[HPA] Add targetCPUUtilization field to collector config

apis/v1alpha1/opentelemetrycollector_types.go

@@ -41,13 +41,11 @@ type OpenTelemetryCollectorSpec struct {
 	// MaxReplicas sets an upper bound to the autoscaling feature. If MaxReplicas is set autoscaling is enabled.
 	// +optional
 	MaxReplicas *int32 `json:"maxReplicas,omitempty"`
-
 	// Autoscaler specifies the pod autoscaling configuration to use
 	// for the OpenTelemetryCollector workload.
 	//
 	// +optional
 	Autoscaler *AutoscalerSpec `json:"autoscaler,omitempty"`
-
 	// SecurityContext will be set as the container security context.
 	// +optional
 	SecurityContext *v1.SecurityContext `json:"securityContext,omitempty"`
@@ -216,6 +214,10 @@ type OpenTelemetryCollectorList struct {
 type AutoscalerSpec struct {
 	// +optional
 	Behavior *autoscalingv2.HorizontalPodAutoscalerBehavior `json:"behavior,omitempty"`
+	// TargetCPUUtilization sets the target average CPU used across all replicas.
+	// If average CPU exceeds this value, the HPA will scale up. Defaults to 90 percent.
+	// +optional
+	TargetCPUUtilization *int32 `json:"targetCPUUtilization,omitempty"`
 }
 
 func init() {

apis/v1alpha1/opentelemetrycollector_webhook.go

@@ -65,6 +65,12 @@ func (r *OpenTelemetryCollector) Default() {
 	if r.Spec.TargetAllocator.Enabled && r.Spec.TargetAllocator.Replicas == nil {
 		r.Spec.TargetAllocator.Replicas = &one
 	}
+
+	// Set default targetCPUUtilization for autoscaler
+	if r.Spec.MaxReplicas != nil && r.Spec.Autoscaler.TargetCPUUtilization == nil {
+		defaultCPUTarget := int32(90)
+		r.Spec.Autoscaler.TargetCPUUtilization = &defaultCPUTarget
+	}
 }
 
 // +kubebuilder:webhook:verbs=create;update,path=/validate-opentelemetry-io-v1alpha1-opentelemetrycollector,mutating=false,failurePolicy=fail,groups=opentelemetry.io,resources=opentelemetrycollectors,versions=v1alpha1,name=vopentelemetrycollectorcreateupdate.kb.io,sideEffects=none,admissionReviewVersions=v1
@@ -141,6 +147,10 @@ func (r *OpenTelemetryCollector) validateCRDSpec() error {
 				return fmt.Errorf("the OpenTelemetry Spec autoscale configuration is incorrect, scaleUp should be one or more")
 			}
 		}
+		if r.Spec.Autoscaler.TargetCPUUtilization != nil && (*r.Spec.Autoscaler.TargetCPUUtilization < int32(1) || *r.Spec.Autoscaler.TargetCPUUtilization > int32(99)) {
+			return fmt.Errorf("the OpenTelemetry Spec autoscale configuration is incorrect, targetCPUUtilization should be greater than 0 and less than 100")
+		}
+
 	}
 
 	return nil

bundle/manifests/opentelemetry.io_opentelemetrycollectors.yaml

@@ -182,6 +182,12 @@ spec:
                             type: integer
                         type: object
                     type: object
+                  targetCPUUtilization:
+                    description: TargetCPUUtilization sets the target average CPU
+                      used across all replicas. If average CPU exceeds this value,
+                      the HPA will scale up. Defaults to 90 percent.
+                    format: int32
+                    type: integer
                 type: object
               config:
                 description: Config is the raw JSON to be used as the collector's

config/crd/bases/opentelemetry.io_opentelemetrycollectors.yaml

@@ -180,6 +180,12 @@ spec:
                             type: integer
                         type: object
                     type: object
+                  targetCPUUtilization:
+                    description: TargetCPUUtilization sets the target average CPU
+                      used across all replicas. If average CPU exceeds this value,
+                      the HPA will scale up. Defaults to 90 percent.
+                    format: int32
+                    type: integer
                 type: object
               config:
                 description: Config is the raw JSON to be used as the collector's

docs/api.md

@@ -1896,6 +1896,15 @@ Autoscaler specifies the pod autoscaling configuration to use for the OpenTeleme
           HorizontalPodAutoscalerBehavior configures the scaling behavior of the target in both Up and Down directions (scaleUp and scaleDown fields respectively).<br/>
         </td>
         <td>false</td>
+      </tr><tr>
+        <td><b>targetCPUUtilization</b></td>
+        <td>integer</td>
+        <td>
+          TargetCPUUtilization sets the target average CPU used across all replicas. If average CPU exceeds this value, the HPA will scale up. Defaults to 90 percent.<br/>
+          <br/>
+            <i>Format</i>: int32<br/>
+        </td>
+        <td>false</td>
       </tr></tbody>
 </table>
 

pkg/collector/horizontalpodautoscaler.go

@@ -28,16 +28,14 @@ import (
 	"github.com/open-telemetry/opentelemetry-operator/pkg/naming"
 )
 
