深入分析Kubernetes Scheduler的优先级队列

type PriorityQueue struct {
	lock sync.RWMutex
	cond sync.Cond
	
	activeQ *Heap
	
	unschedulableQ *UnschedulablePodsMap
	
	nominatedPods map[string][]*v1.Pod
	
	receivedMoveRequest bool
}

type Heap struct {
	data *heapData
}


type heapData struct {
	
	items map[string]*heapItem
	
	queue []string

	keyFunc KeyFunc
	
	lessFunc LessFunc
}

type heapItem struct {
	obj   interface{} // The object which is stored in the heap.
	index int         // The index of the object's key in the Heap.queue.
}

func NewSchedulingQueue() SchedulingQueue {
	if util.PodPriorityEnabled() {
		return NewPriorityQueue()
	}
	return NewFIFO()
}

// NewPriorityQueue creates a PriorityQueue object.
func NewPriorityQueue() *PriorityQueue {
	pq := &PriorityQueue{
		activeQ:        newHeap(cache.MetaNamespaceKeyFunc, util.HigherPriorityPod),
		unschedulableQ: newUnschedulablePodsMap(),
		nominatedPods:  map[string][]*v1.Pod{},
	}
	pq.cond.L = &pq.lock
	return pq
}

func MetaNamespaceKeyFunc(obj interface{}) (string, error) {
	if key, ok := obj.(ExplicitKey); ok {
		return string(key), nil
	}
	meta, err := meta.Accessor(obj)
	if err != nil {
		return "", fmt.Errorf("object has no meta: %v", err)
	}
	if len(meta.GetNamespace()) > 0 {
		return meta.GetNamespace() + "/" + meta.GetName(), nil
	}
	return meta.GetName(), nil
}

const (
	DefaultPriorityWhenNoDefaultClassExists = 0
)

func HigherPriorityPod(pod1, pod2 interface{}) bool {
	return GetPodPriority(pod1.(*v1.Pod)) > GetPodPriority(pod2.(*v1.Pod))
}

func GetPodPriority(pod *v1.Pod) int32 {
	if pod.Spec.Priority != nil {
		return *pod.Spec.Priority
	}
	
	return scheduling.DefaultPriorityWhenNoDefaultClassExists
}

func newUnschedulablePodsMap() *UnschedulablePodsMap {
	return &UnschedulablePodsMap{
		pods:    make(map[string]*v1.Pod),
		keyFunc: util.GetPodFullName,
	}
}

func GetPodFullName(pod *v1.Pod) string {
	return pod.Name + "_" + pod.Namespace
}

func (h *Heap) Add(obj interface{}) error {
	key, err := h.data.keyFunc(obj)
	if err != nil {
		return cache.KeyError{Obj: obj, Err: err}
	}
	if _, exists := h.data.items[key]; exists {
		h.data.items[key].obj = obj
		heap.Fix(h.data, h.data.items[key].index)
	} else {
		heap.Push(h.data, &itemKeyValue{key, obj})
	}
	return nil
}

func Push(h Interface, x interface{}) {
	h.Push(x)
	up(h, h.Len()-1)
}

func up(h Interface, j int) {
	for {
		i := (j - 1) / 2 // parent
		if i == j || !h.Less(j, i) {
			break
		}
		h.Swap(i, j)
		j = i
	}
}

func (h *heapData) Less(i, j int) bool {
	if i > len(h.queue) || j > len(h.queue) {
		return false
	}
	itemi, ok := h.items[h.queue[i]]
	if !ok {
		return false
	}
	itemj, ok := h.items[h.queue[j]]
	if !ok {
		return false
	}
	return h.lessFunc(itemi.obj, itemj.obj)
}

func (h *Heap) Pop() (interface{}, error) {
	obj := heap.Pop(h.data)
	if obj != nil {
		return obj, nil
	}
	return nil, fmt.Errorf("object was removed from heap data")
}

func Pop(h Interface) interface{} {
	n := h.Len() - 1
	h.Swap(0, n)
	down(h, 0, n)
	return h.Pop()
}

func down(h Interface, i, n int) {
	for {
		j1 := 2*i + 1
		if j1 >= n || j1 < 0 { // j1 < 0 after int overflow
			break
		}
		j := j1 // left child
		if j2 := j1 + 1; j2 < n && !h.Less(j1, j2) {
			j = j2 // = 2*i + 2  // right child
		}
		if !h.Less(j, i) {
			break
		}
		h.Swap(i, j)
		i = j
	}
}

