kubernetes/pkg/scheduler/scheduler.go

/*
Copyright 2014 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 scheduler

import (
	"fmt"
	"time"

	"k8s.io/api/core/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/apimachinery/pkg/util/sets"
	"k8s.io/apimachinery/pkg/util/wait"
	utilfeature "k8s.io/apiserver/pkg/util/feature"
	clientset "k8s.io/client-go/kubernetes"
	corelisters "k8s.io/client-go/listers/core/v1"
	"k8s.io/client-go/tools/record"
	"k8s.io/kubernetes/pkg/features"
	"k8s.io/kubernetes/pkg/scheduler/algorithm"
	"k8s.io/kubernetes/pkg/scheduler/algorithm/predicates"
	schedulerapi "k8s.io/kubernetes/pkg/scheduler/api"
	"k8s.io/kubernetes/pkg/scheduler/core"
	"k8s.io/kubernetes/pkg/scheduler/metrics"
	"k8s.io/kubernetes/pkg/scheduler/schedulercache"
	"k8s.io/kubernetes/pkg/scheduler/util"
	"k8s.io/kubernetes/pkg/scheduler/volumebinder"

	"github.com/golang/glog"
)

// Binder knows how to write a binding.
type Binder interface {
	Bind(binding *v1.Binding) error
}

// PodConditionUpdater updates the condition of a pod based on the passed
// PodCondition
type PodConditionUpdater interface {
	Update(pod *v1.Pod, podCondition *v1.PodCondition) error
}

// PodPreemptor has methods needed to delete a pod and to update
// annotations of the preemptor pod.
type PodPreemptor interface {
	GetUpdatedPod(pod *v1.Pod) (*v1.Pod, error)
	DeletePod(pod *v1.Pod) error
	SetNominatedNodeName(pod *v1.Pod, nominatedNode string) error
	RemoveNominatedNodeName(pod *v1.Pod) error
}

// Scheduler watches for new unscheduled pods. It attempts to find
// nodes that they fit on and writes bindings back to the api server.
type Scheduler struct {
	config *Config
}

// StopEverything closes the scheduler config's StopEverything channel, to shut
// down the Scheduler.
func (sched *Scheduler) StopEverything() {
	close(sched.config.StopEverything)
}

// Configurator defines I/O, caching, and other functionality needed to
// construct a new scheduler. An implementation of this can be seen in
// factory.go.
type Configurator interface {
	GetPriorityFunctionConfigs(priorityKeys sets.String) ([]algorithm.PriorityConfig, error)
	GetPriorityMetadataProducer() (algorithm.PriorityMetadataProducer, error)
	GetPredicateMetadataProducer() (algorithm.PredicateMetadataProducer, error)
	GetPredicates(predicateKeys sets.String) (map[string]algorithm.FitPredicate, error)
	GetHardPodAffinitySymmetricWeight() int32
	GetSchedulerName() string
	MakeDefaultErrorFunc(backoff *util.PodBackoff, podQueue core.SchedulingQueue) func(pod *v1.Pod, err error)

	// Needs to be exposed for things like integration tests where we want to make fake nodes.
	GetNodeLister() corelisters.NodeLister
	GetClient() clientset.Interface
	GetScheduledPodLister() corelisters.PodLister

	Create() (*Config, error)
	CreateFromProvider(providerName string) (*Config, error)
	CreateFromConfig(policy schedulerapi.Policy) (*Config, error)
	CreateFromKeys(predicateKeys, priorityKeys sets.String, extenders []algorithm.SchedulerExtender) (*Config, error)
}

// Config is an implementation of the Scheduler's configured input data.
// TODO over time we should make this struct a hidden implementation detail of the scheduler.
type Config struct {
	// It is expected that changes made via SchedulerCache will be observed
	// by NodeLister and Algorithm.
	SchedulerCache schedulercache.Cache
	// Ecache is used for optimistically invalid affected cache items after
	// successfully binding a pod
	Ecache     *core.EquivalenceCache
	NodeLister algorithm.NodeLister
	Algorithm  algorithm.ScheduleAlgorithm
	GetBinder  func(pod *v1.Pod) Binder
	// PodConditionUpdater is used only in case of scheduling errors. If we succeed
	// with scheduling, PodScheduled condition will be updated in apiserver in /bind
	// handler so that binding and setting PodCondition it is atomic.
	PodConditionUpdater PodConditionUpdater
	// PodPreemptor is used to evict pods and update pod annotations.
	PodPreemptor PodPreemptor

