Allow disabling creation of the HPA in raw deployment mode (kserve#3086)

* Refactor autoscaler functionality into an interface. Add no-op autoscaler so users disable creation of autoscaler resources. Add AutoscalerClassNone feature with test cases Signed-off-by: jooho <jlee@redhat.com> Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Rollback getHPAMetrics to set ScaleTarget,ScaleMetric Signed-off-by: jooho <jlee@redhat.com> Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Fix merge conflict Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Fix autoscaler api group Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Fix autoscaler merge Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Fix autoscaling tests Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Updated hpa disable test. Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Replaced "AutoscalerClassNone" with "AutoscalerClassExternal" Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> * Fixed autoscalerClass 'external' go test. Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> --------- Signed-off-by: jooho <jlee@redhat.com> Signed-off-by: Dan Sun <dsun20@bloomberg.net> Signed-off-by: Andrews Arokiam <andrews.arokiam@ideas2it.com> Co-authored-by: Curtis Maddalozzo <curtis@tractable.ai> Co-authored-by: jooho <jlee@redhat.com> Co-authored-by: Dan Sun <dsun20@bloomberg.net>
spolti · Apr 24, 2024 · ea50e18 · ea50e18
1 parent 8777360
commit ea50e18
Show file tree

Hide file tree

Showing 10 changed files with 468 additions and 40 deletions.
diff --git a/pkg/apis/serving/v1beta1/inference_service_validation.go b/pkg/apis/serving/v1beta1/inference_service_validation.go
@@ -139,6 +139,8 @@ func validateInferenceServiceAutoscaler(isvc *InferenceService) error {
 					} else {
 						return nil
 					}
+				case constants.AutoscalerClassExternal:
+					return nil
 				default:
 					return fmt.Errorf("unknown autoscaler class [%s]", class)
 				}

diff --git a/pkg/constants/constants.go b/pkg/constants/constants.go
@@ -154,7 +154,8 @@ var (
 
 // Autoscaler Class
 var (
-	AutoscalerClassHPA AutoscalerClassType = "hpa"
+	AutoscalerClassHPA      AutoscalerClassType = "hpa"
+	AutoscalerClassExternal AutoscalerClassType = "external"
 )
 
 // Autoscaler Metrics
@@ -170,6 +171,7 @@ var (
 // Autoscaler Class Allowed List
 var AutoscalerAllowedClassList = []AutoscalerClassType{
 	AutoscalerClassHPA,
+	AutoscalerClassExternal,
 }
 
 // Autoscaler Metrics Allowed List

diff --git a/pkg/controller/v1beta1/inferenceservice/components/explainer.go b/pkg/controller/v1beta1/inferenceservice/components/explainer.go
@@ -151,10 +151,8 @@ func (e *Explainer) Reconcile(isvc *v1beta1.InferenceService) (ctrl.Result, erro
 			return ctrl.Result{}, errors.Wrapf(err, "fails to set service owner reference for explainer")
 		}
 		//set autoscaler Controller
-		if r.Scaler.Autoscaler.AutoscalerClass == constants.AutoscalerClassHPA {
-			if err := controllerutil.SetControllerReference(isvc, r.Scaler.Autoscaler.HPA.HPA, e.scheme); err != nil {
-				return ctrl.Result{}, errors.Wrapf(err, "fails to set HPA owner reference for explainer")
-			}
+		if err := r.Scaler.Autoscaler.SetControllerReferences(isvc, e.scheme); err != nil {
+			return ctrl.Result{}, errors.Wrapf(err, "fails to set autoscaler owner references for explainer")
 		}
 
 		deployment, err := r.Reconcile()

diff --git a/pkg/controller/v1beta1/inferenceservice/components/predictor.go b/pkg/controller/v1beta1/inferenceservice/components/predictor.go
@@ -319,10 +319,8 @@ func (p *Predictor) Reconcile(isvc *v1beta1.InferenceService) (ctrl.Result, erro
 			return ctrl.Result{}, errors.Wrapf(err, "fails to set service owner reference for predictor")
 		}
 		//set autoscaler Controller
-		if r.Scaler.Autoscaler.AutoscalerClass == constants.AutoscalerClassHPA {
-			if err := controllerutil.SetControllerReference(isvc, r.Scaler.Autoscaler.HPA.HPA, p.scheme); err != nil {
-				return ctrl.Result{}, errors.Wrapf(err, "fails to set HPA owner reference for predictor")
-			}
+		if err := r.Scaler.Autoscaler.SetControllerReferences(isvc, p.scheme); err != nil {
+			return ctrl.Result{}, errors.Wrapf(err, "fails to set autoscaler owner references for predictor")
 		}
 
 		deployment, err := r.Reconcile()

diff --git a/pkg/controller/v1beta1/inferenceservice/components/transformer.go b/pkg/controller/v1beta1/inferenceservice/components/transformer.go
@@ -181,10 +181,8 @@ func (p *Transformer) Reconcile(isvc *v1beta1.InferenceService) (ctrl.Result, er
 			return ctrl.Result{}, errors.Wrapf(err, "fails to set service owner reference for transformer")
 		}
 		//set autoscaler Controller
-		if r.Scaler.Autoscaler.AutoscalerClass == constants.AutoscalerClassHPA {
-			if err := controllerutil.SetControllerReference(isvc, r.Scaler.Autoscaler.HPA.HPA, p.scheme); err != nil {
-				return ctrl.Result{}, errors.Wrapf(err, "fails to set HPA owner reference for transformer")
-			}
+		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")
 		}
 
 		deployment, err := r.Reconcile()