Disable pod preemption by config

cmd/kube-scheduler/app/server.go

@@ -381,6 +381,8 @@ type SchedulerServer struct {
 	HealthzServer *http.Server
 	// MetricsServer is optional.
 	MetricsServer *http.Server
+	// Disable pod preemption or not.
+	DisablePreemption bool
 }
 
 // NewSchedulerServer creates a runnable SchedulerServer from configuration.
@@ -445,6 +447,7 @@ func NewSchedulerServer(config *componentconfig.KubeSchedulerConfiguration, mast
 		LeaderElection:                 leaderElectionConfig,
 		HealthzServer:                  healthzServer,
 		MetricsServer:                  metricsServer,
+		DisablePreemption:              config.DisablePreemption,
 	}, nil
 }
 
@@ -659,6 +662,7 @@ func (s *SchedulerServer) SchedulerConfig() (*scheduler.Config, error) {
 		storageClassInformer,
 		s.HardPodAffinitySymmetricWeight,
 		utilfeature.DefaultFeatureGate.Enabled(features.EnableEquivalenceClassCache),
+		s.DisablePreemption,
 	)
 
 	source := s.AlgorithmSource
@@ -716,5 +720,7 @@ func (s *SchedulerServer) SchedulerConfig() (*scheduler.Config, error) {
 	}
 	// Additional tweaks to the config produced by the configurator.
 	config.Recorder = s.Recorder
+
+	config.DisablePreemption = s.DisablePreemption
 	return config, nil
 }

pkg/apis/componentconfig/types.go

@@ -111,6 +111,9 @@ type KubeSchedulerConfiguration struct {
 	// Indicate the "all topologies" set for empty topologyKey when it's used for PreferredDuringScheduling pod anti-affinity.
 	// DEPRECATED: This is no longer used.
 	FailureDomains string
+
+	// DisablePreemption disables the pod preemption feature.
+	DisablePreemption bool
 }
 
 // KubeSchedulerLeaderElectionConfiguration expands LeaderElectionConfiguration

pkg/apis/componentconfig/v1alpha1/types.go

@@ -106,6 +106,9 @@ type KubeSchedulerConfiguration struct {
 
 	// Indicate the "all topologies" set for empty topologyKey when it's used for PreferredDuringScheduling pod anti-affinity.
 	FailureDomains string `json:"failureDomains"`
+
+	// DisablePreemption disables the pod preemption feature.
+	DisablePreemption bool `json:"disablePreemption"`
 }
 
 // LeaderElectionConfiguration defines the configuration of leader election

pkg/apis/componentconfig/v1alpha1/zz_generated.conversion.go

@@ -309,6 +309,7 @@ func autoConvert_v1alpha1_KubeSchedulerConfiguration_To_componentconfig_KubeSche
 	out.EnableProfiling = in.EnableProfiling
 	out.EnableContentionProfiling = in.EnableContentionProfiling
 	out.FailureDomains = in.FailureDomains
+	out.DisablePreemption = in.DisablePreemption
 	return nil
 }
 
@@ -334,6 +335,7 @@ func autoConvert_componentconfig_KubeSchedulerConfiguration_To_v1alpha1_KubeSche
 	out.EnableProfiling = in.EnableProfiling
 	out.EnableContentionProfiling = in.EnableContentionProfiling
 	out.FailureDomains = in.FailureDomains
+	out.DisablePreemption = in.DisablePreemption
 	return nil
 }
 

pkg/scheduler/algorithmprovider/defaults/compatibility_test.go

@@ -579,6 +579,7 @@ func TestCompatibility_v1_Scheduler(t *testing.T) {
 			informerFactory.Storage().V1().StorageClasses(),
 			v1.DefaultHardPodAffinitySymmetricWeight,
 			enableEquivalenceCache,
+			false,
 		).CreateFromConfig(policy); err != nil {
 			t.Errorf("%s: Error constructing: %v", v, err)
 			continue

pkg/scheduler/core/extender_test.go

@@ -506,7 +506,18 @@ func TestGenericSchedulerWithExtenders(t *testing.T) {
 		}
 		queue := NewSchedulingQueue()
 		scheduler := NewGenericScheduler(
-			cache, nil, queue, test.predicates, algorithm.EmptyPredicateMetadataProducer, test.prioritizers, algorithm.EmptyPriorityMetadataProducer, extenders, nil, schedulertesting.FakePersistentVolumeClaimLister{}, false)
+			cache,
+			nil,
+			queue,
+			test.predicates,
+			algorithm.EmptyPredicateMetadataProducer,
+			test.prioritizers,
+			algorithm.EmptyPriorityMetadataProducer,
+			extenders,
+			nil,
+			schedulertesting.FakePersistentVolumeClaimLister{},
+			false,
+			false)
 		podIgnored := &v1.Pod{}
 		machine, err := scheduler.Schedule(podIgnored, schedulertesting.FakeNodeLister(makeNodeList(test.nodes)))
 		if test.expectsErr {

pkg/scheduler/core/generic_scheduler.go

@@ -98,6 +98,7 @@ type genericScheduler struct {
 	cachedNodeInfoMap        map[string]*schedulercache.NodeInfo
 	volumeBinder             *volumebinder.VolumeBinder
 	pvcLister                corelisters.PersistentVolumeClaimLister
+	disablePreemption        bool
 }
 