	// NextPod should be a function that blocks until the next pod
	// is available. We don't use a channel for this, because scheduling
	// a pod may take some amount of time and we don't want pods to get
	// stale while they sit in a channel.
	NextPod func() *v1.Pod

	// WaitForCacheSync waits for scheduler cache to populate.
	// It returns true if it was successful, false if the controller should shutdown.
	WaitForCacheSync func() bool

	// Error is called if there is an error. It is passed the pod in
	// question, and the error
	Error func(*v1.Pod, error)

	// Recorder is the EventRecorder to use
	Recorder record.EventRecorder

	// Close this to shut down the scheduler.
	StopEverything chan struct{}

	// VolumeBinder handles PVC/PV binding for the pod.
	VolumeBinder *volumebinder.VolumeBinder
}

// NewFromConfigurator returns a new scheduler that is created entirely by the Configurator.  Assumes Create() is implemented.
// Supports intermediate Config mutation for now if you provide modifier functions which will run after Config is created.
func NewFromConfigurator(c Configurator, modifiers ...func(c *Config)) (*Scheduler, error) {
	cfg, err := c.Create()
	if err != nil {
		return nil, err
	}
	// Mutate it if any functions were provided, changes might be required for certain types of tests (i.e. change the recorder).
	for _, modifier := range modifiers {
		modifier(cfg)
	}
	// From this point on the config is immutable to the outside.
	s := &Scheduler{
		config: cfg,
	}
	metrics.Register()
	return s, nil
}

// NewFromConfig returns a new scheduler using the provided Config.
func NewFromConfig(config *Config) *Scheduler {
	metrics.Register()
	return &Scheduler{
		config: config,
	}
}

// Run begins watching and scheduling. It waits for cache to be synced, then starts a goroutine and returns immediately.
func (sched *Scheduler) Run() {
	if !sched.config.WaitForCacheSync() {
		return
	}

	if utilfeature.DefaultFeatureGate.Enabled(features.VolumeScheduling) {
		go sched.config.VolumeBinder.Run(sched.bindVolumesWorker, sched.config.StopEverything)
	}

	go wait.Until(sched.scheduleOne, 0, sched.config.StopEverything)
}

// Config return scheduler's config pointer. It is exposed for testing purposes.
func (sched *Scheduler) Config() *Config {
	return sched.config
}

// schedule implements the scheduling algorithm and returns the suggested host.
func (sched *Scheduler) schedule(pod *v1.Pod) (string, error) {
	host, err := sched.config.Algorithm.Schedule(pod, sched.config.NodeLister)
	if err != nil {
		glog.V(1).Infof("Failed to schedule pod: %v/%v", pod.Namespace, pod.Name)
		pod = pod.DeepCopy()
		sched.config.Error(pod, err)
		sched.config.Recorder.Eventf(pod, v1.EventTypeWarning, "FailedScheduling", "%v", err)
		sched.config.PodConditionUpdater.Update(pod, &v1.PodCondition{
			Type:    v1.PodScheduled,
			Status:  v1.ConditionFalse,
			Reason:  v1.PodReasonUnschedulable,
			Message: err.Error(),
		})
		return "", err
	}
	return host, err
}

