Merge pull request kubernetes#124926 from kerthcet/feat/sharing-waitingPods

enhancement(scheduler): share waitingPods among profiles

Author: k8s-ci-robot

pkg/scheduler/framework/plugins/defaultpreemption/default_preemption_test.go

@@ -364,6 +364,7 @@ func TestPostFilter(t *testing.T) {
 				frameworkruntime.WithExtenders(extenders),
 				frameworkruntime.WithSnapshotSharedLister(internalcache.NewSnapshot(tt.pods, tt.nodes)),
 				frameworkruntime.WithLogger(logger),
+				frameworkruntime.WithWaitingPods(frameworkruntime.NewWaitingPodsMap()),
 			)
 			if err != nil {
 				t.Fatal(err)
@@ -1702,6 +1703,8 @@ func TestPreempt(t *testing.T) {
 			ctx, cancel := context.WithCancel(ctx)
 			defer cancel()
 
+			waitingPods := frameworkruntime.NewWaitingPodsMap()
+
 			cache := internalcache.New(ctx, time.Duration(0))
 			for _, pod := range test.pods {
 				cache.AddPod(logger, pod)
@@ -1745,6 +1748,7 @@ func TestPreempt(t *testing.T) {
 				frameworkruntime.WithPodNominator(internalqueue.NewPodNominator(informerFactory.Core().V1().Pods().Lister())),
 				frameworkruntime.WithSnapshotSharedLister(internalcache.NewSnapshot(test.pods, nodes)),
 				frameworkruntime.WithInformerFactory(informerFactory),
+				frameworkruntime.WithWaitingPods(waitingPods),
 				frameworkruntime.WithLogger(logger),
 			)
 			if err != nil {

pkg/scheduler/framework/runtime/framework.go

@@ -132,6 +132,7 @@ type frameworkOptions struct {
 	extenders              []framework.Extender
 	captureProfile         CaptureProfile
 	parallelizer           parallelize.Parallelizer
+	waitingPods            *waitingPodsMap
 	logger                 *klog.Logger
 }
 
@@ -221,6 +222,13 @@ func WithMetricsRecorder(r *metrics.MetricAsyncRecorder) Option {
 	}
 }
 
+// WithWaitingPods sets waitingPods for the scheduling frameworkImpl.
+func WithWaitingPods(wp *waitingPodsMap) Option {
+	return func(o *frameworkOptions) {
+		o.waitingPods = wp
+	}
+}
+
 // WithLogger overrides the default logger from k8s.io/klog.
 func WithLogger(logger klog.Logger) Option {
 	return func(o *frameworkOptions) {
@@ -254,7 +262,7 @@ func NewFramework(ctx context.Context, r Registry, profile *config.KubeScheduler
 		registry:             r,
 		snapshotSharedLister: options.snapshotSharedLister,
 		scorePluginWeight:    make(map[string]int),
-		waitingPods:          newWaitingPodsMap(),
+		waitingPods:          options.waitingPods,
 		clientSet:            options.clientSet,
 		kubeConfig:           options.kubeConfig,
 		eventRecorder:        options.eventRecorder,

pkg/scheduler/framework/runtime/framework_test.go

@@ -2808,7 +2808,9 @@ func TestPermitPlugins(t *testing.T) {
 			profile := config.KubeSchedulerProfile{Plugins: configPlugins}
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
-			f, err := newFrameworkWithQueueSortAndBind(ctx, registry, profile)
+			f, err := newFrameworkWithQueueSortAndBind(ctx, registry, profile,
+				WithWaitingPods(NewWaitingPodsMap()),
+			)
 			if err != nil {
 				t.Fatalf("fail to create framework: %s", err)
 			}
@@ -2990,7 +2992,10 @@ func TestRecordingMetrics(t *testing.T) {
 				SchedulerName:            testProfileName,
 				Plugins:                  plugins,
 			}
-			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile, withMetricsRecorder(recorder))
+			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile,
+				withMetricsRecorder(recorder),
+				WithWaitingPods(NewWaitingPodsMap()),
+			)
 			if err != nil {
 				cancel()
 				t.Fatalf("Failed to create framework for testing: %v", err)
@@ -3160,7 +3165,9 @@ func TestPermitWaitDurationMetric(t *testing.T) {
 			profile := config.KubeSchedulerProfile{Plugins: plugins}
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
-			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile)
+			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile,
+				WithWaitingPods(NewWaitingPodsMap()),
+			)
 			if err != nil {
 				t.Fatalf("Failed to create framework for testing: %v", err)
 			}
@@ -3216,7 +3223,9 @@ func TestWaitOnPermit(t *testing.T) {
 			profile := config.KubeSchedulerProfile{Plugins: plugins}
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
-			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile)
+			f, err := newFrameworkWithQueueSortAndBind(ctx, r, profile,
+				WithWaitingPods(NewWaitingPodsMap()),
+			)
 			if err != nil {
 				t.Fatalf("Failed to create framework for testing: %v", err)
 			}

