Merge pull request kubernetes#88200 from liu-cong/benchmark

Make MetricCollector configurable for scheduler benchmark tests

Author: k8s-ci-robot

test/integration/scheduler_perf/scheduler_perf_test.go

@@ -19,13 +19,13 @@ package benchmark
 import (
 	"fmt"
 	"io/ioutil"
-	"sync/atomic"
 	"testing"
 	"time"
 
 	v1 "k8s.io/api/core/v1"
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
-	"k8s.io/client-go/tools/cache"
+	coreinformers "k8s.io/client-go/informers/core/v1"
+	clientset "k8s.io/client-go/kubernetes"
 	"k8s.io/component-base/featuregate"
 	featuregatetesting "k8s.io/component-base/featuregate/testing"
 	"k8s.io/klog"
@@ -39,11 +39,13 @@ const (
 )
 
 var (
-	defaultMetrics = []string{
-		"scheduler_scheduling_algorithm_predicate_evaluation_seconds",
-		"scheduler_scheduling_algorithm_priority_evaluation_seconds",
-		"scheduler_binding_duration_seconds",
-		"scheduler_e2e_scheduling_duration_seconds",
+	defaultMetricsCollectorConfig = metricsCollectorConfig{
+		Metrics: []string{
+			"scheduler_scheduling_algorithm_predicate_evaluation_seconds",
+			"scheduler_scheduling_algorithm_priority_evaluation_seconds",
+			"scheduler_binding_duration_seconds",
+			"scheduler_e2e_scheduling_duration_seconds",
+		},
 	}
 )
 
@@ -52,8 +54,8 @@ var (
 //
 // It specifies nodes and pods in the cluster before running the test. It also specifies the pods to
 // schedule during the test. The config can be as simple as just specify number of nodes/pods, where
-// default spec will be applied. It also allows the user to specify a pod spec template for more compicated
-// test cases.
+// default spec will be applied. It also allows the user to specify a pod spec template for more
+// complicated test cases.
 //
 // It also specifies the metrics to be collected after the test. If nothing is specified, default metrics
 // such as scheduling throughput and latencies will be collected.
@@ -68,6 +70,8 @@ type testCase struct {
 	PodsToSchedule podCase
 	// optional, feature gates to set before running the test
 	FeatureGates map[featuregate.Feature]bool
+	// optional, replaces default defaultMetricsCollectorConfig if supplied.
+	MetricsCollectorConfig *metricsCollectorConfig
 }
 
 type nodeCase struct {
@@ -100,6 +104,11 @@ type testParams struct {
 	NumPodsToSchedule int
 }
 
+type testDataCollector interface {
+	run(stopCh chan struct{})
+	collect() []DataItem
+}
+
 func BenchmarkPerfScheduling(b *testing.B) {
 	dataItems := DataItems{Version: "v1"}
 	tests := getSimpleTestCases(configFile)
@@ -119,119 +128,97 @@ func BenchmarkPerfScheduling(b *testing.B) {
 }
 
 func perfScheduling(test testCase, b *testing.B) []DataItem {
-	var nodeStrategy testutils.PrepareNodeStrategy = &testutils.TrivialNodePrepareStrategy{}
-	if test.Nodes.NodeAllocatableStrategy != nil {
-		nodeStrategy = test.Nodes.NodeAllocatableStrategy
-	} else if test.Nodes.LabelNodePrepareStrategy != nil {
-		nodeStrategy = test.Nodes.LabelNodePrepareStrategy
-	} else if test.Nodes.UniqueNodeLabelStrategy != nil {
-		nodeStrategy = test.Nodes.UniqueNodeLabelStrategy
-	}
-
-	setupPodStrategy := getPodStrategy(test.InitPods)
-	testPodStrategy := getPodStrategy(test.PodsToSchedule)
-
-	var nodeSpec *v1.Node
-	if test.Nodes.NodeTemplatePath != nil {
-		nodeSpec = getNodeSpecFromFile(test.Nodes.NodeTemplatePath)
-	}
-
 	finalFunc, podInformer, clientset := mustSetupScheduler()
 	defer finalFunc()
 
-	var nodePreparer testutils.TestNodePreparer
-	if nodeSpec != nil {
-		nodePreparer = framework.NewIntegrationTestNodePreparerWithNodeSpec(
-			clientset,
-			[]testutils.CountToStrategy{{Count: test.Nodes.Num, Strategy: nodeStrategy}},
-			nodeSpec,
-		)
-	} else {
-		nodePreparer = framework.NewIntegrationTestNodePreparer(
-			clientset,
-			[]testutils.CountToStrategy{{Count: test.Nodes.Num, Strategy: nodeStrategy}},
-			"scheduler-perf-",
-		)
-	}
-
+	nodePreparer := getNodePreparer(test.Nodes, clientset)
 	if err := nodePreparer.PrepareNodes(); err != nil {
 		klog.Fatalf("%v", err)
 	}
 	defer nodePreparer.CleanupNodes()
 