// preempt tries to create room for a pod that has failed to schedule, by preempting lower priority pods if possible.
// 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() {
		glog.V(3).Infof("Pod priority feature is not enabled. No preemption is performed.")
		return "", nil
	}
	preemptor, err := sched.config.PodPreemptor.GetUpdatedPod(preemptor)
	if err != nil {
		glog.Errorf("Error getting the updated preemptor pod object: %v", err)
		return "", err
	}

	node, victims, nominatedPodsToClear, err := sched.config.Algorithm.Preempt(preemptor, sched.config.NodeLister, scheduleErr)
	metrics.PreemptionVictims.Set(float64(len(victims)))
	if err != nil {
		glog.Errorf("Error preempting victims to make room for %v/%v.", preemptor.Namespace, preemptor.Name)
		return "", err
	}
	var nodeName = ""
	if node != nil {
		nodeName = node.Name
		err = sched.config.PodPreemptor.SetNominatedNodeName(preemptor, nodeName)
		if err != nil {
			glog.Errorf("Error in preemption process. Cannot update pod %v annotations: %v", preemptor.Name, err)
			return "", err
		}
		for _, victim := range victims {
			if err := sched.config.PodPreemptor.DeletePod(victim); err != nil {
				glog.Errorf("Error preempting pod %v/%v: %v", victim.Namespace, victim.Name, err)
				return "", err
			}
			sched.config.Recorder.Eventf(victim, v1.EventTypeNormal, "Preempted", "by %v/%v on node %v", preemptor.Namespace, preemptor.Name, nodeName)
		}
	}
	// Clearing nominated pods should happen outside of "if node != nil". Node could
	// be nil when a pod with nominated node name is eligible to preempt again,
	// but preemption logic does not find any node for it. In that case Preempt()
	// function of generic_scheduler.go returns the pod itself for removal of the annotation.
	for _, p := range nominatedPodsToClear {
		rErr := sched.config.PodPreemptor.RemoveNominatedNodeName(p)
		if rErr != nil {
			glog.Errorf("Cannot remove nominated node annotation of pod: %v", rErr)
			// We do not return as this error is not critical.
		}
	}
	return nodeName, err
}

// assumeAndBindVolumes will update the volume cache and then asynchronously bind volumes if required.
//
// If volume binding is required, then the bind volumes routine will update the pod to send it back through
// the scheduler.
//
// Otherwise, return nil error and continue to assume the pod.
//
// This function modifies assumed if volume binding is required.
func (sched *Scheduler) assumeAndBindVolumes(assumed *v1.Pod, host string) error {
	if utilfeature.DefaultFeatureGate.Enabled(features.VolumeScheduling) {
		allBound, bindingRequired, err := sched.config.VolumeBinder.Binder.AssumePodVolumes(assumed, host)
		if err != nil {
			sched.config.Error(assumed, err)
			sched.config.Recorder.Eventf(assumed, v1.EventTypeWarning, "FailedScheduling", "AssumePodVolumes failed: %v", err)
			sched.config.PodConditionUpdater.Update(assumed, &v1.PodCondition{
				Type:    v1.PodScheduled,
				Status:  v1.ConditionFalse,
				Reason:  "SchedulerError",
				Message: err.Error(),
			})
			return err
		}
		if !allBound {
			err = fmt.Errorf("Volume binding started, waiting for completion")
			if bindingRequired {
				if sched.config.Ecache != nil {
					invalidPredicates := sets.NewString(predicates.CheckVolumeBindingPred)
					sched.config.Ecache.InvalidateCachedPredicateItemOfAllNodes(invalidPredicates)
				}

				// bindVolumesWorker() will update the Pod object to put it back in the scheduler queue
				sched.config.VolumeBinder.BindQueue.Add(assumed)
			} else {
				// We are just waiting for PV controller to finish binding, put it back in the
				// scheduler queue
				sched.config.Error(assumed, err)
				sched.config.Recorder.Eventf(assumed, v1.EventTypeNormal, "FailedScheduling", "%v", err)
				sched.config.PodConditionUpdater.Update(assumed, &v1.PodCondition{
					Type:   v1.PodScheduled,
					Status: v1.ConditionFalse,
					Reason: "VolumeBindingWaiting",
				})
			}
			return err
		}
	}
	return nil
}

// bindVolumesWorker() processes pods queued in assumeAndBindVolumes() and tries to
// make the API update for volume binding.
// This function runs forever until the volume BindQueue is closed.
func (sched *Scheduler) bindVolumesWorker() {
	workFunc := func() bool {
		keyObj, quit := sched.config.VolumeBinder.BindQueue.Get()
		if quit {
			return true
		}
		defer sched.config.VolumeBinder.BindQueue.Done(keyObj)

		assumed, ok := keyObj.(*v1.Pod)
		if !ok {
			glog.V(4).Infof("Object is not a *v1.Pod")
			return false
		}

