Merge pull request #73454 from denkensk/automated-cherry-pick-of-#72558-upstream-release-1.13

pkg/scheduler/core/extender_test.go

@@ -507,7 +507,7 @@ func TestGenericSchedulerWithExtenders(t *testing.T) {
 			for _, name := range test.nodes {
 				cache.AddNode(createNode(name))
 			}
-			queue := internalqueue.NewSchedulingQueue()
+			queue := internalqueue.NewSchedulingQueue(nil)
 			scheduler := NewGenericScheduler(
 				cache,
 				nil,

pkg/scheduler/core/generic_scheduler_test.go

@@ -449,7 +449,7 @@ func TestGenericScheduler(t *testing.T) {
 			scheduler := NewGenericScheduler(
 				cache,
 				nil,
-				internalqueue.NewSchedulingQueue(),
+				internalqueue.NewSchedulingQueue(nil),
 				test.predicates,
 				algorithm.EmptyPredicateMetadataProducer,
 				test.prioritizers,
@@ -485,7 +485,7 @@ func makeScheduler(predicates map[string]algorithm.FitPredicate, nodes []*v1.Nod
 	s := NewGenericScheduler(
 		cache,
 		nil,
-		internalqueue.NewSchedulingQueue(),
+		internalqueue.NewSchedulingQueue(nil),
 		predicates,
 		algorithm.EmptyPredicateMetadataProducer,
 		prioritizers,
@@ -1412,7 +1412,7 @@ func TestPreempt(t *testing.T) {
 			scheduler := NewGenericScheduler(
 				cache,
 				nil,
-				internalqueue.NewSchedulingQueue(),
+				internalqueue.NewSchedulingQueue(nil),
 				map[string]algorithm.FitPredicate{"matches": algorithmpredicates.PodFitsResources},
 				algorithm.EmptyPredicateMetadataProducer,
 				[]algorithm.PriorityConfig{{Function: numericPriority, Weight: 1}},
@@ -1543,7 +1543,7 @@ func TestCacheInvalidationRace(t *testing.T) {
 	scheduler := NewGenericScheduler(
 		mockCache,
 		eCache,
-		internalqueue.NewSchedulingQueue(),
+		internalqueue.NewSchedulingQueue(nil),
 		ps,
 		algorithm.EmptyPredicateMetadataProducer,
 		prioritizers,
@@ -1626,7 +1626,7 @@ func TestCacheInvalidationRace2(t *testing.T) {
 	scheduler := NewGenericScheduler(
 		cache,
 		eCache,
-		internalqueue.NewSchedulingQueue(),
+		internalqueue.NewSchedulingQueue(nil),
 		ps,
 		algorithm.EmptyPredicateMetadataProducer,
 		prioritizers,

pkg/scheduler/factory/factory.go

@@ -277,7 +277,7 @@ func NewConfigFactory(args *ConfigFactoryArgs) Configurator {
 	c := &configFactory{
 		client:                         args.Client,
 		podLister:                      schedulerCache,
-		podQueue:                       internalqueue.NewSchedulingQueue(),
+		podQueue:                       internalqueue.NewSchedulingQueue(stopEverything),
 		nodeLister:                     args.NodeInformer.Lister(),
 		pVLister:                       args.PvInformer.Lister(),
 		pVCLister:                      args.PvcInformer.Lister(),

pkg/scheduler/internal/queue/BUILD

@@ -13,6 +13,7 @@ go_library(
         "//staging/src/k8s.io/api/core/v1:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/types:go_default_library",
+        "//staging/src/k8s.io/apimachinery/pkg/util/wait:go_default_library",
         "//staging/src/k8s.io/client-go/tools/cache:go_default_library",
         "//vendor/k8s.io/klog:go_default_library",
     ],

pkg/scheduler/internal/queue/scheduling_queue.go

@@ -31,12 +31,14 @@ import (
 	"fmt"
 	"reflect"
 	"sync"
+	"time"
 