pkg/scheduler/framework/runtime/waiting_pods_map.go

@@ -32,8 +32,8 @@ type waitingPodsMap struct {
 	mu   sync.RWMutex
 }
 
-// newWaitingPodsMap returns a new waitingPodsMap.
-func newWaitingPodsMap() *waitingPodsMap {
+// NewWaitingPodsMap returns a new waitingPodsMap.
+func NewWaitingPodsMap() *waitingPodsMap {
 	return &waitingPodsMap{
 		pods: make(map[types.UID]*waitingPod),
 	}

pkg/scheduler/schedule_one_test.go

@@ -778,7 +778,9 @@ func TestSchedulerScheduleOne(t *testing.T) {
 				registerPluginFuncs,
 				testSchedulerName,
 				frameworkruntime.WithClientSet(client),
-				frameworkruntime.WithEventRecorder(eventBroadcaster.NewRecorder(scheme.Scheme, testSchedulerName)))
+				frameworkruntime.WithEventRecorder(eventBroadcaster.NewRecorder(scheme.Scheme, testSchedulerName)),
+				frameworkruntime.WithWaitingPods(frameworkruntime.NewWaitingPodsMap()),
+			)
 			if err != nil {
 				t.Fatal(err)
 			}
@@ -3539,6 +3541,7 @@ func setupTestScheduler(ctx context.Context, t *testing.T, queuedPodStore *clien
 		informerFactory = informers.NewSharedInformerFactory(clientsetfake.NewSimpleClientset(), 0)
 	}
 	schedulingQueue := internalqueue.NewTestQueueWithInformerFactory(ctx, nil, informerFactory)
+	waitingPods := frameworkruntime.NewWaitingPodsMap()
 
 	fwk, _ := tf.NewFramework(
 		ctx,
@@ -3548,6 +3551,7 @@ func setupTestScheduler(ctx context.Context, t *testing.T, queuedPodStore *clien
 		frameworkruntime.WithEventRecorder(recorder),
 		frameworkruntime.WithInformerFactory(informerFactory),
 		frameworkruntime.WithPodNominator(internalqueue.NewPodNominator(informerFactory.Core().V1().Pods().Lister())),
+		frameworkruntime.WithWaitingPods(waitingPods),
 	)
 
 	errChan := make(chan error, 1)

pkg/scheduler/scheduler.go

@@ -292,6 +292,8 @@ func New(ctx context.Context,
 
 	snapshot := internalcache.NewEmptySnapshot()
 	metricsRecorder := metrics.NewMetricsAsyncRecorder(1000, time.Second, stopEverything)
+	// waitingPods holds all the pods that are in the scheduler and waiting in the permit stage
+	waitingPods := frameworkruntime.NewWaitingPodsMap()
 
 	profiles, err := profile.NewMap(ctx, options.profiles, registry, recorderFactory,
 		frameworkruntime.WithComponentConfigVersion(options.componentConfigVersion),
@@ -303,6 +305,7 @@ func New(ctx context.Context,
 		frameworkruntime.WithParallelism(int(options.parallelism)),
 		frameworkruntime.WithExtenders(extenders),
 		frameworkruntime.WithMetricsRecorder(metricsRecorder),
+		frameworkruntime.WithWaitingPods(waitingPods),
 	)
 	if err != nil {
 		return nil, fmt.Errorf("initializing profiles: %v", err)

pkg/scheduler/scheduler_test.go

@@ -21,13 +21,16 @@ import (
 	"fmt"
 	"sort"
 	"strings"
+	"sync"
 	"testing"
 	"time"
 
 	"github.com/google/go-cmp/cmp"
 	v1 "k8s.io/api/core/v1"
+	eventsv1 "k8s.io/api/events/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/util/sets"
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	"k8s.io/client-go/informers"
 	"k8s.io/client-go/kubernetes"
@@ -437,12 +440,14 @@ func initScheduler(ctx context.Context, cache internalcache.Cache, queue interna
 		tf.RegisterBindPlugin(defaultbinder.Name, defaultbinder.New),
 	}
 	eventBroadcaster := events.NewBroadcaster(&events.EventSinkImpl{Interface: client.EventsV1()})
+	waitingPods := frameworkruntime.NewWaitingPodsMap()
 	fwk, err := tf.NewFramework(ctx,
 		registerPluginFuncs,
 		testSchedulerName,
 		frameworkruntime.WithClientSet(client),
 		frameworkruntime.WithInformerFactory(informerFactory),
 		frameworkruntime.WithEventRecorder(eventBroadcaster.NewRecorder(scheme.Scheme, testSchedulerName)),
+		frameworkruntime.WithWaitingPods(waitingPods),
 	)
 	if err != nil {
 		return nil, nil, err
@@ -591,6 +596,7 @@ const (
 	queueSort                      = "no-op-queue-sort-plugin"
 	fakeBind                       = "bind-plugin"
 	emptyEventExtensions           = "emptyEventExtensions"
+	fakePermit                     = "fakePermit"
 )
 
