Stop and Resume a transformer (kserve#4534)

Signed-off-by: Hannah DeFazio <h2defazio@gmail.com>
Co-authored-by: Jooho Lee <jlee@redhat.com>

Author: hdefazio

Makefile

@@ -161,7 +161,7 @@ clean:
 
 # Run tests
 test: fmt vet manifests envtest test-qpext
-	KUBEBUILDER_ASSETS="$$($(ENVTEST) use $(ENVTEST_K8S_VERSION) -p path)" go test $$(go list ./pkg/...) ./cmd/... -coverprofile coverage.out -coverpkg ./pkg/... ./cmd...
+	KUBEBUILDER_ASSETS="$$($(ENVTEST) use $(ENVTEST_K8S_VERSION) -p path)" go test --timeout 20m $$(go list ./pkg/...) ./cmd/... -coverprofile coverage.out -coverpkg ./pkg/... ./cmd...
 
 test-qpext:
 	cd qpext && go test -v ./... -cover

pkg/controller/v1beta1/inferenceservice/components/predictor.go

@@ -198,42 +198,27 @@ func (p *Predictor) Reconcile(ctx context.Context, isvc *v1beta1.InferenceServic
 	// Handle InferenceService status updates based on the force stop annotation.
 	// If true, transition the service to a stopped and unready state; otherwise, ensure it's not marked as stopped.
 	if utils.GetForceStopRuntime(isvc) {
-		// Exit early if we have already set the status to stopped
-		existingStoppedCondition := isvc.Status.GetCondition(v1beta1.Stopped)
-		if existingStoppedCondition != nil && existingStoppedCondition.Status == corev1.ConditionTrue {
+		// Exit early if we have already set the predictor's status to stopped
+		existingPredictorCondition := isvc.Status.GetCondition(v1beta1.PredictorReady)
+		if existingPredictorCondition != nil && existingPredictorCondition.Status == corev1.ConditionFalse && existingPredictorCondition.Reason == v1beta1.StoppedISVCReason {
 			return ctrl.Result{}, nil
 		}
 
+		// Preserve the deployment mode value
 		deployMode := isvc.Status.DeploymentMode
 
 		// Clear all statuses
 		isvc.Status = v1beta1.InferenceServiceStatus{}
-
-		// Preserve the deployment mode value
 		isvc.Status.DeploymentMode = deployMode
 
-		// Set the ready condition
-		predictorReadyCondition := &apis.Condition{
+		// Set the predictor's ready condition to false
+		isvc.Status.SetCondition(v1beta1.PredictorReady, &apis.Condition{
 			Type:   v1beta1.PredictorReady,
 			Status: corev1.ConditionFalse,
 			Reason: v1beta1.StoppedISVCReason,
-		}
-		isvc.Status.SetCondition(v1beta1.PredictorReady, predictorReadyCondition)
-
-		// Add the stopped condition
-		stoppedCondition := &apis.Condition{
-			Type:   v1beta1.Stopped,
-			Status: corev1.ConditionTrue,
-		}
-		isvc.Status.SetCondition(v1beta1.Stopped, stoppedCondition)
+		})
 
 		return ctrl.Result{}, nil
-	} else {
-		resumeCondition := &apis.Condition{
-			Type:   v1beta1.Stopped,
-			Status: corev1.ConditionFalse,
-		}
-		isvc.Status.SetCondition(v1beta1.Stopped, resumeCondition)
 	}
 
 	statusSpec := isvc.Status.Components[v1beta1.PredictorComponent]

pkg/controller/v1beta1/inferenceservice/components/transformer.go

@@ -28,6 +28,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/client-go/kubernetes"
+	"knative.dev/pkg/apis"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
@@ -152,45 +153,78 @@ func (p *Transformer) Reconcile(ctx context.Context, isvc *v1beta1.InferenceServ
 