func NewConfigFactory(...) scheduler.Configurator {
	...
	// scheduled pod cache
	podInformer.Informer().AddEventHandler(
		cache.FilteringResourceEventHandler{
			FilterFunc: func(obj interface{}) bool {
				switch t := obj.(type) {
				case *v1.Pod:
					return assignedNonTerminatedPod(t)
				case cache.DeletedFinalStateUnknown:
					if pod, ok := t.Obj.(*v1.Pod); ok {
						return assignedNonTerminatedPod(pod)
					}
					runtime.HandleError(fmt.Errorf("unable to convert object %T to *v1.Pod in %T", obj, c))
					return false
				default:
					runtime.HandleError(fmt.Errorf("unable to handle object in %T: %T", c, obj))
					return false
				}
			},
			Handler: cache.ResourceEventHandlerFuncs{
				AddFunc:    c.addPodToCache,
				UpdateFunc: c.updatePodInCache,
				DeleteFunc: c.deletePodFromCache,
			},
		},
	)
	// unscheduled pod queue
	podInformer.Informer().AddEventHandler(
		cache.FilteringResourceEventHandler{
			FilterFunc: func(obj interface{}) bool {
				switch t := obj.(type) {
				case *v1.Pod:
					return unassignedNonTerminatedPod(t)
				case cache.DeletedFinalStateUnknown:
					if pod, ok := t.Obj.(*v1.Pod); ok {
						return unassignedNonTerminatedPod(pod)
					}
					runtime.HandleError(fmt.Errorf("unable to convert object %T to *v1.Pod in %T", obj, c))
					return false
				default:
					runtime.HandleError(fmt.Errorf("unable to handle object in %T: %T", c, obj))
					return false
				}
			},
			Handler: cache.ResourceEventHandlerFuncs{
				AddFunc:    c.addPodToSchedulingQueue,
				UpdateFunc: c.updatePodInSchedulingQueue,
				DeleteFunc: c.deletePodFromSchedulingQueue,
			},
		},
	)
	// ScheduledPodLister is something we provide to plug-in functions that
	// they may need to call.
	c.scheduledPodLister = assignedPodLister{podInformer.Lister()}

	nodeInformer.Informer().AddEventHandler(
		cache.ResourceEventHandlerFuncs{
			AddFunc:    c.addNodeToCache,
			UpdateFunc: c.updateNodeInCache,
			DeleteFunc: c.deleteNodeFromCache,
		},
	)
	c.nodeLister = nodeInformer.Lister()

	...

	// This is for MaxPDVolumeCountPredicate: add/delete PVC will affect counts of PV when it is bound.
	pvcInformer.Informer().AddEventHandler(
		cache.ResourceEventHandlerFuncs{
			AddFunc:    c.onPvcAdd,
			UpdateFunc: c.onPvcUpdate,
			DeleteFunc: c.onPvcDelete,
		},
	)
	c.pVCLister = pvcInformer.Lister()

	// This is for ServiceAffinity: affected by the selector of the service is updated.
	// Also, if new service is added, equivalence cache will also become invalid since
	// existing pods may be "captured" by this service and change this predicate result.
	serviceInformer.Informer().AddEventHandler(
		cache.ResourceEventHandlerFuncs{
			AddFunc:    c.onServiceAdd,
			UpdateFunc: c.onServiceUpdate,
			DeleteFunc: c.onServiceDelete,
		},
	)
	c.serviceLister = serviceInformer.Lister()

	...
}

// assignedNonTerminatedPod selects pods that are assigned and non-terminal (scheduled and running).
func assignedNonTerminatedPod(pod *v1.Pod) bool {
	if len(pod.Spec.NodeName) == 0 {
		return false
	}
	if pod.Status.Phase == v1.PodSucceeded || pod.Status.Phase == v1.PodFailed {
		return false
	}
	return true
}

func (c *configFactory) addPodToCache(obj interface{}) {
    ...

	c.podQueue.AssignedPodAdded(pod)
}

// AssignedPodAdded is called when a bound pod is added. Creation of this pod
// may make pending pods with matching affinity terms schedulable.
func (p *PriorityQueue) AssignedPodAdded(pod *v1.Pod) {
	p.movePodsToActiveQueue(p.getUnschedulablePodsWithMatchingAffinityTerm(pod))
}

func (p *PriorityQueue) movePodsToActiveQueue(pods []*v1.Pod) {
	p.lock.Lock()
	defer p.lock.Unlock()
	for _, pod := range pods {
		if err := p.activeQ.Add(pod); err == nil {
			p.unschedulableQ.delete(pod)
		} else {
			glog.Errorf("Error adding pod %v to the scheduling queue: %v", pod.Name, err)
		}
	}
	p.receivedMoveRequest = true
	p.cond.Broadcast()
}

