Merge pull request #124578 from sanposhiho/scheduler_perf_scheduler_plugin_execution_duration_seconds

support `scheduler_plugin_execution_duration_seconds` in scheduler_perf

pkg/scheduler/metrics/metrics.go

@@ -38,7 +38,24 @@ const (
 	Binding = "binding"
 )
 
-// Below are possible values for the extension_point label.
+// ExtentionPoints is a list of possible values for the extension_point label.
+var ExtentionPoints = []string{
+	PreFilter,
+	Filter,
+	PreFilterExtensionAddPod,
+	PreFilterExtensionRemovePod,
+	PostFilter,
+	PreScore,
+	Score,
+	ScoreExtensionNormalize,
+	PreBind,
+	Bind,
+	PostBind,
+	Reserve,
+	Unreserve,
+	Permit,
+}
+
 const (
 	PreFilter                   = "PreFilter"
 	Filter                      = "Filter"

test/integration/scheduler_perf/scheduler_perf.go

@@ -53,6 +53,7 @@ import (
 	"k8s.io/kubernetes/pkg/scheduler/apis/config"
 	"k8s.io/kubernetes/pkg/scheduler/apis/config/scheme"
 	"k8s.io/kubernetes/pkg/scheduler/apis/config/validation"
+	"k8s.io/kubernetes/pkg/scheduler/framework/plugins/names"
 	frameworkruntime "k8s.io/kubernetes/pkg/scheduler/framework/runtime"
 	"k8s.io/kubernetes/pkg/scheduler/metrics"
 	"k8s.io/kubernetes/test/integration/framework"
@@ -90,23 +91,65 @@ const (
 	configFile               = "config/performance-config.yaml"
 	extensionPointsLabelName = "extension_point"
 	resultLabelName          = "result"
+	pluginLabelName          = "plugin"
 )
 
 var (
 	defaultMetricsCollectorConfig = metricsCollectorConfig{
-		Metrics: map[string]*labelValues{
+		Metrics: map[string][]*labelValues{
 			"scheduler_framework_extension_point_duration_seconds": {
-				label:  extensionPointsLabelName,
+				{
-				values: []string{"Filter", "Score"},
+					label:  extensionPointsLabelName,
+					values: metrics.ExtentionPoints,
+				},
 			},
 			"scheduler_scheduling_attempt_duration_seconds": {
-				label:  resultLabelName,
+				{
-				values: []string{metrics.ScheduledResult, metrics.UnschedulableResult, metrics.ErrorResult},
+					label:  resultLabelName,
+					values: []string{metrics.ScheduledResult, metrics.UnschedulableResult, metrics.ErrorResult},
+				},
+			},
+			"scheduler_pod_scheduling_duration_seconds": nil,
+			"scheduler_plugin_execution_duration_seconds": {
+				{
+					label:  pluginLabelName,
+					values: PluginNames,
+				},
+				{
+					label:  extensionPointsLabelName,
+					values: metrics.ExtentionPoints,
+				},
 			},
-			"scheduler_pod_scheduling_duration_seconds":     nil,
-			"scheduler_pod_scheduling_sli_duration_seconds": nil,
 		},
 	}
+
+	// PluginNames is the names of the plugins that scheduler_perf collects metrics for.
+	// We export this variable because people outside k/k may want to put their custom plugins.
+	PluginNames = []string{
+		names.PrioritySort,
+		names.DefaultBinder,
+		names.DefaultPreemption,
+		names.DynamicResources,
+		names.ImageLocality,
+		names.InterPodAffinity,
+		names.NodeAffinity,
+		names.NodeName,
+		names.NodePorts,
+		names.NodeResourcesBalancedAllocation,
+		names.NodeResourcesFit,
+		names.NodeUnschedulable,
+		names.NodeVolumeLimits,
+		names.AzureDiskLimits,
+		names.CinderLimits,
+		names.EBSLimits,
+		names.GCEPDLimits,
+		names.PodTopologySpread,
+		names.SchedulingGates,
+		names.TaintToleration,
+		names.VolumeBinding,
+		names.VolumeRestrictions,
+		names.VolumeZone,
+	}
 )
 
 // testCase defines a set of test cases that intends to test the performance of
@@ -668,7 +711,9 @@ func withCleanup(tCtx ktesting.TContext, enabled bool) ktesting.TContext {
 var perfSchedulingLabelFilter = flag.String("perf-scheduling-label-filter", "performance", "comma-separated list of labels which a testcase must have (no prefix or +) or must not have (-), used by BenchmarkPerfScheduling")
 