 	// Here we allow switch between knative and vanilla deployment
 	if p.deploymentMode == constants.RawDeployment {
-		r, err := raw.NewRawKubeReconciler(ctx, p.client, p.clientset, p.scheme, objectMeta, metav1.ObjectMeta{},
-			&isvc.Spec.Transformer.ComponentExtensionSpec, &podSpec, nil)
-		if err != nil {
-			return ctrl.Result{}, errors.Wrapf(err, "fails to create NewRawKubeReconciler for transformer")
-		}
-		// set Deployment Controller
-		for _, deployment := range r.Deployment.DeploymentList {
-			if err := controllerutil.SetControllerReference(isvc, deployment, p.scheme); err != nil {
-				return ctrl.Result{}, errors.Wrapf(err, "fails to set deployment owner reference for transformer")
-			}
+		if err := p.reconcileTransformerRawDeployment(ctx, isvc, &objectMeta, &podSpec); err != nil {
+			return ctrl.Result{}, err
 		}
-		// set Service Controller
-		for _, svc := range r.Service.ServiceList {
-			if err := controllerutil.SetControllerReference(isvc, svc, p.scheme); err != nil {
-				return ctrl.Result{}, errors.Wrapf(err, "fails to set service owner reference for transformer")
-			}
-		}
-		// set autoscaler Controller
-		if err := r.Scaler.Autoscaler.SetControllerReferences(isvc, p.scheme); err != nil {
-			return ctrl.Result{}, errors.Wrapf(err, "fails to set autoscaler owner references for transformer")
+	} else {
+		if err := p.reconcileTransformerKnativeDeployment(ctx, isvc, &objectMeta, &podSpec); err != nil {
+			return ctrl.Result{}, err
 		}
+	}
 
-		deployment, err := r.Reconcile(ctx)
-		if err != nil {
-			return ctrl.Result{}, errors.Wrapf(err, "fails to reconcile transformer")
+	if utils.GetForceStopRuntime(isvc) {
+		// Exit early if we have already set the transformer's status to stopped
+		existingTransformerCondition := isvc.Status.GetCondition(v1beta1.TransformerReady)
+		if existingTransformerCondition != nil && existingTransformerCondition.Status == corev1.ConditionFalse && existingTransformerCondition.Reason == v1beta1.StoppedISVCReason {
+			return ctrl.Result{}, nil
 		}
-		isvc.Status.PropagateRawStatus(v1beta1.TransformerComponent, deployment, r.URL)
-	} else {
-		r := knative.NewKsvcReconciler(p.client, p.scheme, objectMeta, &isvc.Spec.Transformer.ComponentExtensionSpec,
-			&podSpec, isvc.Status.Components[v1beta1.TransformerComponent], p.inferenceServiceConfig.ServiceLabelDisallowedList)
 
-		if err := controllerutil.SetControllerReference(isvc, r.Service, p.scheme); err != nil {
-			return ctrl.Result{}, errors.Wrapf(err, "fails to set owner reference for transformer")
+		// Set the ready condition to false
+		isvc.Status.SetCondition(v1beta1.TransformerReady, &apis.Condition{
+			Type:   v1beta1.TransformerReady,
+			Status: corev1.ConditionFalse,
+			Reason: v1beta1.StoppedISVCReason,
+		})
+	}
+	return ctrl.Result{}, nil
+}
+
+func (p *Transformer) reconcileTransformerRawDeployment(ctx context.Context, isvc *v1beta1.InferenceService, objectMeta *metav1.ObjectMeta, podSpec *corev1.PodSpec) error {
+	r, err := raw.NewRawKubeReconciler(ctx, p.client, p.clientset, p.scheme, *objectMeta, metav1.ObjectMeta{},
+		&isvc.Spec.Transformer.ComponentExtensionSpec, podSpec, nil)
+	if err != nil {
+		return errors.Wrapf(err, "fails to create NewRawKubeReconciler for transformer")
+	}
+	// set Deployment Controller
+	for _, deployment := range r.Deployment.DeploymentList {
+		if err := controllerutil.SetControllerReference(isvc, deployment, p.scheme); err != nil {
+			return errors.Wrapf(err, "fails to set deployment owner reference for transformer")
 		}
-		status, err := r.Reconcile(ctx)
-		if err != nil {
-			return ctrl.Result{}, errors.Wrapf(err, "fails to reconcile transformer")
+	}
+	// set Service Controller
+	for _, svc := range r.Service.ServiceList {
+		if err := controllerutil.SetControllerReference(isvc, svc, p.scheme); err != nil {
+			return errors.Wrapf(err, "fails to set service owner reference for transformer")
 		}
-		isvc.Status.PropagateStatus(v1beta1.TransformerComponent, status)
 	}
-	return ctrl.Result{}, nil
+	// set autoscaler Controller
+	if err := r.Scaler.Autoscaler.SetControllerReferences(isvc, p.scheme); err != nil {
+		return errors.Wrapf(err, "fails to set autoscaler owner references for transformer")
+	}
+
+	deployment, err := r.Reconcile(ctx)
+	if err != nil {
+		return errors.Wrapf(err, "fails to reconcile transformer")
+	}
+	if !utils.GetForceStopRuntime(isvc) {
+		isvc.Status.PropagateRawStatus(v1beta1.TransformerComponent, deployment, r.URL)
+	}
+	return nil
+}
+
+func (p *Transformer) reconcileTransformerKnativeDeployment(ctx context.Context, isvc *v1beta1.InferenceService, objectMeta *metav1.ObjectMeta, podSpec *corev1.PodSpec) error {
+	r := knative.NewKsvcReconciler(p.client, p.scheme, *objectMeta, &isvc.Spec.Transformer.ComponentExtensionSpec,
+		podSpec, isvc.Status.Components[v1beta1.TransformerComponent], p.inferenceServiceConfig.ServiceLabelDisallowedList)
+
+	if err := controllerutil.SetControllerReference(isvc, r.Service, p.scheme); err != nil {
+		return errors.Wrapf(err, "fails to set owner reference for transformer")
+	}
+	kstatus, err := r.Reconcile(ctx)
+	if err != nil {
+		return errors.Wrapf(err, "fails to reconcile transformer")
+	}
+	if !utils.GetForceStopRuntime(isvc) {
+		isvc.Status.PropagateStatus(v1beta1.TransformerComponent, kstatus)
+	}
+	return nil
 }

