fix(eventgenerator): use PromQL API for metric types throughput and responsetime

src/acceptance/app/dynamic_policy_test.go

@@ -4,6 +4,7 @@ import (
 	"acceptance"
 	. "acceptance/helpers"
 	"fmt"
+	"sync"
 	"time"
 
 	cfh "github.com/cloudfoundry/cf-test-helpers/v2/helpers"
@@ -19,6 +20,7 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 		doneAcceptChan chan bool
 		ticker         *time.Ticker
 		maxHeapLimitMb int
+		wg             sync.WaitGroup
 	)
 
 	const minimalMemoryUsage = 28 // observed minimal memory usage by the test app
@@ -30,6 +32,8 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 		Expect(err).NotTo(HaveOccurred())
 		StartApp(appName, cfg.CfPushTimeoutDuration())
 		instanceName = CreatePolicy(cfg, appName, appGUID, policy)
+
+		wg = sync.WaitGroup{}
 	})
 	BeforeEach(func() {
 		maxHeapLimitMb = cfg.NodeMemoryLimit - minimalMemoryUsage
@@ -105,12 +109,14 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 		Context("when responsetime is greater than scaling out threshold", func() {
 
 			BeforeEach(func() {
-				policy = GenerateDynamicScaleOutPolicy(1, 2, "responsetime", 2000)
+				policy = GenerateDynamicScaleOutPolicy(1, 2, "responsetime", 800)
 				initialInstanceCount = 1
 			})
 
 			JustBeforeEach(func() {
-				ticker = time.NewTicker(10 * time.Second)
+				// simulate ongoing ~20 slow requests per second
+				ticker = time.NewTicker(50 * time.Millisecond)
+				appUri := cfh.AppUri(appName, "/responsetime/slow/1000", cfg)
 				go func(chan bool) {
 					defer GinkgoRecover()
 					for {
@@ -120,7 +126,11 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 							doneAcceptChan <- true
 							return
 						case <-ticker.C:
-							cfh.CurlApp(cfg, appName, "/responsetime/slow/3000", "-f")
+							wg.Add(1)
+							go func() {
+								defer wg.Done()
+								cfh.Curl(cfg, appUri)
+							}()
 						}
 					}
 				}(doneChan)
@@ -129,17 +139,25 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 			It("should scale out", Label(acceptance.LabelSmokeTests), func() {
 				WaitForNInstancesRunning(appGUID, 2, 5*time.Minute)
 			})
+
+			AfterEach(func() {
+				// ginkgo requires all go-routines to be finished before reporting the result.
+				// let's wait for all curl executing go-routines to return.
+				wg.Wait()
+			})
 		})
 
 		Context("when responsetime is less than scaling in threshold", func() {
 
 			BeforeEach(func() {
-				policy = GenerateDynamicScaleInPolicy(1, 2, "responsetime", 1000)
+				policy = GenerateDynamicScaleInPolicyBetween("responsetime", 250, 350)
 				initialInstanceCount = 2
 			})
 
 			JustBeforeEach(func() {
-				ticker = time.NewTicker(2 * time.Second)
+				// simulate ongoing ~20 fast requests per second
+				ticker = time.NewTicker(50 * time.Millisecond)
+				appUri := cfh.AppUri(appName, "/responsetime/300", cfg)
 				go func(chan bool) {
 					defer GinkgoRecover()
 					for {
@@ -149,7 +167,11 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 							doneAcceptChan <- true
 							return
 						case <-ticker.C:
-							cfh.CurlApp(cfg, appName, "/responsetime/fast", "-f")
+							wg.Add(1)
+							go func() {
+								defer wg.Done()
+								cfh.Curl(cfg, appUri)
+							}()
 						}
 					}
 				}(doneChan)
@@ -158,12 +180,17 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 			It("should scale in", func() {
 				WaitForNInstancesRunning(appGUID, 1, 5*time.Minute)
 			})
+
+			AfterEach(func() {
+				// ginkgo requires all go-routines to be finished before reporting the result.
+				// let's wait for all curl executing go-routines to return.
+				wg.Wait()
+			})
 		})
 
 	})
 