 // Schedule tries to schedule the given pod to one of the nodes in the node list.
@@ -1107,7 +1108,9 @@ func NewGenericScheduler(
 	extenders []algorithm.SchedulerExtender,
 	volumeBinder *volumebinder.VolumeBinder,
 	pvcLister corelisters.PersistentVolumeClaimLister,
-	alwaysCheckAllPredicates bool) algorithm.ScheduleAlgorithm {
+	alwaysCheckAllPredicates bool,
+	disablePreemption bool,
+) algorithm.ScheduleAlgorithm {
 	return &genericScheduler{
 		cache:                    cache,
 		equivalenceCache:         eCache,
@@ -1121,5 +1124,6 @@ func NewGenericScheduler(
 		volumeBinder:             volumeBinder,
 		pvcLister:                pvcLister,
 		alwaysCheckAllPredicates: alwaysCheckAllPredicates,
+		disablePreemption:        disablePreemption,
 	}
 }

pkg/scheduler/core/generic_scheduler_test.go

@@ -409,7 +409,18 @@ func TestGenericScheduler(t *testing.T) {
 		pvcLister := schedulertesting.FakePersistentVolumeClaimLister(pvcs)
 
 		scheduler := NewGenericScheduler(
-			cache, nil, NewSchedulingQueue(), test.predicates, algorithm.EmptyPredicateMetadataProducer, test.prioritizers, algorithm.EmptyPriorityMetadataProducer, []algorithm.SchedulerExtender{}, nil, pvcLister, test.alwaysCheckAllPredicates)
+			cache,
+			nil,
+			NewSchedulingQueue(),
+			test.predicates,
+			algorithm.EmptyPredicateMetadataProducer,
+			test.prioritizers,
+			algorithm.EmptyPriorityMetadataProducer,
+			[]algorithm.SchedulerExtender{},
+			nil,
+			pvcLister,
+			test.alwaysCheckAllPredicates,
+			false)
 		machine, err := scheduler.Schedule(test.pod, schedulertesting.FakeNodeLister(makeNodeList(test.nodes)))
 
 		if !reflect.DeepEqual(err, test.wErr) {
@@ -1323,7 +1334,18 @@ func TestPreempt(t *testing.T) {
 			extenders = append(extenders, extender)
 		}
 		scheduler := NewGenericScheduler(
-			cache, nil, NewSchedulingQueue(), map[string]algorithm.FitPredicate{"matches": algorithmpredicates.PodFitsResources}, algorithm.EmptyPredicateMetadataProducer, []algorithm.PriorityConfig{{Function: numericPriority, Weight: 1}}, algorithm.EmptyPriorityMetadataProducer, extenders, nil, schedulertesting.FakePersistentVolumeClaimLister{}, false)
+			cache,
+			nil,
+			NewSchedulingQueue(),
+			map[string]algorithm.FitPredicate{"matches": algorithmpredicates.PodFitsResources},
+			algorithm.EmptyPredicateMetadataProducer,
+			[]algorithm.PriorityConfig{{Function: numericPriority, Weight: 1}},
+			algorithm.EmptyPriorityMetadataProducer,
+			extenders,
+			nil,
+			schedulertesting.FakePersistentVolumeClaimLister{},
+			false,
+			false)
 		// Call Preempt and check the expected results.
 		node, victims, _, err := scheduler.Preempt(test.pod, schedulertesting.FakeNodeLister(makeNodeList(nodeNames)), error(&FitError{Pod: test.pod, FailedPredicates: failedPredMap}))
 		if err != nil {

pkg/scheduler/factory/factory.go

@@ -133,6 +133,9 @@ type configFactory struct {
 
 	// always check all predicates even if the middle of one predicate fails.
 	alwaysCheckAllPredicates bool
+
+	// Disable pod preemption or not.
+	disablePreemption bool
 }
 
 // NewConfigFactory initializes the default implementation of a Configurator To encourage eventual privatization of the struct type, we only
@@ -152,6 +155,7 @@ func NewConfigFactory(
 	storageClassInformer storageinformers.StorageClassInformer,
 	hardPodAffinitySymmetricWeight int32,
 	enableEquivalenceClassCache bool,
+	disablePreemption bool,
 ) scheduler.Configurator {
 	stopEverything := make(chan struct{})
 	schedulerCache := schedulercache.New(30*time.Second, stopEverything)
@@ -179,6 +183,7 @@ func NewConfigFactory(
 		schedulerName:                  schedulerName,
 		hardPodAffinitySymmetricWeight: hardPodAffinitySymmetricWeight,
 		enableEquivalenceClassCache:    enableEquivalenceClassCache,
+		disablePreemption:              disablePreemption,
 	}
 
 	c.scheduledPodsHasSynced = podInformer.Informer().HasSynced
@@ -1064,7 +1069,20 @@ func (c *configFactory) CreateFromKeys(predicateKeys, priorityKeys sets.String,
 		glog.Info("Created equivalence class cache")
 	}
 
