add targetCPUUTilization to HPA

open-telemetry · Aug 30, 2022 · 012fd4a · 012fd4a
1 parent 81b51b1
commit 012fd4a
Show file tree

Hide file tree

Showing 10 changed files with 61 additions and 3 deletions.
diff --git a/apis/v1alpha1/opentelemetrycollector_types.go b/apis/v1alpha1/opentelemetrycollector_types.go
@@ -40,6 +40,10 @@ 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"`
+	// 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"`
 	// SecurityContext will be set as the container security context.
 	// +optional
 	SecurityContext *v1.SecurityContext `json:"securityContext,omitempty"`

diff --git a/apis/v1alpha1/opentelemetrycollector_webhook.go b/apis/v1alpha1/opentelemetrycollector_webhook.go
@@ -129,6 +129,10 @@ func (r *OpenTelemetryCollector) validateCRDSpec() error {
 			return fmt.Errorf("the OpenTelemetry Spec autoscale configuration is incorrect, minReplicas should be one or more")
 		}
 
+		if r.Spec.TargetCPUUtilization != nil && (*r.Spec.TargetCPUUtilization < int32(1) || *r.Spec.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

diff --git a/apis/v1alpha1/zz_generated.deepcopy.go b/apis/v1alpha1/zz_generated.deepcopy.go
diff --git a/bundle/manifests/opentelemetry.io_opentelemetrycollectors.yaml b/bundle/manifests/opentelemetry.io_opentelemetrycollectors.yaml
@@ -708,6 +708,12 @@ spec:
                       service account to use with this instance.
                     type: string
                 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
               tolerations:
                 description: Toleration to schedule OpenTelemetry Collector pods.
                   This is only relevant to daemonset, statefulset, and deployment

diff --git a/config/crd/bases/opentelemetry.io_opentelemetrycollectors.yaml b/config/crd/bases/opentelemetry.io_opentelemetrycollectors.yaml
@@ -706,6 +706,12 @@ spec:
                       service account to use with this instance.
                     type: string
                 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
               tolerations:
                 description: Toleration to schedule OpenTelemetry Collector pods.
                   This is only relevant to daemonset, statefulset, and deployment

diff --git a/docs/api.md b/docs/api.md
@@ -1825,6 +1825,16 @@ OpenTelemetryCollectorSpec defines the desired state of OpenTelemetryCollector.
           TargetAllocator indicates a value which determines whether to spawn a target allocation resource or not.<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><tr>
         <td><b><a href="#opentelemetrycollectorspectolerationsindex">tolerations</a></b></td>
         <td>[]object</td>

diff --git a/pkg/collector/horizontalpodautoscaler.go b/pkg/collector/horizontalpodautoscaler.go
@@ -37,7 +37,12 @@ func HorizontalPodAutoscaler(cfg config.Config, logger logr.Logger, otelcol v1al
 	labels["app.kubernetes.io/name"] = naming.Collector(otelcol)
 
 	annotations := Annotations(otelcol)
-	cpuTarget := defaultCPUTarget
+	var cpuTarget int32
+	if otelcol.Spec.TargetCPUUtilization != nil {
+		cpuTarget = *otelcol.Spec.TargetCPUUtilization
+	} else {
+		cpuTarget = defaultCPUTarget
+	}
 	var result client.Object
 
 	objectMeta := metav1.ObjectMeta{

diff --git a/pkg/collector/reconcile/horizontalpodautoscaler.go b/pkg/collector/reconcile/horizontalpodautoscaler.go
@@ -111,6 +111,7 @@ func expectedHorizontalPodAutoscalers(ctx context.Context, params Params, expect
 
 func setAutoscalerSpec(params Params, autoscalingVersion autodetect.AutoscalingVersion, updated client.Object) {
 	one := int32(1)
+	ninety := int32(90)
 	if params.Instance.Spec.MaxReplicas != nil {
 		if autoscalingVersion == autodetect.AutoscalingVersionV2Beta2 {
 			updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.MaxReplicas = *params.Instance.Spec.MaxReplicas
@@ -119,13 +120,23 @@ func setAutoscalerSpec(params Params, autoscalingVersion autodetect.AutoscalingV
 			} else {
 				updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.MinReplicas = &one
 			}
+			if params.Instance.Spec.TargetCPUUtilization != nil {
+				updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = params.Instance.Spec.TargetCPUUtilization
+			} else {
+				updated.(*autoscalingv2beta2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = &ninety
+			}
 		} 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
 			}
+			if params.Instance.Spec.TargetCPUUtilization != nil {
+				updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = params.Instance.Spec.TargetCPUUtilization
+			} else {
+				updated.(*autoscalingv2.HorizontalPodAutoscaler).Spec.Metrics[0].Resource.Target.AverageUtilization = &ninety
+			}
 		}
 	}
 }

diff --git a/tests/e2e/autoscale/00-assert.yaml b/tests/e2e/autoscale/00-assert.yaml
@@ -6,11 +6,17 @@ status:
   readyReplicas: 1
 
 ---
-apiVersion: autoscaling/v1
+apiVersion: autoscaling/v2beta2
 kind: HorizontalPodAutoscaler
 metadata:
   name: simplest-collector
 spec:
   minReplicas: 1
   maxReplicas: 2
-
+  metrics:
+  - type: Resource
+    resource:
+      name: cpu
+      target:
+        type: Utilization
+        averageUtilization: 50
diff --git a/tests/e2e/autoscale/00-install.yaml b/tests/e2e/autoscale/00-install.yaml
@@ -5,6 +5,7 @@ metadata:
 spec:
   minReplicas: 1
   maxReplicas: 2
+  targetCPUUtilization: 50
   resources:
     limits:
       cpu: 500m