-	config := testutils.NewTestPodCreatorConfig()
-	config.AddStrategy(setupNamespace, test.InitPods.Num, setupPodStrategy)
-	podCreator := testutils.NewTestPodCreator(clientset, config)
-	podCreator.CreatePods()
-
-	for {
-		scheduled, err := getScheduledPods(podInformer)
-		if err != nil {
-			klog.Fatalf("%v", err)
-		}
-		if len(scheduled) >= test.InitPods.Num {
-			break
-		}
-		klog.Infof("got %d existing pods, required: %d", len(scheduled), test.InitPods.Num)
-		time.Sleep(1 * time.Second)
-	}
-
-	scheduled := int32(0)
-	completedCh := make(chan struct{})
-	podInformer.Informer().AddEventHandler(cache.ResourceEventHandlerFuncs{
-		UpdateFunc: func(old, cur interface{}) {
-			curPod := cur.(*v1.Pod)
-			oldPod := old.(*v1.Pod)
-
-			if len(oldPod.Spec.NodeName) == 0 && len(curPod.Spec.NodeName) > 0 {
-				if atomic.AddInt32(&scheduled, 1) >= int32(test.PodsToSchedule.Num) {
-					completedCh <- struct{}{}
-				}
-			}
-		},
-	})
+	createPods(setupNamespace, test.InitPods, clientset)
+	waitNumPodsScheduled(test.InitPods.Num, podInformer)
 
 	// start benchmark
 	b.ResetTimer()
 
-	// Start measuring throughput
+	// Start test data collectors.
 	stopCh := make(chan struct{})
-	throughputCollector := newThroughputCollector(podInformer)
-	go throughputCollector.run(stopCh)
+	collectors := getTestDataCollectors(test, podInformer, b)
+	for _, collector := range collectors {
+		go collector.run(stopCh)
+	}
 
-	// Scheduling the main workload
-	config = testutils.NewTestPodCreatorConfig()
-	config.AddStrategy(testNamespace, test.PodsToSchedule.Num, testPodStrategy)
-	podCreator = testutils.NewTestPodCreator(clientset, config)
-	podCreator.CreatePods()
+	// Schedule the main workload
+	createPods(testNamespace, test.PodsToSchedule, clientset)
+	waitNumPodsScheduled(test.InitPods.Num+test.PodsToSchedule.Num, podInformer)
 
-	<-completedCh
 	close(stopCh)
-
 	// Note: without this line we're taking the overhead of defer() into account.
 	b.StopTimer()
 
-	setNameLabel := func(dataItem *DataItem) DataItem {
-		if dataItem.Labels == nil {
-			dataItem.Labels = map[string]string{}
+	var dataItems []DataItem
+	for _, collector := range collectors {
+		dataItems = append(dataItems, collector.collect()...)
+	}
+	return dataItems
+}
+
+func waitNumPodsScheduled(num int, podInformer coreinformers.PodInformer) {
+	for {
+		scheduled, err := getScheduledPods(podInformer)
+		if err != nil {
+			klog.Fatalf("%v", err)
+		}
+		if len(scheduled) >= num {
+			break
 		}
-		dataItem.Labels["Name"] = b.Name()
-		return *dataItem
+		klog.Infof("got %d existing pods, required: %d", len(scheduled), num)
+		time.Sleep(1 * time.Second)
 	}
+}
 
-	dataItems := []DataItem{
-		setNameLabel(throughputCollector.collect()),
+func getTestDataCollectors(tc testCase, podInformer coreinformers.PodInformer, b *testing.B) []testDataCollector {
+	collectors := []testDataCollector{newThroughputCollector(podInformer, map[string]string{"Name": b.Name()})}
+	metricsCollectorConfig := defaultMetricsCollectorConfig
+	if tc.MetricsCollectorConfig != nil {
+		metricsCollectorConfig = *tc.MetricsCollectorConfig
 	}
+	collectors = append(collectors, newMetricsCollector(metricsCollectorConfig, map[string]string{"Name": b.Name()}))
+	return collectors
+}
 
-	for _, metric := range defaultMetrics {
-		dataItem := newMetricsCollector(metric).collect()
-		if dataItem == nil {
-			continue
-		}
-		dataItems = append(dataItems, setNameLabel(dataItem))
+func getNodePreparer(nc nodeCase, clientset clientset.Interface) testutils.TestNodePreparer {
+	var nodeStrategy testutils.PrepareNodeStrategy = &testutils.TrivialNodePrepareStrategy{}
+	if nc.NodeAllocatableStrategy != nil {
+		nodeStrategy = nc.NodeAllocatableStrategy
+	} else if nc.LabelNodePrepareStrategy != nil {
+		nodeStrategy = nc.LabelNodePrepareStrategy
+	} else if nc.UniqueNodeLabelStrategy != nil {
+		nodeStrategy = nc.UniqueNodeLabelStrategy
 	}
 