 func Test_buildQueueingHintMap(t *testing.T) {
@@ -905,6 +911,160 @@ func newFramework(ctx context.Context, r frameworkruntime.Registry, profile sche
 	)
 }
 
+func TestFrameworkHandler_IterateOverWaitingPods(t *testing.T) {
+	const (
+		testSchedulerProfile1 = "test-scheduler-profile-1"
+		testSchedulerProfile2 = "test-scheduler-profile-2"
+		testSchedulerProfile3 = "test-scheduler-profile-3"
+	)
+
+	nodes := []runtime.Object{
+		st.MakeNode().Name("node1").UID("node1").Obj(),
+		st.MakeNode().Name("node2").UID("node2").Obj(),
+		st.MakeNode().Name("node3").UID("node3").Obj(),
+	}
+
+	cases := []struct {
+		name                        string
+		profiles                    []schedulerapi.KubeSchedulerProfile
+		waitSchedulingPods          []*v1.Pod
+		expectPodNamesInWaitingPods []string
+	}{
+		{
+			name: "pods with same profile are waiting on permit stage",
+			profiles: []schedulerapi.KubeSchedulerProfile{
+				{
+					SchedulerName: testSchedulerProfile1,
+					Plugins: &schedulerapi.Plugins{
+						QueueSort: schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "PrioritySort"}}},
+						Permit:    schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: fakePermit}}},
+						Bind:      schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "DefaultBinder"}}},
+					},
+				},
+			},
+			waitSchedulingPods: []*v1.Pod{
+				st.MakePod().Name("pod1").UID("pod1").SchedulerName(testSchedulerProfile1).Obj(),
+				st.MakePod().Name("pod2").UID("pod2").SchedulerName(testSchedulerProfile1).Obj(),
+				st.MakePod().Name("pod3").UID("pod3").SchedulerName(testSchedulerProfile1).Obj(),
+			},
+			expectPodNamesInWaitingPods: []string{"pod1", "pod2", "pod3"},
+		},
+		{
+			name: "pods with different profiles are waiting on permit stage",
+			profiles: []schedulerapi.KubeSchedulerProfile{
+				{
+					SchedulerName: testSchedulerProfile1,
+					Plugins: &schedulerapi.Plugins{
+						QueueSort: schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "PrioritySort"}}},
+						Permit:    schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: fakePermit}}},
+						Bind:      schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "DefaultBinder"}}},
+					},
+				},
+				{
+					SchedulerName: testSchedulerProfile2,
+					Plugins: &schedulerapi.Plugins{
+						QueueSort: schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "PrioritySort"}}},
+						Permit:    schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: fakePermit}}},
+						Bind:      schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "DefaultBinder"}}},
+					},
+				},
+				{
+					SchedulerName: testSchedulerProfile3,
+					Plugins: &schedulerapi.Plugins{
+						QueueSort: schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "PrioritySort"}}},
+						Permit:    schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: fakePermit}}},
+						Bind:      schedulerapi.PluginSet{Enabled: []schedulerapi.Plugin{{Name: "DefaultBinder"}}},
+					},
+				},
+			},
+			waitSchedulingPods: []*v1.Pod{
+				st.MakePod().Name("pod1").UID("pod1").SchedulerName(testSchedulerProfile1).Obj(),
+				st.MakePod().Name("pod2").UID("pod2").SchedulerName(testSchedulerProfile1).Obj(),
+				st.MakePod().Name("pod3").UID("pod3").SchedulerName(testSchedulerProfile2).Obj(),
+				st.MakePod().Name("pod4").UID("pod4").SchedulerName(testSchedulerProfile3).Obj(),
+			},
+			expectPodNamesInWaitingPods: []string{"pod1", "pod2", "pod3", "pod4"},
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			// Set up scheduler for the 3 nodes.
+			objs := append([]runtime.Object{&v1.Namespace{ObjectMeta: metav1.ObjectMeta{Name: ""}}}, nodes...)
+			fakeClient := fake.NewSimpleClientset(objs...)
+			informerFactory := informers.NewSharedInformerFactory(fakeClient, 0)
+			eventBroadcaster := events.NewBroadcaster(&events.EventSinkImpl{Interface: fakeClient.EventsV1()})
+			eventRecorder := eventBroadcaster.NewRecorder(scheme.Scheme, fakePermit)
+
+			outOfTreeRegistry := frameworkruntime.Registry{
+				fakePermit: newFakePermitPlugin(eventRecorder),
+			}
+
+			_, ctx := ktesting.NewTestContext(t)
+			ctx, cancel := context.WithCancel(ctx)
+			defer cancel()
+
+			scheduler, err := New(
+				ctx,
+				fakeClient,
+				informerFactory,
+				nil,
+				profile.NewRecorderFactory(eventBroadcaster),
+				WithProfiles(tc.profiles...),
+				WithFrameworkOutOfTreeRegistry(outOfTreeRegistry),
+			)
+
+			if err != nil {
+				t.Fatalf("Failed to create scheduler: %v", err)
+			}
+
+			var wg sync.WaitGroup
+			waitSchedulingPodNumber := len(tc.waitSchedulingPods)
+			wg.Add(waitSchedulingPodNumber)
+			stopFn, err := eventBroadcaster.StartEventWatcher(func(obj runtime.Object) {
+				e, ok := obj.(*eventsv1.Event)
+				if !ok || (e.Reason != podWaitingReason) {
+					return
+				}
+				wg.Done()
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			defer stopFn()
+
+			// Run scheduler.
+			informerFactory.Start(ctx.Done())
+			informerFactory.WaitForCacheSync(ctx.Done())
+			go scheduler.Run(ctx)
+
+			// Send pods to be scheduled.
+			for _, p := range tc.waitSchedulingPods {
+				_, err = fakeClient.CoreV1().Pods("").Create(ctx, p, metav1.CreateOptions{})
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			// Wait all pods in waitSchedulingPods to be scheduled.
+			wg.Wait()
+
+			// Ensure that all waitingPods in scheduler can be obtained from any profiles.
+			for _, fwk := range scheduler.Profiles {
+				actualPodNamesInWaitingPods := sets.NewString()
+				fwk.IterateOverWaitingPods(func(pod framework.WaitingPod) {
+					actualPodNamesInWaitingPods.Insert(pod.GetPod().Name)
+				})
+				// Validate the name of pods in waitingPods matches expectations.
+				if actualPodNamesInWaitingPods.Len() != len(tc.expectPodNamesInWaitingPods) ||
+					!actualPodNamesInWaitingPods.HasAll(tc.expectPodNamesInWaitingPods...) {
+					t.Fatalf("Unexpected waitingPods in scheduler profile %s, expect: %#v, got: %#v", fwk.ProfileName(), tc.expectPodNamesInWaitingPods, actualPodNamesInWaitingPods.List())
+				}
+			}
+		})
+	}
+}
+
 var _ framework.QueueSortPlugin = &fakeQueueSortPlugin{}
 
