func getMatchingAntiAffinityTerms(pod *v1.Pod, nodeInfoMap map[string]*schedulercache.NodeInfo) ([]matchingPodAntiAffinityTerm, error) {
allNodeNames := make([]string, 0, len(nodeInfoMap))
for name := range nodeInfoMap {
allNodeNames = append(allNodeNames, name)
}
var lock sync.Mutex
var result []matchingPodAntiAffinityTerm
var firstError error
appendResult := func(toAppend []matchingPodAntiAffinityTerm) {
lock.Lock()
defer lock.Unlock()
result = append(result, toAppend...)
}
catchError := func(err error) {
lock.Lock()
defer lock.Unlock()
if firstError == nil {
firstError = err
}
}
processNode := func(i int) {
nodeInfo := nodeInfoMap[allNodeNames[i]]
node := nodeInfo.Node()
if node == nil {
catchError(fmt.Errorf("node not found"))
return
}
var nodeResult []matchingPodAntiAffinityTerm
for _, existingPod := range nodeInfo.PodsWithAffinity() {
affinity, err := v1.GetAffinityFromPodAnnotations(existingPod.Annotations)
if err != nil {
catchError(err)
return
}
if affinity == nil {
continue
}
for _, term := range getPodAntiAffinityTerms(affinity.PodAntiAffinity) {
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(pod, &term)
selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
if err != nil {
catchError(err)
return
}
match := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, namespaces, selector)
if match {
nodeResult = append(nodeResult, matchingPodAntiAffinityTerm{term: &term, node: node})
}
}
}
if len(nodeResult) > 0 {
appendResult(nodeResult)
}
}
workqueue.Parallelize(16, len(allNodeNames), processNode)
return result, firstError
}
// Filters the nodes to find the ones that fit based on the given predicate functions
// Each node is passed through the predicate functions to determine if it is a fit
func findNodesThatFit(
pod *api.Pod,
nodeNameToInfo map[string]*schedulercache.NodeInfo,
nodes []*api.Node,
predicateFuncs map[string]algorithm.FitPredicate,
extenders []algorithm.SchedulerExtender) ([]*api.Node, FailedPredicateMap, error) {
var filtered []*api.Node
failedPredicateMap := FailedPredicateMap{}
if len(predicateFuncs) == 0 {
filtered = nodes
} else {
// Create filtered list with enough space to avoid growing it
// and allow assigning.
filtered = make([]*api.Node, len(nodes))
meta := predicates.PredicateMetadata(pod, nodeNameToInfo)
errs := []error{}
var predicateResultLock sync.Mutex
var filteredLen int32
checkNode := func(i int) {
nodeName := nodes[i].Name
fits, failedPredicate, err := podFitsOnNode(pod, meta, nodeNameToInfo[nodeName], predicateFuncs)
if err != nil {
predicateResultLock.Lock()
errs = append(errs, err)
predicateResultLock.Unlock()
return
}
if fits {
filtered[atomic.AddInt32(&filteredLen, 1)-1] = nodes[i]
} else {
predicateResultLock.Lock()
failedPredicateMap[nodeName] = failedPredicate
predicateResultLock.Unlock()
}
}
workqueue.Parallelize(16, len(nodes), checkNode)
filtered = filtered[:filteredLen]
if len(errs) > 0 {
return []*api.Node{}, FailedPredicateMap{}, errors.NewAggregate(errs)
}
}
if len(filtered) > 0 && len(extenders) != 0 {
for _, extender := range extenders {
filteredList, err := extender.Filter(pod, filtered)
if err != nil {
return []*api.Node{}, FailedPredicateMap{}, err
}
filtered = filteredList
if len(filtered) == 0 {
break
}
}
}
return filtered, failedPredicateMap, nil
}
func createPod(client clientset.Interface, namespace string, podCount int, podTemplate *api.Pod) error {
var createError error
lock := sync.Mutex{}
createPodFunc := func(i int) {
if err := makeCreatePod(client, namespace, podTemplate); err != nil {
lock.Lock()
defer lock.Unlock()
createError = err
}
}
if podCount < 30 {
workqueue.Parallelize(podCount, podCount, createPodFunc)
} else {
workqueue.Parallelize(30, podCount, createPodFunc)
}
return createError
}
// Filters the nodes to find the ones that fit based on the given predicate functions
// Each node is passed through the predicate functions to determine if it is a fit
func findNodesThatFit(pod *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, predicateFuncs map[string]algorithm.FitPredicate, nodes api.NodeList, extenders []algorithm.SchedulerExtender) (api.NodeList, FailedPredicateMap, error) {
// Create filtered list with enough space to avoid growing it.