		// TODO: add metrics
		var reason string
		var eventType string

		glog.V(5).Infof("Trying to bind volumes for pod \"%v/%v\"", assumed.Namespace, assumed.Name)

		// The Pod is always sent back to the scheduler afterwards.
		err := sched.config.VolumeBinder.Binder.BindPodVolumes(assumed)
		if err != nil {
			glog.V(1).Infof("Failed to bind volumes for pod \"%v/%v\"", assumed.Namespace, assumed.Name, err)
			reason = "VolumeBindingFailed"
			eventType = v1.EventTypeWarning
		} else {
			glog.V(4).Infof("Successfully bound volumes for pod \"%v/%v\"", assumed.Namespace, assumed.Name)
			reason = "VolumeBindingWaiting"
			eventType = v1.EventTypeNormal
			err = fmt.Errorf("Volume binding started, waiting for completion")
		}

		// Always fail scheduling regardless of binding success.
		// The Pod needs to be sent back through the scheduler to:
		// * Retry volume binding if it fails.
		// * Retry volume binding if dynamic provisioning fails.
		// * Bind the Pod to the Node once all volumes are bound.
		sched.config.Error(assumed, err)
		sched.config.Recorder.Eventf(assumed, eventType, "FailedScheduling", "%v", err)
		sched.config.PodConditionUpdater.Update(assumed, &v1.PodCondition{
			Type:   v1.PodScheduled,
			Status: v1.ConditionFalse,
			Reason: reason,
		})
		return false
	}

	for {
		if quit := workFunc(); quit {
			glog.V(4).Infof("bindVolumesWorker shutting down")
			break
		}
	}
}

// assume signals to the cache that a pod is already in the cache, so that binding can be asynchronous.
// assume modifies `assumed`.
func (sched *Scheduler) assume(assumed *v1.Pod, host string) error {
	// Optimistically assume that the binding will succeed and send it to apiserver
	// in the background.
	// If the binding fails, scheduler will release resources allocated to assumed pod
	// immediately.
	assumed.Spec.NodeName = host
	if err := sched.config.SchedulerCache.AssumePod(assumed); err != nil {
		glog.Errorf("scheduler cache AssumePod failed: %v", err)

		// This is most probably result of a BUG in retrying logic.
		// We report an error here so that pod scheduling can be retried.
		// This relies on the fact that Error will check if the pod has been bound
		// to a node and if so will not add it back to the unscheduled pods queue
		// (otherwise this would cause an infinite loop).
		sched.config.Error(assumed, err)
		sched.config.Recorder.Eventf(assumed, v1.EventTypeWarning, "FailedScheduling", "AssumePod failed: %v", err)
		sched.config.PodConditionUpdater.Update(assumed, &v1.PodCondition{
			Type:    v1.PodScheduled,
			Status:  v1.ConditionFalse,
			Reason:  "SchedulerError",
			Message: err.Error(),
		})
		return err
	}

	// Optimistically assume that the binding will succeed, so we need to invalidate affected
	// predicates in equivalence cache.
	// If the binding fails, these invalidated item will not break anything.
	if sched.config.Ecache != nil {
		sched.config.Ecache.InvalidateCachedPredicateItemForPodAdd(assumed, host)
	}
	return nil
}

// bind binds a pod to a given node defined in a binding object.  We expect this to run asynchronously, so we
// handle binding metrics internally.
func (sched *Scheduler) bind(assumed *v1.Pod, b *v1.Binding) error {
	bindingStart := time.Now()
	// If binding succeeded then PodScheduled condition will be updated in apiserver so that
	// it's atomic with setting host.
	err := sched.config.GetBinder(assumed).Bind(b)
	if err := sched.config.SchedulerCache.FinishBinding(assumed); err != nil {
		glog.Errorf("scheduler cache FinishBinding failed: %v", err)
	}
	if err != nil {
		glog.V(1).Infof("Failed to bind pod: %v/%v", assumed.Namespace, assumed.Name)
		if err := sched.config.SchedulerCache.ForgetPod(assumed); err != nil {
			glog.Errorf("scheduler cache ForgetPod failed: %v", err)
		}
		sched.config.Error(assumed, err)
		sched.config.Recorder.Eventf(assumed, v1.EventTypeWarning, "FailedScheduling", "Binding rejected: %v", err)
		sched.config.PodConditionUpdater.Update(assumed, &v1.PodCondition{
			Type:   v1.PodScheduled,
			Status: v1.ConditionFalse,
			Reason: "BindingRejected",
		})
		return err
	}

