*: init node100_dp5_pod10k for runkperf bench

api/types/load_traffic.go

@@ -104,6 +104,8 @@ type RequestList struct {
 	Limit int `json:"limit" yaml:"limit"`
 	// Selector defines how to identify a set of objects.
 	Selector string `json:"seletor" yaml:"seletor"`
+	// FieldSelector defines how to identify a set of objects with field selector.
+	FieldSelector string `json:"fieldSelector" yaml:"fieldSelector"`
 }
 
 // RequestPut defines PUT request for target resource type.

api/types/load_traffic_test.go

@@ -39,6 +39,7 @@ spec:
       resource: pods
       namespace: default
       seletor: app=x2
+      fieldSelector: spec.nodeName=x
     shares: 200
   - quorumList:
       group: core
@@ -94,6 +95,7 @@ spec:
 	assert.Equal(t, "default", target.Spec.Requests[2].StaleList.Namespace)
 	assert.Equal(t, 0, target.Spec.Requests[2].StaleList.Limit)
 	assert.Equal(t, "app=x2", target.Spec.Requests[2].StaleList.Selector)
+	assert.Equal(t, "spec.nodeName=x", target.Spec.Requests[2].StaleList.FieldSelector)
 
 	assert.NotNil(t, target.Spec.Requests[3].QuorumList)
 	assert.Equal(t, 400, target.Spec.Requests[3].Shares)

contrib/cmd/runkperf/commands/bench/node100_dp5_pod10k.go

@@ -0,0 +1,108 @@
+package bench
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	internaltypes "github.com/Azure/kperf/contrib/internal/types"
+	"github.com/Azure/kperf/contrib/internal/utils"
+
+	"github.com/urfave/cli"
+)
+
+var benchNode100Deployment5Pod10KCase = cli.Command{
+	Name: "node100_dp5_pod10k",
+	Usage: `
+
+The test suite is to setup 100 virtual nodes and deploy 5 deployments for 10k
+pods on that nodes. It repeats to rolling-update deployments one by one during
+benchmark.
+	`,
+	Flags: []cli.Flag{
+		cli.IntFlag{
+			Name:  "total",
+			Usage: "Total requests per runner (There are 10 runners totally and runner's rate is 10)",
+			Value: 36000,
+		},
+		cli.IntFlag{
+			Name:  "podsize",
+			Usage: "Add <key=data, value=randomStringByLen(podsize)> in pod's annotation to increase pod size. The value is close to pod's size",
+			Value: 0,
+		},
+	},
+	Action: func(cliCtx *cli.Context) error {
+		_, err := renderBenchmarkReportInterceptor(
+			addAPIServerCoresInfoInterceptor(benchNode100Deployment5Pod10KRun),
+		)(cliCtx)
+		return err
+	},
+}
+
+// benchNode100Deployment5Pod10KCase is for subcommand benchNode100Deployment5Pod10KCase.
+func benchNode100Deployment5Pod10KRun(cliCtx *cli.Context) (*internaltypes.BenchmarkReport, error) {
+	ctx := context.Background()
+	kubeCfgPath := cliCtx.GlobalString("kubeconfig")
+
+	rgCfgFile, rgSpec, rgCfgFileDone, err := newLoadProfileFromEmbed(cliCtx,
+		"loadprofile/node100_dp5_pod10k.yaml")
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = rgCfgFileDone() }()
+
+	vcDone, err := deployVirtualNodepool(ctx, cliCtx, "node100dp5pod10k", 100, 150)
+	if err != nil {
+		return nil, fmt.Errorf("failed to deploy virtual node: %w", err)
+	}
+	defer func() { _ = vcDone() }()
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+
+	restartInterval := 10 * time.Second
+	dpCtx, dpCancel := context.WithCancel(ctx)
+
+	podSize := cliCtx.Int("podsize")
+	rollingUpdateFn, err := utils.RepeatRollingUpdate10KPod(dpCtx, kubeCfgPath, "dp5pod10k", podSize, restartInterval)
+	if err != nil {
+		dpCancel()
+		return nil, fmt.Errorf("failed to setup workload: %w", err)
+	}
+
+	go func() {
+		defer wg.Done()
+
+		// FIXME(weifu):
+		//
+		// DeployRunnerGroup should return ready notification.
+		// The rolling update should run after runners.
+		rollingUpdateFn()
+	}()
+
+	rgResult, derr := utils.DeployRunnerGroup(ctx,
+		cliCtx.GlobalString("kubeconfig"),
+		cliCtx.GlobalString("runner-image"),
+		rgCfgFile,
+		cliCtx.GlobalString("runner-flowcontrol"),
+		cliCtx.GlobalString("rg-affinity"),
+	)
+	dpCancel()
+	wg.Wait()
+
+	if derr != nil {
+		return nil, derr
+	}
+
+	return &internaltypes.BenchmarkReport{
+		Description: fmt.Sprintf(`
+Environment: 100 virtual nodes managed by kwok-controller,
+Workload: Deploy 5 deployments with 10,000 pods. Rolling-update deployments one by one and the interval is %v`, restartInterval),
+		LoadSpec: *rgSpec,
+		Result:   *rgResult,
+		Info: map[string]interface{}{
+			"podSizeInBytes": podSize,
+		},
+	}, nil
+}

contrib/cmd/runkperf/commands/bench/node100_job1_pod3k.go

@@ -6,14 +6,10 @@ import (
 	"sync"
 	"time"
 