filtered := make([]api.Node, 0, len(nodes.Items))
failedPredicateMap := FailedPredicateMap{}
if len(predicateFuncs) == 0 {
filtered = nodes.Items
} else {
predicateResultLock := sync.Mutex{}
errs := []error{}
meta := predicates.PredicateMetadata(pod)
checkNode := func(i int) {
nodeName := nodes.Items[i].Name
fits, failedPredicate, err := podFitsOnNode(pod, meta, nodeNameToInfo[nodeName], predicateFuncs)
predicateResultLock.Lock()
defer predicateResultLock.Unlock()
if err != nil {
errs = append(errs, err)
return
}
if fits {
filtered = append(filtered, nodes.Items[i])
} else {
failedPredicateMap[nodeName] = failedPredicate
}
}
workqueue.Parallelize(16, len(nodes.Items), checkNode)
if len(errs) > 0 {
return api.NodeList{}, FailedPredicateMap{}, errors.NewAggregate(errs)
}
}
if len(filtered) > 0 && len(extenders) != 0 {
for _, extender := range extenders {
filteredList, err := extender.Filter(pod, &api.NodeList{Items: filtered})
if err != nil {
return api.NodeList{}, FailedPredicateMap{}, err
}
filtered = filteredList.Items
if len(filtered) == 0 {
break
}
}
}
return api.NodeList{Items: filtered}, failedPredicateMap, nil
}
// makePodsFromRC will create a ReplicationController object and
// a given number of pods (imitating the controller).
func makePodsFromRC(c client.Interface, name string, podCount int) {
rc := &api.ReplicationController{
ObjectMeta: api.ObjectMeta{
Name: name,
},
Spec: api.ReplicationControllerSpec{
Replicas: int32(podCount),
Selector: map[string]string{"name": name},
Template: &api.PodTemplateSpec{
ObjectMeta: api.ObjectMeta{
Labels: map[string]string{"name": name},
},
Spec: makePodSpec(),
},
},
}
if _, err := c.ReplicationControllers("default").Create(rc); err != nil {
glog.Fatalf("unexpected error: %v", err)
}
basePod := &api.Pod{
ObjectMeta: api.ObjectMeta{
GenerateName: "scheduler-test-pod-",
Labels: map[string]string{"name": name},
},
Spec: makePodSpec(),
}
createPod := func(i int) {
for {
if _, err := c.Pods("default").Create(basePod); err == nil {
break
}
}
}
workqueue.Parallelize(30, podCount, createPod)
}
framework.PrintLatencies(watchLag, "worst watch latencies")
framework.PrintLatencies(schedToWatchLag, "worst scheduled-to-end total latencies")
framework.PrintLatencies(e2eLag, "worst e2e total latencies")
// Test whether e2e pod startup time is acceptable.