-	algo := core.NewGenericScheduler(c.schedulerCache, c.equivalencePodCache, c.podQueue, predicateFuncs, predicateMetaProducer, priorityConfigs, priorityMetaProducer, extenders, c.volumeBinder, c.pVCLister, c.alwaysCheckAllPredicates)
+	algo := core.NewGenericScheduler(
+		c.schedulerCache,
+		c.equivalencePodCache,
+		c.podQueue,
+		predicateFuncs,
+		predicateMetaProducer,
+		priorityConfigs,
+		priorityMetaProducer,
+		extenders,
+		c.volumeBinder,
+		c.pVCLister,
+		c.alwaysCheckAllPredicates,
+		c.disablePreemption,
+	)
 
 	podBackoff := util.CreateDefaultPodBackoff()
 	return &scheduler.Config{

pkg/scheduler/factory/factory_test.go

@@ -46,7 +46,10 @@ import (
 	"k8s.io/kubernetes/pkg/scheduler/util"
 )
 
-const enableEquivalenceCache = true
+const (
+	enableEquivalenceCache = true
+	disablePodPreemption   = false
+)
 
 func TestCreate(t *testing.T) {
 	handler := utiltesting.FakeHandler{
@@ -533,6 +536,7 @@ func newConfigFactory(client *clientset.Clientset, hardPodAffinitySymmetricWeigh
 		informerFactory.Storage().V1().StorageClasses(),
 		hardPodAffinitySymmetricWeight,
 		enableEquivalenceCache,
+		disablePodPreemption,
 	)
 }
 

pkg/scheduler/scheduler.go

@@ -137,6 +137,9 @@ type Config struct {
 
 	// VolumeBinder handles PVC/PV binding for the pod.
 	VolumeBinder *volumebinder.VolumeBinder
+
+	// Disable pod preemption or not.
+	DisablePreemption bool
 }
 
 // NewFromConfigurator returns a new scheduler that is created entirely by the Configurator.  Assumes Create() is implemented.
@@ -207,8 +210,9 @@ func (sched *Scheduler) schedule(pod *v1.Pod) (string, error) {
 // If it succeeds, it adds the name of the node where preemption has happened to the pod annotations.
 // It returns the node name and an error if any.
 func (sched *Scheduler) preempt(preemptor *v1.Pod, scheduleErr error) (string, error) {
-	if !util.PodPriorityEnabled() {
+	if !util.PodPriorityEnabled() || sched.config.DisablePreemption {
-		glog.V(3).Infof("Pod priority feature is not enabled. No preemption is performed.")
+		glog.V(3).Infof("Pod priority feature is not enabled or preemption is disabled by scheduler configuration." +
+			" No preemption is performed.")
 		return "", nil
 	}
 	preemptor, err := sched.config.PodPreemptor.GetUpdatedPod(preemptor)

pkg/scheduler/scheduler_test.go

@@ -548,6 +548,7 @@ func setupTestScheduler(queuedPodStore *clientcache.FIFO, scache schedulercache.
 		[]algorithm.SchedulerExtender{},
 		nil,
 		schedulertesting.FakePersistentVolumeClaimLister{},
+		false,
 		false)
 	bindingChan := make(chan *v1.Binding, 1)
 	errChan := make(chan error, 1)
@@ -596,6 +597,7 @@ func setupTestSchedulerLongBindingWithRetry(queuedPodStore *clientcache.FIFO, sc
 		[]algorithm.SchedulerExtender{},
 		nil,
 		schedulertesting.FakePersistentVolumeClaimLister{},
+		false,
 		false)
 	bindingChan := make(chan *v1.Binding, 2)
 	configurator := &FakeConfigurator{

test/integration/scheduler/util.go

@@ -85,6 +85,7 @@ func createConfiguratorWithPodInformer(
 		informerFactory.Storage().V1().StorageClasses(),
 		v1.DefaultHardPodAffinitySymmetricWeight,
 		utilfeature.DefaultFeatureGate.Enabled(features.EnableEquivalenceClassCache),
+		false,
 	)
 }
 

test/integration/util/util.go

@@ -112,5 +112,6 @@ func createSchedulerConfigurator(
 		informerFactory.Storage().V1().StorageClasses(),
 		v1.DefaultHardPodAffinitySymmetricWeight,
 		utilfeature.DefaultFeatureGate.Enabled(features.EnableEquivalenceClassCache),
+		false,
 	)
 }