Merge pull request #22171 from piosz/hpa-default

Added default cpu target for HPA

api/swagger-spec/autoscaling_v1.json

@@ -1001,7 +1001,7 @@
      "targetCPUUtilizationPercentage": {
       "type": "integer",
       "format": "int32",
-      "description": "target average CPU utilization (represented as a percentage of requested CPU) over all the pods;"
+      "description": "target average CPU utilization (represented as a percentage of requested CPU) over all the pods; if not specified the default autoscaling policy will be used."
      }
     }
    },

pkg/apis/autoscaling/v1/types.go

@@ -41,6 +41,7 @@ type HorizontalPodAutoscalerSpec struct {
 	// upper limit for the number of pods that can be set by the autoscaler; cannot be smaller than MinReplicas.
 	MaxReplicas int32 `json:"maxReplicas"`
 	// target average CPU utilization (represented as a percentage of requested CPU) over all the pods;
+	// if not specified the default autoscaling policy will be used.
 	TargetCPUUtilizationPercentage *int32 `json:"targetCPUUtilizationPercentage,omitempty"`
 }
 

pkg/apis/autoscaling/v1/types_swagger_doc_generated.go

@@ -64,7 +64,7 @@ var map_HorizontalPodAutoscalerSpec = map[string]string{
 	"scaleTargetRef":                 "reference to scaled resource; horizontal pod autoscaler will learn the current resource consumption and will set the desired number of pods by using its Scale subresource.",
 	"minReplicas":                    "lower limit for the number of pods that can be set by the autoscaler, default 1.",
 	"maxReplicas":                    "upper limit for the number of pods that can be set by the autoscaler; cannot be smaller than MinReplicas.",
-	"targetCPUUtilizationPercentage": "target average CPU utilization (represented as a percentage of requested CPU) over all the pods;",
+	"targetCPUUtilizationPercentage": "target average CPU utilization (represented as a percentage of requested CPU) over all the pods; if not specified the default autoscaling policy will be used.",
 }
 
 func (HorizontalPodAutoscalerSpec) SwaggerDoc() map[string]string {

pkg/controller/podautoscaler/horizontal.go

@@ -39,6 +39,8 @@ const (
 	// TODO: make it a flag or HPA spec element.
 	tolerance = 0.1
 
+	defaultTargetCPUUtilizationPercentage = 80
+
 	HpaCustomMetricsTargetAnnotationName = "alpha/target.custom-metrics.podautoscaler.kubernetes.io"
 	HpaCustomMetricsStatusAnnotationName = "alpha/status.custom-metrics.podautoscaler.kubernetes.io"
 )
@@ -76,11 +78,9 @@ func (a *HorizontalController) Run(syncPeriod time.Duration) {
 }
 
 func (a *HorizontalController) computeReplicasForCPUUtilization(hpa extensions.HorizontalPodAutoscaler, scale *extensions.Scale) (int, *int, time.Time, error) {
-	if hpa.Spec.CPUUtilization == nil {
-		// If CPUTarget is not specified than we should return some default values.
-		// Since we always take maximum number of replicas from all policies it is safe
-		// to just return 0.
-		return 0, nil, time.Time{}, nil
+	targetUtilization := defaultTargetCPUUtilizationPercentage
+	if hpa.Spec.CPUUtilization != nil {
+		targetUtilization = hpa.Spec.CPUUtilization.TargetPercentage
 	}
 	currentReplicas := scale.Status.Replicas
 	currentUtilization, timestamp, err := a.metricsClient.GetCPUUtilization(hpa.Namespace, scale.Status.Selector)
@@ -91,7 +91,7 @@ func (a *HorizontalController) computeReplicasForCPUUtilization(hpa extensions.H
 		return 0, nil, time.Time{}, fmt.Errorf("failed to get cpu utilization: %v", err)
 	}
 
-	usageRatio := float64(*currentUtilization) / float64(hpa.Spec.CPUUtilization.TargetPercentage)
+	usageRatio := float64(*currentUtilization) / float64(targetUtilization)
 	if math.Abs(1.0-usageRatio) > tolerance {
 		return int(math.Ceil(usageRatio * float64(currentReplicas))), currentUtilization, timestamp, nil
 	} else {
@@ -192,8 +192,9 @@ func (a *HorizontalController) reconcileAutoscaler(hpa extensions.HorizontalPodA
 		desiredReplicas = 1
 	} else {
 		// All basic scenarios covered, the state should be sane, lets use metrics.
+		cmAnnotation, cmAnnotationFound := hpa.Annotations[HpaCustomMetricsTargetAnnotationName]
 
-		if hpa.Spec.CPUUtilization != nil {
+		if hpa.Spec.CPUUtilization != nil || !cmAnnotationFound {
 			cpuDesiredReplicas, cpuCurrentUtilization, cpuTimestamp, err = a.computeReplicasForCPUUtilization(hpa, scale)
 			if err != nil {
 				a.updateCurrentReplicasInStatus(hpa, currentReplicas)
@@ -202,7 +203,7 @@ func (a *HorizontalController) reconcileAutoscaler(hpa extensions.HorizontalPodA
 			}
 		}
 
-		if cmAnnotation, cmAnnotationFound := hpa.Annotations[HpaCustomMetricsTargetAnnotationName]; cmAnnotationFound {
+		if cmAnnotationFound {
 			cmDesiredReplicas, cmStatus, cmTimestamp, err = a.computeReplicasForCustomMetrics(hpa, scale, cmAnnotation)
 			if err != nil {
 				a.updateCurrentReplicasInStatus(hpa, currentReplicas)

pkg/controller/podautoscaler/horizontal_test.go

@@ -257,6 +257,19 @@ func (tc *testCase) runTest(t *testing.T) {
 	tc.verifyResults(t)
 }
 
+func TestDefaultScaleUp(t *testing.T) {
+	tc := testCase{
+		minReplicas:         2,
+		maxReplicas:         6,
+		initialReplicas:     4,
+		desiredReplicas:     5,
+		verifyCPUCurrent:    true,
+		reportedLevels:      []uint64{900, 950, 950, 1000},
+		reportedCPURequests: []resource.Quantity{resource.MustParse("1.0"), resource.MustParse("1.0"), resource.MustParse("1.0"), resource.MustParse("1.0")},
+	}
+	tc.runTest(t)
+}
+
 func TestScaleUp(t *testing.T) {
 	tc := testCase{
 		minReplicas:         2,
@@ -290,6 +303,19 @@ func TestScaleUpCM(t *testing.T) {
 	tc.runTest(t)
 }
 
+func TestDefaultScaleDown(t *testing.T) {
+	tc := testCase{
+		minReplicas:         2,
+		maxReplicas:         6,
+		initialReplicas:     5,
+		desiredReplicas:     4,
+		verifyCPUCurrent:    true,
+		reportedLevels:      []uint64{400, 500, 600, 700, 800},
+		reportedCPURequests: []resource.Quantity{resource.MustParse("1.0"), resource.MustParse("1.0"), resource.MustParse("1.0"), resource.MustParse("1.0"), resource.MustParse("1.0")},
+	}
+	tc.runTest(t)
+}
+
 func TestScaleDown(t *testing.T) {
 	tc := testCase{
 		minReplicas:         2,