test: add test to measure rate of growth of metrics

Signed-off-by: Alex Castilio dos Santos <[email protected]>

deploy/legacy/prometheus/values.yaml

@@ -67,3 +67,15 @@ prometheus:
           - source_labels: [__name__]
             action: keep
             regex: (.*)
+    resources:
+      limits:
+        memory: 10Gi
+    enableAdminAPI: true
+    storageSpec: 
+     volumeClaimTemplate:
+       spec:
+         # storageClassName: gp2
+         accessModes: ["ReadWriteOnce"]
+         resources:
+           requests:
+             storage: 50Gi

test/e2e/common/common.go

@@ -16,12 +16,14 @@ import (
 )
 
 const (
-	RetinaPort int = 10093
+	RetinaPort     int = 10093
+	PrometheusPort int = 9090
 	// netObsRGtag is used to tag resources created by this test suite
 	NetObsRGtag            = "-e2e-netobs-"
 	KubeSystemNamespace    = "kube-system"
 	TestPodNamespace       = "kube-system-test"
 	AzureAppInsightsKeyEnv = "AZURE_APP_INSIGHTS_KEY"
+	OutputFilePathEnv      = "OUTPUT_FILEPATH"
 )
 
 var (

test/e2e/framework/metrics/query-publish.go

@@ -0,0 +1,174 @@
+package metrics
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io/fs"
+	"log"
+	"os"
+	"sync"
+	"time"
+
+	"github.com/microsoft/retina/pkg/telemetry"
+	"github.com/microsoft/retina/test/e2e/common"
+	prom_client "github.com/prometheus/client_golang/api"
+	prom_v1 "github.com/prometheus/client_golang/api/prometheus/v1"
+	prom_model "github.com/prometheus/common/model"
+)
+
+type QueryAndPublish struct {
+	Query                       string
+	Endpoint                    string
+	AdditionalTelemetryProperty map[string]string
+	outputFilePath              string
+	stop                        chan struct{}
+	wg                          sync.WaitGroup
+	telemetryClient             *telemetry.TelemetryClient
+	appInsightsKey              string
+}
+
+func (q *QueryAndPublish) Run() error {
+	if q.appInsightsKey != "" {
+		telemetry.InitAppInsights(q.appInsightsKey, q.AdditionalTelemetryProperty["retinaVersion"])
+
+		telemetryClient, err := telemetry.NewAppInsightsTelemetryClient("retina-rate-of-growth", q.AdditionalTelemetryProperty)
+		if err != nil {
+			return fmt.Errorf("error creating telemetry client: %w", err)
+		}
+
+		q.telemetryClient = telemetryClient
+	}
+
+	q.stop = make(chan struct{})
+	q.wg.Add(1)
+
+	go func() {
+
+		t := time.NewTicker(2 * time.Second)
+
+		// First execution
+		err := q.getAndPublishMetrics()
+		if err != nil {
+			log.Fatalf("error getting and publishing metrics: %v", err)
+			return
+		}
+
+		for {
+			select {
+
+			case <-t.C:
+				err := q.getAndPublishMetrics()
+				if err != nil {
+					log.Fatalf("error getting and publishing metrics: %v", err)
+					return
+				}
+
+			case <-q.stop:
+				q.wg.Done()
+				return
+
+			}
+		}
+
+	}()
+
+	return nil
+}
+
+func (q *QueryAndPublish) getAndPublishMetrics() error {
+	// ctx, cancel := context.WithTimeout(context.Background(), defaultTimeoutSeconds*time.Second)
+	// defer cancel()
+
+	client, err := prom_client.NewClient(prom_client.Config{
+		Address: q.Endpoint,
+	})
+	if err != nil {
+		return fmt.Errorf("error creating prometheus client: %w", err)
+	}
+
+	promApi := prom_v1.NewAPI(client)
+	ctx := context.TODO()
+
+	result, warnings, err := promApi.Query(ctx, q.Query, time.Now())
+	if err != nil {
+		return fmt.Errorf("error querying prometheus: %w", err)
+	}
+	if len(warnings) > 0 {
+		log.Println("query warnings: ", warnings)
+	}
+	type metrics map[string]string
+
+	allMetrics := []metrics{}
+
+	for _, sample := range result.(prom_model.Vector) {
+		instance := string(sample.Metric["instance"])
+		samplesScraped := sample.Value.String()
+
+		m := map[string]string{
+			"instance":       instance,
+			"samplesScraped": samplesScraped,
+		}
+		allMetrics = append(allMetrics, m)
+	}
+
+	// Publish metrics
+	if q.telemetryClient != nil {
+		log.Println("Publishing metrics to AppInsights")
+		for _, metric := range allMetrics {
+			q.telemetryClient.TrackEvent("metrics-scraped", metric)
+
+		}
+	}
+
+	// Write metrics to file
+	if q.outputFilePath != "" {
+		log.Println("Writing metrics to file ", q.outputFilePath)
+
+		permissions := 0o644
+		file, err := os.OpenFile(q.outputFilePath, os.O_APPEND|os.O_CREATE|os.O_WRONLY, fs.FileMode(permissions))
+		if err != nil {
+			return fmt.Errorf("error writing to csv file: %w", err)
+		}
+		defer file.Close()
+
+		for _, m := range allMetrics {
+			b, err := json.Marshal(m)
+			if err != nil {
+				return fmt.Errorf("error marshalling metric: %w", err)
+			}
+			file.Write(b)
+			file.WriteString("\n")
+		}
+
+	}
+
+	return nil
+}
+
+func (q *QueryAndPublish) Stop() error {
+	telemetry.ShutdownAppInsights()
+	close(q.stop)
+	q.wg.Wait()
+	return nil
+}
+
+func (q *QueryAndPublish) Prevalidate() error {
+	if os.Getenv(common.AzureAppInsightsKeyEnv) == "" {
+		log.Println("env ", common.AzureAppInsightsKeyEnv, " not provided")
+	}
+	q.appInsightsKey = os.Getenv(common.AzureAppInsightsKeyEnv)
+
+	if _, ok := q.AdditionalTelemetryProperty["retinaVersion"]; !ok {
+		return fmt.Errorf("retinaVersion is required in AdditionalTelemetryProperty")
+	}
+
+	if os.Getenv(common.OutputFilePathEnv) == "" {
+		log.Println("Output file path not provided. Metrics will not be written to file")
+		return nil
+	}
+	q.outputFilePath = os.Getenv(common.OutputFilePathEnv)
+
+	log.Println("Output file path provided: ", q.outputFilePath)
+	return nil
+}