 	"k8s.io/klog"
 
 	"k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	ktypes "k8s.io/apimachinery/pkg/types"
+	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/client-go/tools/cache"
 	podutil "k8s.io/kubernetes/pkg/api/v1/pod"
 	"k8s.io/kubernetes/pkg/scheduler/algorithm/predicates"
@@ -48,6 +50,10 @@ var (
 	queueClosed = "scheduling queue is closed"
 )
 
+// If the pod stays in unschedulableQ longer than the unschedulableQTimeInterval,
+// the pod will be moved from unschedulableQ to activeQ.
+const unschedulableQTimeInterval = 60 * time.Second
+
 // SchedulingQueue is an interface for a queue to store pods waiting to be scheduled.
 // The interface follows a pattern similar to cache.FIFO and cache.Heap and
 // makes it easy to use those data structures as a SchedulingQueue.
@@ -79,9 +85,9 @@ type SchedulingQueue interface {
 
 // NewSchedulingQueue initializes a new scheduling queue. If pod priority is
 // enabled a priority queue is returned. If it is disabled, a FIFO is returned.
-func NewSchedulingQueue() SchedulingQueue {
+func NewSchedulingQueue(stop <-chan struct{}) SchedulingQueue {
 	if util.PodPriorityEnabled() {
-		return NewPriorityQueue()
+		return NewPriorityQueue(stop)
 	}
 	return NewFIFO()
 }
@@ -193,8 +199,10 @@ func NominatedNodeName(pod *v1.Pod) string {
 // pods that are already tried and are determined to be unschedulable. The latter
 // is called unschedulableQ.
 type PriorityQueue struct {
-	lock sync.RWMutex
-	cond sync.Cond
+	stop  <-chan struct{}
+	clock util.Clock
+	lock  sync.RWMutex
+	cond  sync.Cond
 
 	// activeQ is heap structure that scheduler actively looks at to find pods to
 	// schedule. Head of heap is the highest priority pod.
@@ -244,16 +252,25 @@ func activeQComp(pod1, pod2 interface{}) bool {
 }
 
 // NewPriorityQueue creates a PriorityQueue object.
-func NewPriorityQueue() *PriorityQueue {
+func NewPriorityQueue(stop <-chan struct{}) *PriorityQueue {
 	pq := &PriorityQueue{
+		clock:          util.RealClock{},
+		stop:           stop,
 		activeQ:        newHeap(cache.MetaNamespaceKeyFunc, activeQComp),
 		unschedulableQ: newUnschedulablePodsMap(),
 		nominatedPods:  newNominatedPodMap(),
 	}
 	pq.cond.L = &pq.lock
+
+	pq.run()
 	return pq
 }
 
+// run starts the goroutine to pump from unschedulableQ to activeQ
+func (p *PriorityQueue) run() {
+	go wait.Until(p.flushUnschedulableQLeftover, 30*time.Second, p.stop)
+}
+
 // Add adds a pod to the active queue. It should be called only when a new pod
 // is added so there is no chance the pod is already in either queue.
 func (p *PriorityQueue) Add(pod *v1.Pod) error {
@@ -324,6 +341,26 @@ func (p *PriorityQueue) AddUnschedulableIfNotPresent(pod *v1.Pod) error {
 	return err
 }
 
+// flushUnschedulableQLeftover moves pod which stays in unschedulableQ longer than the durationStayUnschedulableQ
+// to activeQ.
+func (p *PriorityQueue) flushUnschedulableQLeftover() {
+	p.lock.Lock()
+	defer p.lock.Unlock()
+
+	var podsToMove []*v1.Pod
+	currentTime := p.clock.Now()
+	for _, pod := range p.unschedulableQ.pods {
+		lastScheduleTime := podTimestamp(pod)
+		if !lastScheduleTime.IsZero() && currentTime.Sub(lastScheduleTime.Time) > unschedulableQTimeInterval {
+			podsToMove = append(podsToMove, pod)
+		}
+	}
+
+	if len(podsToMove) > 0 {
+		p.movePodsToActiveQueue(podsToMove)
+	}
+}
+
 // Pop removes the head of the active queue and returns it. It blocks if the
 // activeQ is empty and waits until a new item is added to the queue. It also
 // clears receivedMoveRequest to mark the beginning of a new scheduling cycle.