podStartupLatency := framework.PodStartupLatency{Latency: framework.ExtractLatencyMetrics(e2eLag)}
framework.ExpectNoError(framework.VerifyPodStartupLatency(podStartupLatency))
framework.LogSuspiciousLatency(startupLag, e2eLag, nodeCount, c)
By("Removing additional replication controllers")
deleteRC := func(i int) {
name := additionalPodsPrefix + "-" + strconv.Itoa(i+1)
framework.ExpectNoError(framework.DeleteRCAndWaitForGC(c, ns, name))
}
workqueue.Parallelize(16, nodeCount, deleteRC)
}
cleanupDensityTest(dConfig)
})
}
// Calculate total number of pods from each node's max-pod
It("[Feature:ManualPerformance] should allow running maximum capacity pods on nodes", func() {
totalPods = 0
for _, n := range nodes.Items {
totalPods += int(n.Status.Capacity.Pods().Value())
}
totalPods -= framework.WaitForStableCluster(c, masters)
fileHndl, err := os.Create(fmt.Sprintf(framework.TestContext.OutputDir+"/%s/pod_states.csv", uuid))
// compute a sum by iterating through the elements of weightedPodAffinityTerm and adding
// "weight" to the sum if the corresponding PodAffinityTerm is satisfied for
// that node; the node(s) with the highest sum are the most preferred.
// Symmetry need to be considered for preferredDuringSchedulingIgnoredDuringExecution from podAffinity & podAntiAffinity,
// symmetry need to be considered for hard requirements from podAffinity
func (ipa *InterPodAffinity) CalculateInterPodAffinityPriority(pod *v1.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodes []*v1.Node) (schedulerapi.HostPriorityList, error) {
affinity := pod.Spec.Affinity
hasAffinityConstraints := affinity != nil && affinity.PodAffinity != nil
hasAntiAffinityConstraints := affinity != nil && affinity.PodAntiAffinity != nil
allNodeNames := make([]string, 0, len(nodeNameToInfo))
for name := range nodeNameToInfo {
allNodeNames = append(allNodeNames, name)
}
// convert the topology key based weights to the node name based weights
var maxCount float64
var minCount float64
// priorityMap stores the mapping from node name to so-far computed score of
// the node.
pm := newPodAffinityPriorityMap(nodes, ipa.failureDomains)
processPod := func(existingPod *v1.Pod) error {
existingPodNode, err := ipa.info.GetNodeInfo(existingPod.Spec.NodeName)
if err != nil {
return err
}
existingPodAffinity := existingPod.Spec.Affinity
existingHasAffinityConstraints := existingPodAffinity != nil && existingPodAffinity.PodAffinity != nil
existingHasAntiAffinityConstraints := existingPodAffinity != nil && existingPodAffinity.PodAntiAffinity != nil
if hasAffinityConstraints {
// For every soft pod affinity term of <pod>, if <existingPod> matches the term,
// increment <pm.counts> for every node in the cluster with the same <term.TopologyKey>
// value as that of <existingPods>`s node by the term`s weight.
terms := affinity.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution
pm.processTerms(terms, pod, existingPod, existingPodNode, 1)
}
if hasAntiAffinityConstraints {
// For every soft pod anti-affinity term of <pod>, if <existingPod> matches the term,
// decrement <pm.counts> for every node in the cluster with the same <term.TopologyKey>
// value as that of <existingPod>`s node by the term`s weight.
terms := affinity.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution
pm.processTerms(terms, pod, existingPod, existingPodNode, -1)
}
if existingHasAffinityConstraints {
// For every hard pod affinity term of <existingPod>, if <pod> matches the term,
// increment <pm.counts> for every node in the cluster with the same <term.TopologyKey>
// value as that of <existingPod>'s node by the constant <ipa.hardPodAffinityWeight>
if ipa.hardPodAffinityWeight > 0 {
terms := existingPodAffinity.PodAffinity.RequiredDuringSchedulingIgnoredDuringExecution
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(existingPodAffinity.PodAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// terms = append(terms, existingPodAffinity.PodAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, term := range terms {
pm.processTerm(&term, existingPod, pod, existingPodNode, float64(ipa.hardPodAffinityWeight))
}
}
// For every soft pod affinity term of <existingPod>, if <pod> matches the term,
// increment <pm.counts> for every node in the cluster with the same <term.TopologyKey>
// value as that of <existingPod>'s node by the term's weight.