-const defaultCPUTarget int32 = 90
-
 func HorizontalPodAutoscaler(cfg config.Config, logger logr.Logger, otelcol v1alpha1.OpenTelemetryCollector) client.Object {
 	autoscalingVersion := cfg.AutoscalingVersion()
 
 	labels := Labels(otelcol, cfg.LabelsFilter())
 	labels["app.kubernetes.io/name"] = naming.Collector(otelcol)
 
 	annotations := Annotations(otelcol)
-	cpuTarget := defaultCPUTarget
+
 	var result client.Object
 
 	objectMeta := metav1.ObjectMeta{
@@ -54,7 +52,7 @@ func HorizontalPodAutoscaler(cfg config.Config, logger logr.Logger, otelcol v1al
 				Name: corev1.ResourceCPU,
 				Target: autoscalingv2beta2.MetricTarget{
 					Type:               autoscalingv2beta2.UtilizationMetricType,
-					AverageUtilization: &cpuTarget,
+					AverageUtilization: otelcol.Spec.Autoscaler.TargetCPUUtilization,
 				},
 			},
 		}
@@ -87,7 +85,7 @@ func HorizontalPodAutoscaler(cfg config.Config, logger logr.Logger, otelcol v1al
 				Name: corev1.ResourceCPU,
 				Target: autoscalingv2.MetricTarget{
 					Type:               autoscalingv2.UtilizationMetricType,
-					AverageUtilization: &cpuTarget,
+					AverageUtilization: otelcol.Spec.Autoscaler.TargetCPUUtilization,
 				},
 			},
 		}

pkg/collector/horizontalpodautoscaler_test.go

@@ -41,6 +41,7 @@ func TestHPA(t *testing.T) {
 
 	var minReplicas int32 = 3
 	var maxReplicas int32 = 5
+	var cpuUtilization int32 = 90
 
 	otelcol := v1alpha1.OpenTelemetryCollector{
 		ObjectMeta: metav1.ObjectMeta{
@@ -49,6 +50,9 @@ func TestHPA(t *testing.T) {
 		Spec: v1alpha1.OpenTelemetryCollectorSpec{
 			Replicas:    &minReplicas,
 			MaxReplicas: &maxReplicas,
+			Autoscaler: &v1alpha1.AutoscalerSpec{
+				TargetCPUUtilization: &cpuUtilization,
+			},
 		},
 	}
 

pkg/collector/reconcile/horizontalpodautoscaler.go

@@ -119,13 +119,15 @@ func setAutoscalerSpec(params Params, autoscalingVersion autodetect.AutoscalingV
 			} else {
 				updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.MinReplicas = &one
 			}
+			updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = params.Instance.Spec.Autoscaler.TargetCPUUtilization
 		} else {
 			updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.MaxReplicas = *params.Instance.Spec.MaxReplicas
 			if params.Instance.Spec.MinReplicas != nil {
 				updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.MinReplicas = params.Instance.Spec.MinReplicas
 			} else {
 				updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.MinReplicas = &one
 			}
+			updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = params.Instance.Spec.Autoscaler.TargetCPUUtilization
 		}
 	}
 }

pkg/collector/reconcile/horizontalpodautoscaler_test.go

@@ -109,6 +109,7 @@ func paramsWithHPA(autoscalingVersion autodetect.AutoscalingVersion) Params {
 
 	minReplicas := int32(3)
 	maxReplicas := int32(5)
+	cpuUtilization := int32(90)
 
 	mockAutoDetector := &mockAutoDetect{
 		HPAVersionFunc: func() (autodetect.AutoscalingVersion, error) {
@@ -147,6 +148,9 @@ func paramsWithHPA(autoscalingVersion autodetect.AutoscalingVersion) Params {
 				Config:      string(configYAML),
 				Replicas:    &minReplicas,
 				MaxReplicas: &maxReplicas,
+				Autoscaler: &v1alpha1.AutoscalerSpec{
+					TargetCPUUtilization: &cpuUtilization,
+				},
 			},
 		},
 		Scheme:   testScheme,

tests/e2e/autoscale/00-assert.yaml

@@ -4,7 +4,6 @@ metadata:
   name: simplest-collector
 status:
   readyReplicas: 1
-
 ---
 apiVersion: autoscaling/v1
 kind: HorizontalPodAutoscaler
@@ -16,4 +15,18 @@ spec:
 # This is not neccesarily exact.  We really just want to wait until this is no longer <unknown>
 status:
   currentCPUUtilizationPercentage: 20
-
+---
+apiVersion: autoscaling/v2beta2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: simplest-set-utilization-collector
+spec:
+  minReplicas: 1
+  maxReplicas: 2
+  metrics:
+  - type: Resource
+    resource:
+      name: cpu
+      target:
+        type: Utilization
+        averageUtilization: 50

tests/e2e/autoscale/00-install.yaml

@@ -36,3 +36,38 @@ spec:
           receivers: [otlp]
           processors: []
           exporters: [logging]
+---
+apiVersion: opentelemetry.io/v1alpha1
+kind: OpenTelemetryCollector
+metadata:
+  name: simplest-set-utilization
+spec:
+  minReplicas: 1
+  maxReplicas: 2
+  autoscaler:
+    targetCPUUtilization: 50
+  resources:
+    limits:
+      cpu: 500m
+      memory: 128Mi
+    requests:
+      cpu: 5m
+      memory: 64Mi
+
+  config: |
+    receivers:
+      otlp:
+        protocols:
+          grpc:
+          http:
+    processors:
+
+    exporters:
+      logging:
+
+    service:
+      pipelines:
+        traces:
+          receivers: [otlp]
+          processors: []
+          exporters: [logging]