pkg/scheduler/internal/queue/scheduling_queue_test.go

@@ -21,6 +21,7 @@ import (
 	"reflect"
 	"sync"
 	"testing"
+	"time"
 
 	"k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -94,8 +95,20 @@ var highPriorityPod, highPriNominatedPod, medPriorityPod, unschedulablePod = v1.
 		},
 	}
 
+func addOrUpdateUnschedulablePod(p *PriorityQueue, pod *v1.Pod) {
+	p.lock.Lock()
+	defer p.lock.Unlock()
+	p.unschedulableQ.addOrUpdate(pod)
+}
+
+func getUnschedulablePod(p *PriorityQueue, pod *v1.Pod) *v1.Pod {
+	p.lock.Lock()
+	defer p.lock.Unlock()
+	return p.unschedulableQ.get(pod)
+}
+
 func TestPriorityQueue_Add(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Add(&medPriorityPod)
 	q.Add(&unschedulablePod)
 	q.Add(&highPriorityPod)
@@ -126,8 +139,8 @@ func TestPriorityQueue_Add(t *testing.T) {
 }
 
 func TestPriorityQueue_AddIfNotPresent(t *testing.T) {
-	q := NewPriorityQueue()
-	q.unschedulableQ.addOrUpdate(&highPriNominatedPod)
+	q := NewPriorityQueue(nil)
+	addOrUpdateUnschedulablePod(q, &highPriNominatedPod)
 	q.AddIfNotPresent(&highPriNominatedPod) // Must not add anything.
 	q.AddIfNotPresent(&medPriorityPod)
 	q.AddIfNotPresent(&unschedulablePod)
@@ -152,13 +165,13 @@ func TestPriorityQueue_AddIfNotPresent(t *testing.T) {
 	if len(q.nominatedPods.nominatedPods["node1"]) != 2 {
 		t.Errorf("Expected medPriorityPod and unschedulablePod to be still present in nomindatePods: %v", q.nominatedPods.nominatedPods["node1"])
 	}
-	if q.unschedulableQ.get(&highPriNominatedPod) != &highPriNominatedPod {
+	if getUnschedulablePod(q, &highPriNominatedPod) != &highPriNominatedPod {
 		t.Errorf("Pod %v was not found in the unschedulableQ.", highPriNominatedPod.Name)
 	}
 }
 
 func TestPriorityQueue_AddUnschedulableIfNotPresent(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Add(&highPriNominatedPod)
 	q.AddUnschedulableIfNotPresent(&highPriNominatedPod) // Must not add anything.
 	q.AddUnschedulableIfNotPresent(&medPriorityPod)      // This should go to activeQ.
@@ -185,13 +198,13 @@ func TestPriorityQueue_AddUnschedulableIfNotPresent(t *testing.T) {
 	if len(q.nominatedPods.nominatedPods) != 1 {
 		t.Errorf("Expected nomindatePods to have one element: %v", q.nominatedPods)
 	}
-	if q.unschedulableQ.get(&unschedulablePod) != &unschedulablePod {
+	if getUnschedulablePod(q, &unschedulablePod) != &unschedulablePod {
 		t.Errorf("Pod %v was not found in the unschedulableQ.", unschedulablePod.Name)
 	}
 }
 
 func TestPriorityQueue_Pop(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	wg := sync.WaitGroup{}
 	wg.Add(1)
 	go func() {
@@ -208,7 +221,7 @@ func TestPriorityQueue_Pop(t *testing.T) {
 }
 
