test: automate scale test execution

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

.github/workflows/daily-scale-test.yaml

@@ -0,0 +1,23 @@
+name: Daily Scale Test
+
+on:
+  push:
+    branches:
+      - alexcastilio/scale-test-workflow
+  # schedule:
+    # - cron: "0 0 * * *"
+
+permissions:
+  contents: read
+  id-token: write
+
+jobs:
+  call-scale-test:
+    uses: ./.github/workflows/scale-test.yaml
+    with: 
+      num_deployments: 10
+      num_replicas: 10
+        # TODO: Fix value
+      num_netpol: 0
+      cleanup: false
+    secrets: inherit

.github/workflows/scale-test.yaml

@@ -36,25 +36,17 @@ on:
 
   workflow_call:
     inputs:
-      resource_group:
-        description: "Azure Resource Group"
-        required: true
-        type: string
-      cluster_name:
-        description: "AKS Cluster Name"
-        required: true
-        type: string
       num_deployments:
         description: "Number of Traffic Deployments"
-        default: 1000
+        default: 100
         type: number
       num_replicas:
         description: "Number of Traffic Replicas per Deployment"
-        default: 40
+        default: 10
         type: number
       num_netpol: 
         description: "Number of Network Policies"
-        default: 1000
+        default: 100
         type: number
       cleanup: 
         description: "Clean up environment after test"
@@ -100,8 +92,14 @@ jobs:
           IMAGE_NAMESPACE: ${{ github.repository }}
           TAG: ${{ inputs.image_tag }}
           AZURE_APP_INSIGHTS_KEY: ${{ secrets.AZURE_APP_INSIGHTS_KEY }}
+          # TODO: FIX VALUE
+          NODES_PER_POOL: ""
+          # TODO: DELETE
+          WORKFLOW_NAME: ${{ github.event_name }}
+          WORKFLOW_EVENT: ${{ github.event }}
+          WORKFLOW_CALL: ${{ github.event.workflow == '.github/workflows/daily-scale-test.yaml' }}
         shell: bash
         run: |
           set -euo pipefail
           [[ $TAG == "" ]] && TAG=$(make version)
-          go test -v ./test/e2e/. -timeout 300m -tags=scale -count=1  -args -create-infra=false -delete-infra=false 
+          go test -v ./test/e2e/. -timeout 300m -tags=scale -count=1  -args -create-infra=$(echo $WORKFLOW_CALL) -delete-infra=false

test/e2e/common/common.go

@@ -6,12 +6,12 @@ package common
 
 import (
 	"flag"
-	"os"
 	"os/user"
 	"strconv"
 	"testing"
 	"time"
 
+	"github.com/microsoft/retina/test/e2e/framework/params"
 	"github.com/stretchr/testify/require"
 )
 
@@ -30,10 +30,61 @@ var (
 	Architectures  = []string{"amd64", "arm64"}
 	CreateInfra    = flag.Bool("create-infra", true, "create a Resource group, vNET and AKS cluster for testing")
 	DeleteInfra    = flag.Bool("delete-infra", true, "delete a Resource group, vNET and AKS cluster for testing")
+	ScaleTestInfra = ScaleTestInfraHandler{
+		location:       params.Location,
+		subscriptionID: params.SubscriptionID,
+		resourceGroup:  params.ResourceGroup,
+		clusterName:    params.ClusterName,
+		nodesPerPool:   params.NodesPerPool,
+	}
 )
 