// getUnschedulablePodsWithMatchingAffinityTerm returns unschedulable pods which have
// any affinity term that matches "pod".
func (p *PriorityQueue) getUnschedulablePodsWithMatchingAffinityTerm(pod *v1.Pod) []*v1.Pod {
	p.lock.RLock()
	defer p.lock.RUnlock()
	var podsToMove []*v1.Pod
	for _, up := range p.unschedulableQ.pods {
		affinity := up.Spec.Affinity
		if affinity != nil && affinity.PodAffinity != nil {
			terms := predicates.GetPodAffinityTerms(affinity.PodAffinity)
			for _, term := range terms {
				namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(up, &term)
				selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
				if err != nil {
					glog.Errorf("Error getting label selectors for pod: %v.", up.Name)
				}
				if priorityutil.PodMatchesTermsNamespaceAndSelector(pod, namespaces, selector) {
					podsToMove = append(podsToMove, up)
					break
				}
			}
		}
	}
	return podsToMove
}

func (p _PriorityQueue) AddUnschedulableIfNotPresent(pod_ v1.Pod) error {
p.lock.Lock()
defer p.lock.Unlock()

if p.unschedulableQ.get(pod) != nil {

	return fmt.Errorf("pod is already present in unschedulableQ")
}

if _, exists, _ := p.activeQ.Get(pod); exists {

	return fmt.Errorf("pod is already present in the activeQ")

}

if !p.receivedMoveRequest && isPodUnschedulable(pod) {

	p.unschedulableQ.addOrUpdate(pod)

	p.addNominatedPodIfNeeded(pod)

	return nil

}

err := p.activeQ.Add(pod)

if err == nil {

	p.addNominatedPodIfNeeded(pod)

	p.cond.Broadcast()

}

return err


}

func (c *configFactory) updatePodInCache(oldObj, newObj interface{}) {
	...
	c.podQueue.AssignedPodUpdated(newPod)
}


// AssignedPodUpdated is called when a bound pod is updated. Change of labels
// may make pending pods with matching affinity terms schedulable.
func (p *PriorityQueue) AssignedPodUpdated(pod *v1.Pod) {
	p.movePodsToActiveQueue(p.getUnschedulablePodsWithMatchingAffinityTerm(pod))
}

func (c *configFactory) deletePodFromCache(obj interface{}) {
	...
	c.podQueue.MoveAllToActiveQueue()
}

func (p *PriorityQueue) MoveAllToActiveQueue() {
	p.lock.Lock()
	defer p.lock.Unlock()
	for _, pod := range p.unschedulableQ.pods {
		if err := p.activeQ.Add(pod); err != nil {
			glog.Errorf("Error adding pod %v to the scheduling queue: %v", pod.Name, err)
		}
	}
	p.unschedulableQ.clear()
	p.receivedMoveRequest = true
	p.cond.Broadcast()
}

// unassignedNonTerminatedPod selects pods that are unassigned and non-terminal.
func unassignedNonTerminatedPod(pod *v1.Pod) bool {
	if len(pod.Spec.NodeName) != 0 {
		return false
	}
	if pod.Status.Phase == v1.PodSucceeded || pod.Status.Phase == v1.PodFailed {
		return false
	}
	return true
}

func (c *configFactory) addPodToSchedulingQueue(obj interface{}) {
	if err := c.podQueue.Add(obj.(*v1.Pod)); err != nil {
		runtime.HandleError(fmt.Errorf("unable to queue %T: %v", obj, err))
	}
}

func (p *PriorityQueue) Add(pod *v1.Pod) error {
	p.lock.Lock()
	defer p.lock.Unlock()
	err := p.activeQ.Add(pod)
	if err != nil {
		glog.Errorf("Error adding pod %v to the scheduling queue: %v", pod.Name, err)
	} else {
		if p.unschedulableQ.get(pod) != nil {
			glog.Errorf("Error: pod %v is already in the unschedulable queue.", pod.Name)
			p.deleteNominatedPodIfExists(pod)
			p.unschedulableQ.delete(pod)
		}
		p.addNominatedPodIfNeeded(pod)
		p.cond.Broadcast()
	}
	return err
}

func (c *configFactory) updatePodInSchedulingQueue(oldObj, newObj interface{}) {
	pod := newObj.(*v1.Pod)
	if c.skipPodUpdate(pod) {
		return
	}
	if err := c.podQueue.Update(oldObj.(*v1.Pod), pod); err != nil {
		runtime.HandleError(fmt.Errorf("unable to update %T: %v", newObj, err))
	}
}

func (c *configFactory) skipPodUpdate(pod *v1.Pod) bool {
	// Non-assumed pods should never be skipped.
	isAssumed, err := c.schedulerCache.IsAssumedPod(pod)
	if err != nil {
		runtime.HandleError(fmt.Errorf("failed to check whether pod %s/%s is assumed: %v", pod.Namespace, pod.Name, err))
		return false
	}
	if !isAssumed {
		return false
	}