 func TestPriorityQueue_Update(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Update(nil, &highPriorityPod)
 	if _, exists, _ := q.activeQ.Get(&highPriorityPod); !exists {
 		t.Errorf("Expected %v to be added to activeQ.", highPriorityPod.Name)
@@ -244,7 +257,7 @@ func TestPriorityQueue_Update(t *testing.T) {
 }
 
 func TestPriorityQueue_Delete(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Update(&highPriorityPod, &highPriNominatedPod)
 	q.Add(&unschedulablePod)
 	q.Delete(&highPriNominatedPod)
@@ -266,10 +279,10 @@ func TestPriorityQueue_Delete(t *testing.T) {
 }
 
 func TestPriorityQueue_MoveAllToActiveQueue(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Add(&medPriorityPod)
-	q.unschedulableQ.addOrUpdate(&unschedulablePod)
-	q.unschedulableQ.addOrUpdate(&highPriorityPod)
+	addOrUpdateUnschedulablePod(q, &unschedulablePod)
+	addOrUpdateUnschedulablePod(q, &highPriorityPod)
 	q.MoveAllToActiveQueue()
 	if q.activeQ.data.Len() != 3 {
 		t.Error("Expected all items to be in activeQ.")
@@ -312,28 +325,28 @@ func TestPriorityQueue_AssignedPodAdded(t *testing.T) {
 		Spec: v1.PodSpec{NodeName: "machine1"},
 	}
 
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Add(&medPriorityPod)
 	// Add a couple of pods to the unschedulableQ.
-	q.unschedulableQ.addOrUpdate(&unschedulablePod)
-	q.unschedulableQ.addOrUpdate(affinityPod)
+	addOrUpdateUnschedulablePod(q, &unschedulablePod)
+	addOrUpdateUnschedulablePod(q, affinityPod)
 	// Simulate addition of an assigned pod. The pod has matching labels for
 	// affinityPod. So, affinityPod should go to activeQ.
 	q.AssignedPodAdded(&labelPod)
-	if q.unschedulableQ.get(affinityPod) != nil {
+	if getUnschedulablePod(q, affinityPod) != nil {
 		t.Error("affinityPod is still in the unschedulableQ.")
 	}
 	if _, exists, _ := q.activeQ.Get(affinityPod); !exists {
 		t.Error("affinityPod is not moved to activeQ.")
 	}
 	// Check that the other pod is still in the unschedulableQ.
-	if q.unschedulableQ.get(&unschedulablePod) == nil {
+	if getUnschedulablePod(q, &unschedulablePod) == nil {
 		t.Error("unschedulablePod is not in the unschedulableQ.")
 	}
 }
 
 func TestPriorityQueue_NominatedPodsForNode(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	q.Add(&medPriorityPod)
 	q.Add(&unschedulablePod)
 	q.Add(&highPriorityPod)
@@ -350,7 +363,7 @@ func TestPriorityQueue_NominatedPodsForNode(t *testing.T) {
 }
 
 func TestPriorityQueue_UpdateNominatedPodForNode(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	if err := q.Add(&medPriorityPod); err != nil {
 		t.Errorf("add failed: %v", err)
 	}
@@ -581,7 +594,7 @@ func TestSchedulingQueue_Close(t *testing.T) {
 		},
 		{
 			name:        "PriorityQueue close",
-			q:           NewPriorityQueue(),
+			q:           NewPriorityQueue(nil),
 			expectedErr: fmt.Errorf(queueClosed),
 		},
 	}
@@ -610,7 +623,7 @@ func TestSchedulingQueue_Close(t *testing.T) {
 // ensures that an unschedulable pod does not block head of the queue when there
 // are frequent events that move pods to the active queue.
 func TestRecentlyTriedPodsGoBack(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 	// Add a few pods to priority queue.
 	for i := 0; i < 5; i++ {
 		p := v1.Pod{
@@ -664,7 +677,7 @@ func TestRecentlyTriedPodsGoBack(t *testing.T) {
 // This behavior ensures that an unschedulable pod does not block head of the queue when there
 // are frequent events that move pods to the active queue.
 func TestPodFailedSchedulingMultipleTimesDoesNotBlockNewerPod(t *testing.T) {
-	q := NewPriorityQueue()
+	q := NewPriorityQueue(nil)
 