pkg/controller/v1beta1/inferenceservice/controller.go

@@ -139,6 +139,7 @@ func (r *InferenceServiceReconciler) Reconcile(ctx context.Context, req ctrl.Req
 	annotations := utils.Filter(isvc.Annotations, func(key string) bool {
 		return !utils.Includes(isvcConfig.ServiceAnnotationDisallowedList, key)
 	})
+	forceStopRuntime := utils.GetForceStopRuntime(isvc)
 
 	deployConfig, err := v1beta1.NewDeployConfig(isvcConfigMap)
 	if err != nil {
@@ -254,6 +255,28 @@ func (r *InferenceServiceReconciler) Reconcile(ctx context.Context, req ctrl.Req
 			return result, nil
 		}
 	}
+	// Handle InferenceService status updates based on the force stop annotation.
+	// If true, transition the service to a stopped and unready state; otherwise, ensure it's not marked as stopped.
+	if forceStopRuntime {
+		// Exit early if we have already set the status to stopped
+		existingStoppedCondition := isvc.Status.GetCondition(v1beta1.Stopped)
+		if existingStoppedCondition != nil && existingStoppedCondition.Status == corev1.ConditionTrue {
+			// TODO: Set condition to stoppING
+		} else {
+			// Add the stopped condition
+			stoppedCondition := &apis.Condition{
+				Type:   v1beta1.Stopped,
+				Status: corev1.ConditionTrue,
+			}
+			isvc.Status.SetCondition(v1beta1.Stopped, stoppedCondition)
+		}
+	} else {
+		resumeCondition := &apis.Condition{
+			Type:   v1beta1.Stopped,
+			Status: corev1.ConditionFalse,
+		}
+		isvc.Status.SetCondition(v1beta1.Stopped, resumeCondition)
+	}
 	// reconcile RoutesReady and LatestDeploymentReady conditions for serverless deployment
 	if deploymentMode == constants.Serverless {
 		componentList := []v1beta1.ComponentType{v1beta1.PredictorComponent}
@@ -263,7 +286,7 @@ func (r *InferenceServiceReconciler) Reconcile(ctx context.Context, req ctrl.Req
 		if isvc.Spec.Explainer != nil {
 			componentList = append(componentList, v1beta1.ExplainerComponent)
 		}
-		if !utils.GetForceStopRuntime(isvc) {
+		if !forceStopRuntime {
 			isvc.Status.PropagateCrossComponentStatus(componentList, v1beta1.RoutesReady)
 			isvc.Status.PropagateCrossComponentStatus(componentList, v1beta1.LatestDeploymentReady)
 		}