	// Gets the assumed pod from the cache.
	assumedPod, err := c.schedulerCache.GetPod(pod)
	if err != nil {
		runtime.HandleError(fmt.Errorf("failed to get assumed pod %s/%s from cache: %v", pod.Namespace, pod.Name, err))
		return false
	}

	// Compares the assumed pod in the cache with the pod update. If they are
	// equal (with certain fields excluded), this pod update will be skipped.
	f := func(pod *v1.Pod) *v1.Pod {
		p := pod.DeepCopy()
		// ResourceVersion must be excluded because each object update will
		// have a new resource version.
		p.ResourceVersion = ""
		// Spec.NodeName must be excluded because the pod assumed in the cache
		// is expected to have a node assigned while the pod update may nor may
		// not have this field set.
		p.Spec.NodeName = ""
		// Annotations must be excluded for the reasons described in
		// https://github.com/kubernetes/kubernetes/issues/52914.
		p.Annotations = nil
		return p
	}
	assumedPodCopy, podCopy := f(assumedPod), f(pod)
	if !reflect.DeepEqual(assumedPodCopy, podCopy) {
		return false
	}
	glog.V(3).Infof("Skipping pod %s/%s update", pod.Namespace, pod.Name)
	return true
}

func (p *PriorityQueue) Update(oldPod, newPod *v1.Pod) error {
	p.lock.Lock()
	defer p.lock.Unlock()
	// If the pod is already in the active queue, just update it there.
	if _, exists, _ := p.activeQ.Get(newPod); exists {
		p.updateNominatedPod(oldPod, newPod)
		err := p.activeQ.Update(newPod)
		return err
	}
	// If the pod is in the unschedulable queue, updating it may make it schedulable.
	if usPod := p.unschedulableQ.get(newPod); usPod != nil {
		p.updateNominatedPod(oldPod, newPod)
		if isPodUpdated(oldPod, newPod) {
			p.unschedulableQ.delete(usPod)
			err := p.activeQ.Add(newPod)
			if err == nil {
				p.cond.Broadcast()
			}
			return err
		}
		p.unschedulableQ.addOrUpdate(newPod)
		return nil
	}
	// If pod is not in any of the two queue, we put it in the active queue.
	err := p.activeQ.Add(newPod)
	if err == nil {
		p.addNominatedPodIfNeeded(newPod)
		p.cond.Broadcast()
	}
	return err
}

func (c *configFactory) deletePodFromSchedulingQueue(obj interface{}) {
	...
	
	if err := c.podQueue.Delete(pod); err != nil {
		runtime.HandleError(fmt.Errorf("unable to dequeue %T: %v", obj, err))
	}
	...
}

func (p *PriorityQueue) Delete(pod *v1.Pod) error {
	p.lock.Lock()
	defer p.lock.Unlock()
	p.deleteNominatedPodIfExists(pod)
	err := p.activeQ.Delete(pod)
	if err != nil { // The item was probably not found in the activeQ.
		p.unschedulableQ.delete(pod)
	}
	return nil
}

func (c *configFactory) addNodeToCache(obj interface{}) {
	...

	c.podQueue.MoveAllToActiveQueue()
}

func (c *configFactory) updateNodeInCache(oldObj, newObj interface{}) {
	...
	
	c.podQueue.MoveAllToActiveQueue()
}

func (c *configFactory) onServiceAdd(obj interface{}) {
	...
	
	c.podQueue.MoveAllToActiveQueue()
}


func (c *configFactory) onServiceUpdate(oldObj interface{}, newObj interface{}) {
	...
	
	c.podQueue.MoveAllToActiveQueue()
}


func (c *configFactory) onServiceDelete(obj interface{}) {
	...
	
	c.podQueue.MoveAllToActiveQueue()
}

func (c *configFactory) onPvcAdd(obj interface{}) {
	...
	
	c.podQueue.MoveAllToActiveQueue()
}

func (c *configFactory) onPvcUpdate(old, new interface{}) {
	...
	c.podQueue.MoveAllToActiveQueue()
}