 	Context("when scaling by throughput", func() {
-
 		JustBeforeEach(func() {
 			doneChan = make(chan bool)
 			doneAcceptChan = make(chan bool)
@@ -177,12 +204,14 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 		Context("when throughput is greater than scaling out threshold", func() {
 
 			BeforeEach(func() {
-				policy = GenerateDynamicScaleOutPolicy(1, 2, "throughput", 1)
+				policy = GenerateDynamicScaleOutPolicy(1, 2, "throughput", 15)
 				initialInstanceCount = 1
 			})
 
 			JustBeforeEach(func() {
-				ticker = time.NewTicker(25 * time.Millisecond)
+				// simulate ongoing ~20 requests per second
+				ticker = time.NewTicker(50 * time.Millisecond)
+				appUri := cfh.AppUri(appName, "/responsetime/fast", cfg)
 				go func(chan bool) {
 					defer GinkgoRecover()
 					for {
@@ -192,7 +221,11 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 							doneAcceptChan <- true
 							return
 						case <-ticker.C:
-							cfh.CurlApp(cfg, appName, "/responsetime/fast", "-f")
+							wg.Add(1)
+							go func() {
+								defer wg.Done()
+								cfh.Curl(cfg, appUri)
+							}()
 						}
 					}
 				}(doneChan)
@@ -202,17 +235,25 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 				WaitForNInstancesRunning(appGUID, 2, 5*time.Minute)
 			})
 
+			AfterEach(func() {
+				// ginkgo requires all go-routines to be finished before reporting the result.
+				// let's wait for all curl executing go-routines to return.
+				wg.Wait()
+			})
+
 		})
 
 		Context("when throughput is less than scaling in threshold", func() {
 
 			BeforeEach(func() {
-				policy = GenerateDynamicScaleInPolicy(1, 2, "throughput", 100)
+				policy = GenerateDynamicScaleInPolicyBetween("throughput", 15, 25)
 				initialInstanceCount = 2
 			})
 
 			JustBeforeEach(func() {
-				ticker = time.NewTicker(10 * time.Second)
+				// simulate ongoing ~20 requests per second
+				ticker = time.NewTicker(50 * time.Millisecond)
+				appUri := cfh.AppUri(appName, "/responsetime/fast", cfg)
 				go func(chan bool) {
 					defer GinkgoRecover()
 					for {
@@ -222,16 +263,26 @@ var _ = Describe("AutoScaler dynamic policy", func() {
 							doneAcceptChan <- true
 							return
 						case <-ticker.C:
-							cfh.CurlApp(cfg, appName, "/responsetime/fast", "-f")
+							wg.Add(1)
+							go func() {
+								defer wg.Done()
+								cfh.Curl(cfg, appUri)
+							}()
 						}
 					}
 				}(doneChan)
 			})
+
 			It("should scale in", func() {
 				WaitForNInstancesRunning(appGUID, 1, 5*time.Minute)
 			})
-		})
 
+			AfterEach(func() {
+				// ginkgo requires all go-routines to be finished before reporting the result.
+				// let's wait for all curl executing go-routines to return.
+				wg.Wait()
+			})
+		})
 	})
 
 	// To check existing aggregated cpu metrics do: cf asm APP_NAME cpu

src/acceptance/helpers/helpers.go

@@ -276,6 +276,50 @@ func GenerateDynamicScaleOutAndInPolicy(instanceMin, instanceMax int, metricName
 	return string(marshaled)
 }
 