 // fakeQueueSortPlugin is a no-op implementation for QueueSort extension point.
@@ -1004,3 +1164,36 @@ func (*emptyEventsToRegisterPlugin) Filter(_ context.Context, _ *framework.Cycle
 }
 
 func (*emptyEventsToRegisterPlugin) EventsToRegister() []framework.ClusterEventWithHint { return nil }
+
+// fakePermitPlugin only implements PermitPlugin interface.
+type fakePermitPlugin struct {
+	eventRecorder events.EventRecorder
+}
+
+func newFakePermitPlugin(eventRecorder events.EventRecorder) frameworkruntime.PluginFactory {
+	return func(ctx context.Context, configuration runtime.Object, f framework.Handle) (framework.Plugin, error) {
+		pl := &fakePermitPlugin{
+			eventRecorder: eventRecorder,
+		}
+		return pl, nil
+	}
+}
+
+func (f fakePermitPlugin) Name() string {
+	return fakePermit
+}
+
+const (
+	podWaitingReason = "podWaiting"
+)
+
+func (f fakePermitPlugin) Permit(ctx context.Context, state *framework.CycleState, p *v1.Pod, nodeName string) (*framework.Status, time.Duration) {
+	defer func() {
+		// Send event with podWaiting reason to broadcast this pod is already waiting in the permit stage.
+		f.eventRecorder.Eventf(p, nil, v1.EventTypeWarning, podWaitingReason, "", "")
+	}()
+
+	return framework.NewStatus(framework.Wait), 100 * time.Second
+}
+
+var _ framework.PermitPlugin = &fakePermitPlugin{}