategyType
+
+_Underlying type:_ _string_
+
+
+
+
+
+_Appears in:_
+- [RestartStrategy](#restartstrategy)
+
+| Field | Description |
+| --- | --- |
+| `Sequential` |  |
+| `Parallel` |  |
+
+
 #### ScalingAdapter