-	"github.com/Azure/kperf/api/types"
-	kperfcmdutils "github.com/Azure/kperf/cmd/kperf/commands/utils"
 	internaltypes "github.com/Azure/kperf/contrib/internal/types"
 	"github.com/Azure/kperf/contrib/internal/utils"
 
 	"github.com/urfave/cli"
-	"gopkg.in/yaml.v2"
-	"k8s.io/klog/v2"
 )
 
 var benchNode100Job1Pod3KCase = cli.Command{
@@ -43,31 +39,8 @@ func benchNode100Job1Pod3KCaseRun(cliCtx *cli.Context) (*internaltypes.Benchmark
 	ctx := context.Background()
 	kubeCfgPath := cliCtx.GlobalString("kubeconfig")
 
-	var rgSpec types.RunnerGroupSpec
-	rgCfgFile, rgCfgFileDone, err := utils.NewLoadProfileFromEmbed(
-		"loadprofile/node100_job1_pod3k.yaml",
-		func(spec *types.RunnerGroupSpec) error {
-			reqs := cliCtx.Int("total")
-			if reqs < 0 {
-				return fmt.Errorf("invalid total-requests value: %v", reqs)
-			}
-
-			rgAffinity := cliCtx.GlobalString("rg-affinity")
-			affinityLabels, err := kperfcmdutils.KeyValuesMap([]string{rgAffinity})
-			if err != nil {
-				return fmt.Errorf("failed to parse %s affinity: %w", rgAffinity, err)
-			}
-
-			spec.Profile.Spec.Total = reqs
-			spec.NodeAffinity = affinityLabels
-
-			data, _ := yaml.Marshal(spec)
-			klog.V(2).InfoS("Load Profile", "config", string(data))
-
-			rgSpec = *spec
-			return nil
-		},
-	)
+	rgCfgFile, rgSpec, rgCfgFileDone, err := newLoadProfileFromEmbed(cliCtx,
+		"loadprofile/node100_job1_pod3k.yaml")
 	if err != nil {
 		return nil, err
 	}
@@ -108,7 +81,7 @@ func benchNode100Job1Pod3KCaseRun(cliCtx *cli.Context) (*internaltypes.Benchmark
 		Description: fmt.Sprintf(`
 Environment: 100 virtual nodes managed by kwok-controller,
 Workload: Deploy 1 job with 3,000 pods repeatedly. The parallelism is 100. The interval is %v`, jobInterval),
-		LoadSpec: rgSpec,
+		LoadSpec: *rgSpec,
 		Result:   *rgResult,
 		Info:     make(map[string]interface{}),
 	}, nil

contrib/cmd/runkperf/commands/bench/root.go

@@ -53,5 +53,6 @@ var Command = cli.Command{
 	},
 	Subcommands: []cli.Command{
 		benchNode100Job1Pod3KCase,
+		benchNode100Deployment5Pod10KCase,
 	},
 }

contrib/cmd/runkperf/commands/bench/utils.go

@@ -7,10 +7,13 @@ import (
 	"os"
 	"path/filepath"
 
+	"github.com/Azure/kperf/api/types"
+	kperfcmdutils "github.com/Azure/kperf/cmd/kperf/commands/utils"
 	internaltypes "github.com/Azure/kperf/contrib/internal/types"
 	"github.com/Azure/kperf/contrib/internal/utils"
 
 	"github.com/urfave/cli"
+	"gopkg.in/yaml.v2"
 	"k8s.io/klog/v2"
 )
 
@@ -120,3 +123,37 @@ func deployVirtualNodepool(ctx context.Context, cliCtx *cli.Context, target stri
 		return kr.DeleteNodepool(ctx, 0, target)
 	}, nil
 }
+
+// newLoadProfileFromEmbed loads load profile from embed and tweaks that load
+// profile.
+func newLoadProfileFromEmbed(cliCtx *cli.Context, name string) (_name string, _spec *types.RunnerGroupSpec, _cleanup func() error, _err error) {
+	var rgSpec types.RunnerGroupSpec
+	rgCfgFile, rgCfgFileDone, err := utils.NewLoadProfileFromEmbed(
+		name,
+		func(spec *types.RunnerGroupSpec) error {
+			reqs := cliCtx.Int("total")
+			if reqs < 0 {
+				return fmt.Errorf("invalid total-requests value: %v", reqs)
+			}
+
+			rgAffinity := cliCtx.GlobalString("rg-affinity")
+			affinityLabels, err := kperfcmdutils.KeyValuesMap([]string{rgAffinity})
+			if err != nil {
+				return fmt.Errorf("failed to parse %s affinity: %w", rgAffinity, err)
+			}
+
+			spec.Profile.Spec.Total = reqs
+			spec.NodeAffinity = affinityLabels
+
+			data, _ := yaml.Marshal(spec)
+			klog.V(2).InfoS("Load Profile", "config", string(data))
+
+			rgSpec = *spec
+			return nil
+		},
+	)
+	if err != nil {
+		return "", nil, nil, err
+	}
+	return rgCfgFile, &rgSpec, rgCfgFileDone, nil
+}