test/e2e/growth_metrics_test.go

@@ -0,0 +1,112 @@
+package retina
+
+import (
+	"os"
+	"path/filepath"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/microsoft/retina/test/e2e/common"
+	"github.com/microsoft/retina/test/e2e/framework/azure"
+	"github.com/microsoft/retina/test/e2e/framework/generic"
+	"github.com/microsoft/retina/test/e2e/framework/helpers"
+	"github.com/microsoft/retina/test/e2e/framework/kubernetes"
+	"github.com/microsoft/retina/test/e2e/framework/metrics"
+	"github.com/microsoft/retina/test/e2e/framework/types"
+	"github.com/stretchr/testify/require"
+)
+
+func GetKubeconfig(clusterName, subscriptionId, resourceGroup, kubeConfigFilePath string) *types.Job {
+	job := types.NewJob("Get kubeconfig")
+	job.AddStep(&azure.GetAKSKubeConfig{
+		ClusterName:        clusterName,
+		SubscriptionID:     subscriptionId,
+		ResourceGroupName:  resourceGroup,
+		Location:           "why?",
+		KubeConfigFilePath: kubeConfigFilePath,
+	}, nil)
+	return job
+}
+
+func GrowthTest(additionalTelemetryProperty map[string]string, kubeConfigFilePath string) *types.Job {
+	job := types.NewJob("Growth Test")
+	labelAffinity := "app.kubernetes.io/instance=prometheus-kube-prometheus-prometheus"
+	portForwardId := "port-forward"
+	metricsStepId := "metrics"
+
+	job.AddStep(&kubernetes.PortForward{
+		KubeConfigFilePath:    kubeConfigFilePath,
+		Namespace:             common.KubeSystemNamespace,
+		LabelSelector:         "app.kubernetes.io/instance=prometheus-kube-prometheus-prometheus",
+		LocalPort:             strconv.Itoa(common.PrometheusPort),
+		RemotePort:            strconv.Itoa(common.PrometheusPort),
+		Endpoint:              "metrics",
+		OptionalLabelAffinity: labelAffinity,
+	},
+		&types.StepOptions{
+			SkipSavingParametersToJob: true,
+			RunInBackgroundWithID:     portForwardId,
+		})
+
+	job.AddStep(&metrics.QueryAndPublish{
+		Endpoint:                    "http://localhost:" + strconv.Itoa(common.PrometheusPort),
+		Query:                       "scrape_samples_scraped{job=\"retina-pods\"}",
+		AdditionalTelemetryProperty: additionalTelemetryProperty,
+	},
+		&types.StepOptions{
+			SkipSavingParametersToJob: true,
+			RunInBackgroundWithID:     metricsStepId,
+		})
+
+	job.AddStep(&types.Sleep{
+		Duration: 60 * time.Second,
+	}, nil)
+
+	job.AddStep(
+		&types.Stop{
+			BackgroundID: metricsStepId,
+		}, nil)
+
+	job.AddStep(
+		&types.Stop{
+			BackgroundID: portForwardId,
+		}, nil)
+	return job
+}
+
+func Test_GrowthOfMetrics(t *testing.T) {
+	ctx, cancel := helpers.Context(t)
+	defer cancel()
+
+	clusterName := common.ClusterNameForE2ETest(t)
+
+	subID := os.Getenv("AZURE_SUBSCRIPTION_ID")
+	require.NotEmpty(t, subID)
+
+	rg := os.Getenv("AZURE_RESOURCE_GROUP")
+	if rg == "" {
+		// Use the cluster name as the resource group name by default.
+		rg = clusterName
+	}
+
+	RetinaVersion := os.Getenv(generic.DefaultTagEnv)
+	require.NotEmpty(t, RetinaVersion)
+
+	additionalTelemetryProperty := map[string]string{}
+	additionalTelemetryProperty["retinaVersion"] = RetinaVersion
+	additionalTelemetryProperty["clusterName"] = clusterName
+	additionalTelemetryProperty["resourceGroup"] = rg
+
+	cwd, err := os.Getwd()
+	require.NoError(t, err)
+
+	rootDir := filepath.Dir(filepath.Dir(cwd))
+	kubeConfigFilePath := filepath.Join(rootDir, "test", "e2e", "test.pem")
+
+	getKubeconfig := types.NewRunner(t, GetKubeconfig(clusterName, subID, rg, kubeConfigFilePath))
+	getKubeconfig.Run(ctx)
+
+	growth := types.NewRunner(t, GrowthTest(additionalTelemetryProperty, kubeConfigFilePath))
+	growth.Run(ctx)
+}