	metrics.BindingLatency.Observe(metrics.SinceInMicroseconds(bindingStart))
	sched.config.Recorder.Eventf(assumed, v1.EventTypeNormal, "Scheduled", "Successfully assigned %v/%v to %v", assumed.Namespace, assumed.Name, b.Target.Name)
	return nil
}

// scheduleOne does the entire scheduling workflow for a single pod.  It is serialized on the scheduling algorithm's host fitting.
func (sched *Scheduler) scheduleOne() {
	pod := sched.config.NextPod()
	if pod.DeletionTimestamp != nil {
		sched.config.Recorder.Eventf(pod, v1.EventTypeWarning, "FailedScheduling", "skip schedule deleting pod: %v/%v", pod.Namespace, pod.Name)
		glog.V(3).Infof("Skip schedule deleting pod: %v/%v", pod.Namespace, pod.Name)
		return
	}

	glog.V(3).Infof("Attempting to schedule pod: %v/%v", pod.Namespace, pod.Name)

	// Synchronously attempt to find a fit for the pod.
	start := time.Now()
	suggestedHost, err := sched.schedule(pod)
	if err != nil {
		// schedule() may have failed because the pod would not fit on any host, so we try to
		// preempt, with the expectation that the next time the pod is tried for scheduling it
		// will fit due to the preemption. It is also possible that a different pod will schedule
		// into the resources that were preempted, but this is harmless.
		if fitError, ok := err.(*core.FitError); ok {
			preemptionStartTime := time.Now()
			sched.preempt(pod, fitError)
			metrics.PreemptionAttempts.Inc()
			metrics.SchedulingAlgorithmPremptionEvaluationDuration.Observe(metrics.SinceInMicroseconds(preemptionStartTime))
		}
		return
	}
	metrics.SchedulingAlgorithmLatency.Observe(metrics.SinceInMicroseconds(start))
	// Tell the cache to assume that a pod now is running on a given node, even though it hasn't been bound yet.
	// This allows us to keep scheduling without waiting on binding to occur.
	assumedPod := pod.DeepCopy()

	// Assume volumes first before assuming the pod.
	//
	// If no volumes need binding, then nil is returned, and continue to assume the pod.
	//
	// Otherwise, error is returned and volume binding is started asynchronously for all of the pod's volumes.
	// scheduleOne() returns immediately on error, so that it doesn't continue to assume the pod.
	//
	// After the asynchronous volume binding updates are made, it will send the pod back through the scheduler for
	// subsequent passes until all volumes are fully bound.
	//
	// This function modifies 'assumedPod' if volume binding is required.
	err = sched.assumeAndBindVolumes(assumedPod, suggestedHost)
	if err != nil {
		return
	}

	// assume modifies `assumedPod` by setting NodeName=suggestedHost
	err = sched.assume(assumedPod, suggestedHost)
	if err != nil {
		return
	}
	// bind the pod to its host asynchronously (we can do this b/c of the assumption step above).
	go func() {
		err := sched.bind(assumedPod, &v1.Binding{
			ObjectMeta: metav1.ObjectMeta{Namespace: assumedPod.Namespace, Name: assumedPod.Name, UID: assumedPod.UID},
			Target: v1.ObjectReference{
				Kind: "Node",
				Name: suggestedHost,
			},
		})
		metrics.E2eSchedulingLatency.Observe(metrics.SinceInMicroseconds(start))
		if err != nil {
			glog.Errorf("Internal error binding pod: (%v)", err)
		}
	}()
}