 // RunBenchmarkPerfScheduling runs the scheduler performance tests.
-// Optionally, you can pass your own scheduler plugin via outOfTreePluginRegistry.
+//
+// You can pass your own scheduler plugins via outOfTreePluginRegistry.
+// Also, you may want to put your plugins in PluginNames variable in this package.
 func RunBenchmarkPerfScheduling(b *testing.B, outOfTreePluginRegistry frameworkruntime.Registry) {
 	testCases, err := getTestCases(configFile)
 	if err != nil {

test/integration/scheduler_perf/util.go

@@ -247,7 +247,7 @@ type labelValues struct {
 // metricsCollectorConfig is the config to be marshalled to YAML config file.
 // NOTE: The mapping here means only one filter is supported, either value in the list of `values` is able to be collected.
 type metricsCollectorConfig struct {
-	Metrics map[string]*labelValues
+	Metrics map[string][]*labelValues
 }
 
 // metricsCollector collects metrics from legacyregistry.DefaultGatherer.Gather() endpoint.
@@ -270,17 +270,15 @@ func (*metricsCollector) run(tCtx ktesting.TContext) {
 
 func (pc *metricsCollector) collect() []DataItem {
 	var dataItems []DataItem
-	for metric, labelVals := range pc.Metrics {
+	for metric, labelValsSlice := range pc.Metrics {
 		// no filter is specified, aggregate all the metrics within the same metricFamily.
-		if labelVals == nil {
+		if labelValsSlice == nil {
 			dataItem := collectHistogramVec(metric, pc.labels, nil)
 			if dataItem != nil {
 				dataItems = append(dataItems, *dataItem)
 			}
 		} else {
-			// fetch the metric from metricFamily which match each of the lvMap.
+			for _, lvMap := range uniqueLVCombos(labelValsSlice) {
-			for _, value := range labelVals.values {
-				lvMap := map[string]string{labelVals.label: value}
 				dataItem := collectHistogramVec(metric, pc.labels, lvMap)
 				if dataItem != nil {
 					dataItems = append(dataItems, *dataItem)
@@ -291,6 +289,32 @@ func (pc *metricsCollector) collect() []DataItem {
 	return dataItems
 }
 
+// uniqueLVCombos lists up all possible label values combinations.
+// e.g., if there are 3 labelValues, each of which has 2 values,
+// the result would be {A: a1, B: b1, C: c1}, {A: a2, B: b1, C: c1}, {A: a1, B: b2, C: c1}, ... (2^3 = 8 combinations).
+func uniqueLVCombos(lvs []*labelValues) []map[string]string {
+	if len(lvs) == 0 {
+		return []map[string]string{{}}
+	}
+
+	remainingCombos := uniqueLVCombos(lvs[1:])
+
+	results := make([]map[string]string, 0)
+
+	current := lvs[0]
+	for _, value := range current.values {
+		for _, combo := range remainingCombos {
+			newCombo := make(map[string]string, len(combo)+1)
+			for k, v := range combo {
+				newCombo[k] = v
+			}
+			newCombo[current.label] = value
+			results = append(results, newCombo)
+		}
+	}
+	return results
+}
+
 func collectHistogramVec(metric string, labels map[string]string, lvMap map[string]string) *DataItem {
 	vec, err := testutil.GetHistogramVecFromGatherer(legacyregistry.DefaultGatherer, metric, lvMap)
 	if err != nil {

test/integration/scheduler_perf/util_test.go

@@ -0,0 +1,87 @@
+/*
+Copyright 2015 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package benchmark
+
+import (
+	"reflect"
+	"testing"
+)
+
+func Test_uniqueLVCombos(t *testing.T) {
+	type args struct {
+		lvs []*labelValues
+	}
+	tests := []struct {
+		name string
+		args args
+		want []map[string]string
+	}{
+		{
+			name: "empty input",
+			args: args{
+				lvs: []*labelValues{},
+			},
+			want: []map[string]string{{}},
+		},
+		{
+			name: "single label, multiple values",
+			args: args{
+				lvs: []*labelValues{
+					{"A", []string{"a1", "a2"}},
+				},
+			},
+			want: []map[string]string{
+				{"A": "a1"},
+				{"A": "a2"},
+			},
+		},
+		{
+			name: "multiple labels, single value each",
+			args: args{
+				lvs: []*labelValues{
+					{"A", []string{"a1"}},
+					{"B", []string{"b1"}},
+				},
+			},
+			want: []map[string]string{
+				{"A": "a1", "B": "b1"},
+			},
+		},
+		{
+			name: "multiple labels, multiple values",
+			args: args{
+				lvs: []*labelValues{
+					{"A", []string{"a1", "a2"}},
+					{"B", []string{"b1", "b2"}},
+				},
+			},
+			want: []map[string]string{
+				{"A": "a1", "B": "b1"},
+				{"A": "a1", "B": "b2"},
+				{"A": "a2", "B": "b1"},
+				{"A": "a2", "B": "b2"},
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := uniqueLVCombos(tt.args.lvs); !reflect.DeepEqual(got, tt.want) {
+				t.Errorf("uniqueLVCombos() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}