allow external hpa

Signed-off-by: Robin <[email protected]>

Author: Robin

config/internal/base/deployment.yaml.tmpl

@@ -22,7 +22,9 @@ metadata:
     app.kubernetes.io/name: modelmesh-controller
     name: {{.ServiceName}}-{{.Name}}
 spec:
+  {{if ge .Replicas 0}}
   replicas: {{.Replicas}}
+  {{end}}
   selector:
     matchLabels:
       modelmesh-service: {{.ServiceName}}

controllers/autoscaler/autoscaler_reconciler.go

@@ -104,6 +104,10 @@ func createAutoscaler(client client.Client,
 		// Set HPA reconciler even though AutoscalerClass is None to delete existing hpa
 		as.HPA = hpa.NewHPAReconciler(client, scheme, runtimeMeta, mmDeploymentName, mmNamespace)
 		return as, nil
+	case constants.AutoscalerClassExternal:
+		// Set HPA reconciler even though AutoscalerClass is External to delete existing hpa
+		as.HPA = hpa.NewHPAReconciler(client, scheme, runtimeMeta, mmDeploymentName, mmNamespace)
+		return as, nil
 	default:
 		return nil, errors.New("unknown autoscaler class type.")
 	}

controllers/modelmesh/modelmesh.go

@@ -66,7 +66,7 @@ type Deployment struct {
 	PullerImage         string
 	PullerImageCommand  []string
 	PullerResources     *corev1.ResourceRequirements
-	Replicas            uint16
+	Replicas            int32
 	Port                uint16
 	TLSSecretName       string
 	TLSClientAuth       string

controllers/servingruntime_controller.go

@@ -28,6 +28,7 @@ import (
 
 	kserveapi "github.com/kserve/kserve/pkg/apis/serving/v1alpha1"
 	"github.com/kserve/kserve/pkg/apis/serving/v1beta1"
+	"github.com/kserve/kserve/pkg/constants"
 	api "github.com/kserve/modelmesh-serving/apis/serving/v1alpha1"
 	"github.com/kserve/modelmesh-serving/controllers/autoscaler"
 	"github.com/kserve/modelmesh-serving/controllers/modelmesh"
@@ -291,7 +292,12 @@ func (r *ServingRuntimeReconciler) Reconcile(ctx context.Context, req ctrl.Reque
 
 	//ScaleToZero or None autoscaler case
 	if replicas == uint16(0) || as.Autoscaler.AutoscalerClass == autoscaler.AutoscalerClassNone {
-		mmDeployment.Replicas = replicas
+		mmDeployment.Replicas = int32(replicas)
+		if _, err = as.Reconcile(true); err != nil {
+			return ctrl.Result{}, fmt.Errorf("HPA reconcile error: %w", err)
+		}
+	} else if as.Autoscaler.AutoscalerClass == constants.AutoscalerClassExternal {
+		mmDeployment.Replicas = -1
 		if _, err = as.Reconcile(true); err != nil {
 			return ctrl.Result{}, fmt.Errorf("HPA reconcile error: %w", err)
 		}
@@ -309,9 +315,9 @@ func (r *ServingRuntimeReconciler) Reconcile(ctx context.Context, req ctrl.Reque
 				return ctrl.Result{}, fmt.Errorf("Could not get the deployment for the servingruntime : %w", err)
 			}
 			if *existingDeployment.Spec.Replicas == int32(0) {
-				mmDeployment.Replicas = uint16(*(as.Autoscaler.HPA.HPA).Spec.MinReplicas)
+				mmDeployment.Replicas = *(as.Autoscaler.HPA.HPA).Spec.MinReplicas
 			} else {
-				mmDeployment.Replicas = uint16(*(existingDeployment.Spec.Replicas))
+				mmDeployment.Replicas = *(existingDeployment.Spec.Replicas)
 			}
 		}