+type ScaleTestInfraHandler struct {
+	location       string
+	subscriptionID string
+	resourceGroup  string
+	clusterName    string
+	nodesPerPool   string
+}
+
+func (s ScaleTestInfraHandler) GetSubscriptionID(t *testing.T) string {
+	require.NotEmpty(t, s.subscriptionID)
+	return s.subscriptionID
+}
+
+func (s ScaleTestInfraHandler) GetLocation(t *testing.T) string {
+	if s.location == "" {
+		return "westus2"
+	}
+	return s.location
+}
+
+func (s ScaleTestInfraHandler) GetResourceGroup(t *testing.T) string {
+	if s.resourceGroup != "" {
+		return s.resourceGroup
+	}
+	// Use the cluster name as the resource group name by default.
+	return s.GetClusterName(t)
+}
+
+func (s ScaleTestInfraHandler) GetNodesPerPool(t *testing.T) int32 {
+	if s.nodesPerPool == "" {
+		return 5
+	}
+	nodesPerPool, err := strconv.Atoi(s.nodesPerPool)
+	require.NoError(t, err, "NODES_PER_POOL must be an integer")
+	return int32(nodesPerPool)
+}
+
+func (s ScaleTestInfraHandler) GetClusterName(t *testing.T) string {
+	if s.clusterName != "" {
+		return s.clusterName
+	}
+	return "retina-scale-test"
+}
+
 func ClusterNameForE2ETest(t *testing.T) string {
-	clusterName := os.Getenv("CLUSTER_NAME")
+	clusterName := params.ClusterName
 	if clusterName == "" {
 		curuser, err := user.Current()
 		require.NoError(t, err)

test/e2e/framework/azure/create-cluster-with-npm.go

@@ -22,7 +22,6 @@ const (
 	clusterCreateTicker = 30 * time.Second
 	pollFrequency       = 5 * time.Second
 	AgentARMSKU         = "Standard_D4pls_v5"
-	AuxilaryNodeCount   = 1
 )
 
 type CreateNPMCluster struct {
@@ -35,6 +34,7 @@ type CreateNPMCluster struct {
 	PodCidr           string
 	DNSServiceIP      string
 	ServiceCidr       string
+	NodesPerPool      int32
 }
 
 func (c *CreateNPMCluster) Prevalidate() error {
@@ -55,7 +55,7 @@ func (c *CreateNPMCluster) Run() error {
 	npmCluster.Properties.AgentPoolProfiles = append(npmCluster.Properties.AgentPoolProfiles, &armcontainerservice.ManagedClusterAgentPoolProfile{ //nolint:all
 		Type: to.Ptr(armcontainerservice.AgentPoolTypeVirtualMachineScaleSets),
 		// AvailabilityZones:  []*string{to.Ptr("1")},
-		Count:              to.Ptr[int32](AuxilaryNodeCount),
+		Count:              to.Ptr[int32](c.NodesPerPool),
 		EnableNodePublicIP: to.Ptr(false),
 		Mode:               to.Ptr(armcontainerservice.AgentPoolModeUser),
 		OSType:             to.Ptr(armcontainerservice.OSTypeWindows),
@@ -86,7 +86,7 @@ func (c *CreateNPMCluster) Run() error {
 	npmCluster.Properties.AgentPoolProfiles = append(npmCluster.Properties.AgentPoolProfiles, &armcontainerservice.ManagedClusterAgentPoolProfile{ //nolint:all
 		Type: to.Ptr(armcontainerservice.AgentPoolTypeVirtualMachineScaleSets),
 		// AvailabilityZones:  []*string{to.Ptr("1")},
-		Count:              to.Ptr[int32](AuxilaryNodeCount),
+		Count:              to.Ptr[int32](c.NodesPerPool),
 		EnableNodePublicIP: to.Ptr(false),
 		Mode:               to.Ptr(armcontainerservice.AgentPoolModeUser),
 		OSType:             to.Ptr(armcontainerservice.OSTypeLinux),

test/e2e/framework/params/params.go

@@ -0,0 +1,17 @@
+package params
+
+import (
+	"os"
+)
+
+var (
+	Location           = os.Getenv("LOCATION")
+	SubscriptionID     = os.Getenv("AZURE_SUBSCRIPTION_ID")
+	ResourceGroup      = os.Getenv("AZURE_RESOURCE_GROUP")
+	ClusterName        = os.Getenv("CLUSTER_NAME")
+	NodesPerPool       = os.Getenv("NODES_PER_POOL")
+	NumDeployments     = os.Getenv("NUM_DEPLOYMENTS")
+	NumReplicas        = os.Getenv("NUM_REPLICAS")
+	NumNetworkPolicies = os.Getenv("NUM_NET_POL")
+	CleanUp            = os.Getenv("CLEANUP")
+)