-	return dataItems
+	if nc.NodeTemplatePath != nil {
+		return framework.NewIntegrationTestNodePreparerWithNodeSpec(
+			clientset,
+			[]testutils.CountToStrategy{{Count: nc.Num, Strategy: nodeStrategy}},
+			getNodeSpecFromFile(nc.NodeTemplatePath),
+		)
+	}
+	return framework.NewIntegrationTestNodePreparer(
+		clientset,
+		[]testutils.CountToStrategy{{Count: nc.Num, Strategy: nodeStrategy}},
+		"scheduler-perf-",
+	)
+}
+
+func createPods(ns string, pc podCase, clientset clientset.Interface) {
+	strategy := getPodStrategy(pc)
+	config := testutils.NewTestPodCreatorConfig()
+	config.AddStrategy(ns, pc.Num, strategy)
+	podCreator := testutils.NewTestPodCreator(clientset, config)
+	podCreator.CreatePods()
 }
 
 func getPodStrategy(pc podCase) testutils.TestPodCreateStrategy {

test/integration/scheduler_perf/util.go

@@ -118,20 +118,42 @@ func dataItems2JSONFile(dataItems DataItems, namePrefix string) error {
 	return ioutil.WriteFile(destFile, b, 0644)
 }
 
+// metricsCollectorConfig is the config to be marshalled to YAML config file.
+type metricsCollectorConfig struct {
+	Metrics []string
+}
+
 // metricsCollector collects metrics from legacyregistry.DefaultGatherer.Gather() endpoint.
 // Currently only Histrogram metrics are supported.
 type metricsCollector struct {
-	metric string
+	metricsCollectorConfig
+	labels map[string]string
 }
 
-func newMetricsCollector(metric string) *metricsCollector {
+func newMetricsCollector(config metricsCollectorConfig, labels map[string]string) *metricsCollector {
 	return &metricsCollector{
-		metric: metric,
+		metricsCollectorConfig: config,
+		labels:                 labels,
 	}
 }
 
-func (pc *metricsCollector) collect() *DataItem {
-	hist, err := testutil.GetHistogramFromGatherer(legacyregistry.DefaultGatherer, pc.metric)
+func (*metricsCollector) run(stopCh chan struct{}) {
+	// metricCollector doesn't need to start before the tests, so nothing to do here.
+}
+
+func (pc *metricsCollector) collect() []DataItem {
+	var dataItems []DataItem
+	for _, metric := range pc.Metrics {
+		dataItem := collectHistogram(metric, pc.labels)
+		if dataItem != nil {
+			dataItems = append(dataItems, *dataItem)
+		}
+	}
+	return dataItems
+}
+
+func collectHistogram(metric string, labels map[string]string) *DataItem {
+	hist, err := testutil.GetHistogramFromGatherer(legacyregistry.DefaultGatherer, metric)
 	if err != nil {
 		klog.Error(err)
 		return nil
@@ -153,10 +175,13 @@ func (pc *metricsCollector) collect() *DataItem {
 
 	msFactor := float64(time.Second) / float64(time.Millisecond)
 
+	// Copy labels and add "Metric" label for this metric.
+	labelMap := map[string]string{"Metric": metric}
+	for k, v := range labels {
+		labelMap[k] = v
+	}
 	return &DataItem{
-		Labels: map[string]string{
-			"Metric": pc.metric,
-		},
+		Labels: labelMap,
 		Data: map[string]float64{
 			"Perc50":  q50 * msFactor,
 			"Perc90":  q90 * msFactor,
@@ -170,11 +195,13 @@ func (pc *metricsCollector) collect() *DataItem {
 type throughputCollector struct {
 	podInformer           coreinformers.PodInformer
 	schedulingThroughputs []float64
+	labels                map[string]string
 }
 
-func newThroughputCollector(podInformer coreinformers.PodInformer) *throughputCollector {
+func newThroughputCollector(podInformer coreinformers.PodInformer, labels map[string]string) *throughputCollector {
 	return &throughputCollector{
 		podInformer: podInformer,
+		labels:      labels,
 	}
 }
 
@@ -205,8 +232,8 @@ func (tc *throughputCollector) run(stopCh chan struct{}) {
 	}
 }
 
-func (tc *throughputCollector) collect() *DataItem {
-	throughputSummary := &DataItem{}
+func (tc *throughputCollector) collect() []DataItem {
+	throughputSummary := DataItem{Labels: tc.labels}
 	if length := len(tc.schedulingThroughputs); length > 0 {
 		sort.Float64s(tc.schedulingThroughputs)
 		sum := 0.0
@@ -225,5 +252,6 @@ func (tc *throughputCollector) collect() *DataItem {
 		}
 		throughputSummary.Unit = "pods/s"
 	}
-	return throughputSummary
+
+	return []DataItem{throughputSummary}
 }