contrib/internal/manifests/helm.go

@@ -0,0 +1,12 @@
+package manifests
+
+import (
+	rootmainfests "github.com/Azure/kperf/manifests"
+
+	"helm.sh/helm/v3/pkg/chart"
+)
+
+// LoadChart returns chart from current package's embed filesystem.
+func LoadChart(componentName string) (*chart.Chart, error) {
+	return rootmainfests.LoadChartFromEmbedFS(FS, componentName)
+}

contrib/internal/manifests/loadprofile/node100_dp5_pod10k.yaml

@@ -0,0 +1,40 @@
+count: 10
+loadProfile:
+  version: 1
+  description: "node100-deployment5-pod10k"
+  spec:
+    rate: 10
+    total: 36000
+    conns: 10
+    client: 100
+    contentType: json
+    disableHTTP2: false
+    maxRetries: 0
+    requests:
+      - staleList:
+          version: v1
+          resource: pods
+          # NOTE: Please align with ../../utils/utils.go#RepeatRollingUpdate10KPod
+          seletor: "app=benchmark"
+          # NOTE: Please align with ../../../cmd/runkperf/commands/bench/node100_dp5_pod10k.go.
+          # And there are only 100 nodes and each node can run 150 pods. It should
+          # have items in the response.
+          fieldSelector: "spec.nodeName=node100dp5pod10k-49"
+        shares: 1000 # 1000 / (1000 + 100 + 200) * 10 = 7.7 req/s
+      - staleList:
+          version: v1
+          resource: pods
+        shares: 100 # 100 / (1000 + 100 + 200) * 10 = 0.7 req/s
+      - quorumList:
+          version: v1
+          resource: pods
+          namespace: benchmark-0
+          # NOTE: It's to simulate the request created by daemonset to get pods,
+          # including kubelet, when they want to get pods from ETCD. The limit
+          # is 100 because it's close to MaxPods value.
+          limit: 100
+          # NOTE: Please align with ../../../cmd/runkperf/commands/bench/node100_dp5_pod10k.go.
+          fieldSelector: "spec.nodeName=node100dp5pod10k-49"
+          # And there are only 100 nodes and each node can run 150 pods. It should
+          # have items in the response.
+        shares: 200 # 200 / (1000 + 100 + 200) * 10 = 1.5 req/s

contrib/internal/manifests/workload/2kpodper1deployment/Chart.yaml

@@ -0,0 +1,3 @@
+apiVersion: v1
+name: "2k-pods-per-1-deployment"
+version: "0.0.1"

contrib/internal/manifests/workload/2kpodper1deployment/templates/deployments.tpl

@@ -0,0 +1,53 @@
+{{- $pattern := .Values.pattern }}
+{{- $podSizeInBytes := int .Values.podSizeInBytes }}
+{{- range $index := (untilStep 0 (int .Values.total) 1) }}
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: {{ $pattern }}-{{ $index }}
+  labels:
+    name: benchmark-testing
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: {{ $pattern }}-{{ $index }}
+  namespace: {{ $pattern }}-{{ $index }}
+  labels:
+    app: {{ $pattern }}
+spec:
+  replicas: 2000
+  strategy:
+    rollingUpdate:
+      maxSurge: 100
+    type: RollingUpdate
+  selector:
+    matchLabels:
+      app: {{ $pattern }}
+      index: "{{ $index }}"
+  template:
+    metadata:
+      labels:
+        app: {{ $pattern }}
+        index: "{{ $index }}"
+      annotations:
+        data: "{{ randAlphaNum $podSizeInBytes | nospace }}"
+    spec:
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+            - matchExpressions:
+              - key: type
+                operator: In
+                values:
+                - kperf-virtualnodes
+      tolerations:
+      - key: "kperf.io/nodepool"
+        operator: "Exists"
+        effect: "NoSchedule"
+      containers:
+      - name: fake-container
+        image: fake-image
+---
+{{- end}}

contrib/internal/manifests/workload/2kpodper1deployment/values.yaml

@@ -0,0 +1,3 @@
+pattern: "benchmark"
+total: 5
+podSizeInBytes: 2048