test/e2e/jobs/jobs.go

@@ -43,6 +43,7 @@ func CreateTestInfra(subID, rg, clusterName, location, kubeConfigFilePath string
 			PodCidr:      "10.128.0.0/9",
 			DNSServiceIP: "192.168.0.10",
 			ServiceCidr:  "192.168.0.0/28",
+			NodesPerPool: 1,
 		}, nil)
 
 		job.AddStep(&azure.GetAKSKubeConfig{

test/e2e/jobs/scale.go

@@ -5,12 +5,49 @@ import (
 	"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/kubernetes"
 	"github.com/microsoft/retina/test/e2e/framework/scaletest"
 	"github.com/microsoft/retina/test/e2e/framework/types"
 )
 
 func DefaultScaleTestOptions() scaletest.Options {
+	// var NumRealDeployments int
+	// var NumReplicas int
+	// var err error
+	//
+	// if env.NumDeployments != "" {
+	// 	NumRealDeployments, err = strconv.Atoi(env.NumDeployments)
+	// 	require.NoError(t, err, "Failed to convert NUM_DEPLOYMENTS to int")
+	// } else {
+	// 	NumRealDeployments = 1000
+	// }
+	//
+	// if env.NumReplicas != "" {
+	// 	NumReplicas, err = strconv.Atoi(env.NumReplicas)
+	// 	require.NoError(t, err, "Failed to convert NUM_REPLICAS to int")
+	// } else {
+	// 	NumReplicas = "40"
+	// NumNetworkPolicies := env.NumNetworkPolicies
+	// CleanUp := env.CleanUp
+	//
+	// if NumDeployments != "" {
+	// } else {
+	// 	NumRealDeployments = 1000
+	// }
+	// if NumReplicas != "" {
+	// 	opt.NumRealReplicas, err = strconv.Atoi(NumReplicas)
+	// 	require.NoError(t, err)
+	// }
+	// if NumNetworkPolicies != "" {
+	// 	opt.NumNetworkPolicies, err = strconv.Atoi(NumNetworkPolicies)
+	// 	require.NoError(t, err)
+	// }
+	// if CleanUp != "" {
+	// 	opt.DeleteLabels, err = strconv.ParseBool(CleanUp)
+	// 	require.NoError(t, err)
+	// }
 	return scaletest.Options{
 		Namespace:                     "scale-test",
 		MaxKwokPodsPerNode:            0,
@@ -45,6 +82,57 @@ func DefaultScaleTestOptions() scaletest.Options {
 	}
 }
 
+func GetScaleTestInfra(subID, rg, clusterName, location, kubeConfigFilePath string, nodesPerPool int32, createInfra bool) *types.Job {
+	job := types.NewJob("Get scale test infrastructure")
+
+	if createInfra {
+		job.AddStep(&azure.CreateResourceGroup{
+			SubscriptionID:    subID,
+			ResourceGroupName: rg,
+			Location:          location,
+		}, nil)
+
+		job.AddStep(&azure.CreateVNet{
+			VnetName:         "testvnet",
+			VnetAddressSpace: "10.0.0.0/9",
+		}, nil)
+
+		job.AddStep(&azure.CreateSubnet{
+			SubnetName:         "testsubnet",
+			SubnetAddressSpace: "10.0.0.0/12",
+		}, nil)
+
+		job.AddStep(&azure.CreateNPMCluster{
+			ClusterName:  clusterName,
+			PodCidr:      "10.128.0.0/9",
+			DNSServiceIP: "192.168.0.10",
+			ServiceCidr:  "192.168.0.0/28",
+			NodesPerPool: nodesPerPool,
+		}, nil)
+
+		job.AddStep(&azure.GetAKSKubeConfig{
+			KubeConfigFilePath: kubeConfigFilePath,
+		}, nil)
+
+	} else {
+		job.AddStep(&azure.GetAKSKubeConfig{
+			KubeConfigFilePath: kubeConfigFilePath,
+			ClusterName:        clusterName,
+			SubscriptionID:     subID,
+			ResourceGroupName:  rg,
+			Location:           location,
+		}, nil)
+	}
+
+	job.AddStep(&generic.LoadFlags{
+		TagEnv:            generic.DefaultTagEnv,
+		ImageNamespaceEnv: generic.DefaultImageNamespace,
+		ImageRegistryEnv:  generic.DefaultImageRegistry,
+	}, nil)
+
+	return job
+}
+
 func ScaleTest(opt *scaletest.Options) *types.Job {
 	job := types.NewJob("Scale Test")
 