+func GenerateDynamicScaleInPolicyBetween(metricName string, scaleInLowerThreshold int64, scaleInUpperThreshold int64) string {
+	// < 10 => +1 => don't do anything if below 10
+	// < 30 => -1 => scale down if less than 30
+	// > 30 => +1 => don't do anything if above 30
+	noDownscalingWhenBelowLower := ScalingRule{
+		MetricType:            metricName,
+		BreachDurationSeconds: TestBreachDurationSeconds,
+		Threshold:             scaleInLowerThreshold,
+		Operator:              "<",
+		CoolDownSeconds:       TestCoolDownSeconds,
+		Adjustment:            "+1",
+	}
+
+	downscalingWhenBelowUpper := ScalingRule{
+		MetricType:            metricName,
+		BreachDurationSeconds: TestBreachDurationSeconds,
+		Threshold:             scaleInUpperThreshold,
+		Operator:              "<",
+		CoolDownSeconds:       TestCoolDownSeconds,
+		Adjustment:            "-1",
+	}
+
+	noDownscalingWhenAboveUpper := ScalingRule{
+		MetricType:            metricName,
+		BreachDurationSeconds: TestBreachDurationSeconds,
+		Threshold:             scaleInUpperThreshold,
+		Operator:              ">=",
+		CoolDownSeconds:       TestCoolDownSeconds,
+		Adjustment:            "+1",
+	}
+
+	policy := ScalingPolicy{
+		InstanceMin:  1,
+		InstanceMax:  2,
+		ScalingRules: []*ScalingRule{&noDownscalingWhenBelowLower, &downscalingWhenBelowUpper, &noDownscalingWhenAboveUpper},
+	}
+
+	marshaled, err := MarshalWithoutHTMLEscape(policy)
+	Expect(err).NotTo(HaveOccurred())
+
+	return string(marshaled)
+
+}
+
 func GenerateSpecificDateSchedulePolicy(startDateTime, endDateTime time.Time, scheduledInstanceMin, scheduledInstanceMax, scheduledInstanceInit int) string {
 	scalingInRule := ScalingRule{
 		MetricType:            "cpu",

src/autoscaler/envelopeprocessor/envelope_processor.go

@@ -20,7 +20,7 @@ type EnvelopeProcessorCreator interface {
 
 type EnvelopeProcessor interface {
 	GetGaugeMetrics(envelopes []*loggregator_v2.Envelope, currentTimeStamp int64) ([]models.AppInstanceMetric, error)
-	GetTimerMetrics(envelopes []*loggregator_v2.Envelope, appID string, currentTimestamp int64) []models.AppInstanceMetric
+	GetCollectionInterval() time.Duration
 }
 
 var _ EnvelopeProcessor = &Processor{}
@@ -43,10 +43,9 @@ func (p Processor) GetGaugeMetrics(envelopes []*loggregator_v2.Envelope, current
 	p.logger.Debug("Compacted envelopes", lager.Data{"compactedEnvelopes": compactedEnvelopes})
 	return GetGaugeInstanceMetrics(compactedEnvelopes, currentTimeStamp)
 }
-func (p Processor) GetTimerMetrics(envelopes []*loggregator_v2.Envelope, appID string, currentTimestamp int64) []models.AppInstanceMetric {
-	p.logger.Debug("GetTimerMetrics")
-	p.logger.Debug("Compacted envelopes", lager.Data{"Envelopes": envelopes})
-	return GetHttpStartStopInstanceMetrics(envelopes, appID, currentTimestamp, p.collectionInterval)
+
+func (p Processor) GetCollectionInterval() time.Duration {
+	return p.collectionInterval
 }
 
 // Log cache returns instance metrics such as cpu and memory in serparate envelopes, this was not the case with
@@ -91,10 +90,10 @@ func GetHttpStartStopInstanceMetrics(envelopes []*loggregator_v2.Envelope, appID
 	var metrics []models.AppInstanceMetric
 
 	numRequestsPerAppIdx := calcNumReqs(envelopes)
-	sumReponseTimesPerAppIdx := calcSumResponseTimes(envelopes)
+	sumResponseTimesPerAppIdx := calcSumResponseTimes(envelopes)
 
 	throughputMetrics := getThroughputInstanceMetrics(envelopes, appID, numRequestsPerAppIdx, collectionInterval, currentTimestamp)
-	responseTimeMetric := getResponsetimeInstanceMetrics(envelopes, appID, numRequestsPerAppIdx, sumReponseTimesPerAppIdx, currentTimestamp)
+	responseTimeMetric := getResponsetimeInstanceMetrics(envelopes, appID, numRequestsPerAppIdx, sumResponseTimesPerAppIdx, currentTimestamp)
 
 	metrics = append(metrics, throughputMetrics...)
 	metrics = append(metrics, responseTimeMetric...)