 	// Add an unschedulable pod to a priority queue.
 	// This makes a situation that the pod was tried to schedule
@@ -746,3 +759,63 @@ func TestPodFailedSchedulingMultipleTimesDoesNotBlockNewerPod(t *testing.T) {
 		t.Errorf("Expected that test-newer-pod was popped, got %v", p2.Name)
 	}
 }
+
+// TestHighProirotyFlushUnschedulableQLeftover tests that pods will be moved to
+// activeQ after one minutes if it is in unschedulableQ
+func TestHighProirotyFlushUnschedulableQLeftover(t *testing.T) {
+	q := NewPriorityQueue(nil)
+	midPod := v1.Pod{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test-midpod",
+			Namespace: "ns1",
+			UID:       types.UID("tp-mid"),
+		},
+		Spec: v1.PodSpec{
+			Priority: &midPriority,
+		},
+		Status: v1.PodStatus{
+			NominatedNodeName: "node1",
+		},
+	}
+	highPod := v1.Pod{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test-highpod",
+			Namespace: "ns1",
+			UID:       types.UID("tp-high"),
+		},
+		Spec: v1.PodSpec{
+			Priority: &highPriority,
+		},
+		Status: v1.PodStatus{
+			NominatedNodeName: "node1",
+		},
+	}
+
+	addOrUpdateUnschedulablePod(q, &highPod)
+	addOrUpdateUnschedulablePod(q, &midPod)
+
+	// Update pod condition to highPod.
+	podutil.UpdatePodCondition(&highPod.Status, &v1.PodCondition{
+		Type:          v1.PodScheduled,
+		Status:        v1.ConditionFalse,
+		Reason:        v1.PodReasonUnschedulable,
+		Message:       "fake scheduling failure",
+		LastProbeTime: metav1.Time{Time: time.Now().Add(-1 * unschedulableQTimeInterval)},
+	})
+
+	// Update pod condition to midPod.
+	podutil.UpdatePodCondition(&midPod.Status, &v1.PodCondition{
+		Type:          v1.PodScheduled,
+		Status:        v1.ConditionFalse,
+		Reason:        v1.PodReasonUnschedulable,
+		Message:       "fake scheduling failure",
+		LastProbeTime: metav1.Time{Time: time.Now().Add(-1 * unschedulableQTimeInterval)},
+	})
+
+	if p, err := q.Pop(); err != nil || p != &highPod {
+		t.Errorf("Expected: %v after Pop, but got: %v", highPriorityPod.Name, p.Name)
+	}
+	if p, err := q.Pop(); err != nil || p != &midPod {
+		t.Errorf("Expected: %v after Pop, but got: %v", medPriorityPod.Name, p.Name)
+	}
+}

pkg/scheduler/util/BUILD

@@ -25,6 +25,7 @@ go_library(
     name = "go_default_library",
     srcs = [
         "backoff_utils.go",
+        "clock.go",
         "utils.go",
     ],
     importpath = "k8s.io/kubernetes/pkg/scheduler/util",

pkg/scheduler/util/clock.go

+/*
+Copyright 2018 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 util
+
+import (
+	"time"
+)
+
+// Clock provides an interface for getting the current time
+type Clock interface {
+	Now() time.Time
+}
+
+// RealClock implements a clock using time
+type RealClock struct{}
+
+// Now returns the current time with time.Now
+func (RealClock) Now() time.Time {
+	return time.Now()
+}