test/e2e/scale_test.go

@@ -3,8 +3,6 @@
 package retina
 
 import (
-	"crypto/rand"
-	"math/big"
 	"os"
 	"path/filepath"
 	"strconv"
@@ -14,6 +12,7 @@ import (
 	"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/params"
 	"github.com/microsoft/retina/test/e2e/framework/types"
 	jobs "github.com/microsoft/retina/test/e2e/jobs"
 	"github.com/stretchr/testify/require"
@@ -23,25 +22,11 @@ func TestE2ERetina_Scale(t *testing.T) {
 	ctx, cancel := helpers.Context(t)
 	defer cancel()
 
-	clusterName := common.ClusterNameForE2ETest(t)
-
-	subID := os.Getenv("AZURE_SUBSCRIPTION_ID")
-	require.NotEmpty(t, subID)
-
-	location := os.Getenv("AZURE_LOCATION")
-	if location == "" {
-		nBig, err := rand.Int(rand.Reader, big.NewInt(int64(len(common.AzureLocations))))
-		if err != nil {
-			t.Fatal("Failed to generate a secure random index", err)
-		}
-		location = common.AzureLocations[nBig.Int64()]
-	}
-
-	rg := os.Getenv("AZURE_RESOURCE_GROUP")
-	if rg == "" {
-		// Use the cluster name as the resource group name by default.
-		rg = clusterName
-	}
+	clusterName := common.ScaleTestInfra.GetClusterName(t)
+	subID := common.ScaleTestInfra.GetSubscriptionID(t)
+	location := common.ScaleTestInfra.GetLocation(t)
+	rg := common.ScaleTestInfra.GetResourceGroup(t)
+	nodesPerPool := common.ScaleTestInfra.GetNodesPerPool(t)
 
 	cwd, err := os.Getwd()
 	require.NoError(t, err)
@@ -56,10 +41,10 @@ func TestE2ERetina_Scale(t *testing.T) {
 	opt := jobs.DefaultScaleTestOptions()
 	opt.KubeconfigPath = kubeConfigFilePath
 
-	NumDeployments := os.Getenv("NUM_DEPLOYMENTS")
-	NumReplicas := os.Getenv("NUM_REPLICAS")
-	NumNetworkPolicies := os.Getenv("NUM_NETPOLS")
-	CleanUp := os.Getenv("CLEANUP")
+	NumDeployments := params.NumDeployments
+	NumReplicas := params.NumReplicas
+	NumNetworkPolicies := params.NumNetworkPolicies
+	CleanUp := params.CleanUp
 
 	if NumDeployments != "" {
 		opt.NumRealDeployments, err = strconv.Atoi(NumDeployments)
@@ -89,9 +74,11 @@ func TestE2ERetina_Scale(t *testing.T) {
 
 	opt.LabelsToGetMetrics = map[string]string{"k8s-app": "retina"}
 
+	createInfra := *common.CreateInfra
+
 	// CreateTestInfra
-	createTestInfra := types.NewRunner(t, jobs.CreateTestInfra(subID, rg, clusterName, location, kubeConfigFilePath, *common.CreateInfra))
-	createTestInfra.Run(ctx)
+	infra := types.NewRunner(t, jobs.GetScaleTestInfra(subID, rg, clusterName, location, kubeConfigFilePath, nodesPerPool, createInfra))
+	infra.Run(ctx)
 
 	t.Cleanup(func() {
 		if *common.DeleteInfra {