terms := existingPodAffinity.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution
pm.processTerms(terms, existingPod, pod, existingPodNode, 1)
}
if existingHasAntiAffinityConstraints {
// For every soft pod anti-affinity term of <existingPod>, if <pod> matches the term,
// decrement <pm.counts> for every node in the cluster with the same <term.TopologyKey>
// value as that of <existingPod>'s node by the term's weight.
terms := existingPodAffinity.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution
pm.processTerms(terms, existingPod, pod, existingPodNode, -1)
}
return nil
}
processNode := func(i int) {
nodeInfo := nodeNameToInfo[allNodeNames[i]]
if hasAffinityConstraints || hasAntiAffinityConstraints {
// We need to process all the nodes.
for _, existingPod := range nodeInfo.Pods() {
if err := processPod(existingPod); err != nil {
pm.setError(err)
}
}
} else {
// The pod doesn't have any constraints - we need to check only existing
// ones that have some.
for _, existingPod := range nodeInfo.PodsWithAffinity() {
if err := processPod(existingPod); err != nil {
pm.setError(err)
}
}
}
}
workqueue.Parallelize(16, len(allNodeNames), processNode)
if pm.firstError != nil {
return nil, pm.firstError
}
for _, node := range nodes {
if pm.counts[node.Name] > maxCount {
maxCount = pm.counts[node.Name]
}
if pm.counts[node.Name] < minCount {
minCount = pm.counts[node.Name]
}
}
// calculate final priority score for each node
result := make(schedulerapi.HostPriorityList, 0, len(nodes))
for _, node := range nodes {
fScore := float64(0)
if (maxCount - minCount) > 0 {
fScore = 10 * ((pm.counts[node.Name] - minCount) / (maxCount - minCount))
}
result = append(result, schedulerapi.HostPriority{Host: node.Name, Score: int(fScore)})
if glog.V(10) {
// We explicitly don't do glog.V(10).Infof() to avoid computing all the parameters if this is
// not logged. There is visible performance gain from it.
glog.V(10).Infof("%v -> %v: InterPodAffinityPriority, Score: (%d)", pod.Name, node.Name, int(fScore))
}
}
return result, nil
}
// CalculateSpreadPriority spreads pods across hosts and zones, considering pods belonging to the same service or replication controller.
// When a pod is scheduled, it looks for services or RCs that match the pod, then finds existing pods that match those selectors.
// It favors nodes that have fewer existing matching pods.
// i.e. it pushes the scheduler towards a node where there's the smallest number of
// pods which match the same service selectors or RC selectors as the pod being scheduled.
// Where zone information is included on the nodes, it favors nodes in zones with fewer existing matching pods.
func (s *SelectorSpread) CalculateSpreadPriority(pod *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodes []*api.Node) (schedulerapi.HostPriorityList, error) {
selectors := make([]labels.Selector, 0, 3)
if services, err := s.serviceLister.GetPodServices(pod); err == nil {
for _, service := range services {
selectors = append(selectors, labels.SelectorFromSet(service.Spec.Selector))
}
}
if rcs, err := s.controllerLister.GetPodControllers(pod); err == nil {
for _, rc := range rcs {
selectors = append(selectors, labels.SelectorFromSet(rc.Spec.Selector))
}
}
if rss, err := s.replicaSetLister.GetPodReplicaSets(pod); err == nil {
for _, rs := range rss {
if selector, err := unversioned.LabelSelectorAsSelector(rs.Spec.Selector); err == nil {
selectors = append(selectors, selector)
}
}
}
// Count similar pods by node
countsByNodeName := make(map[string]float32, len(nodes))
countsByZone := make(map[string]float32, 10)
maxCountByNodeName := float32(0)
countsByNodeNameLock := sync.Mutex{}
if len(selectors) > 0 {
processNodeFunc := func(i int) {
nodeName := nodes[i].Name
count := float32(0)
for _, nodePod := range nodeNameToInfo[nodeName].Pods() {
if pod.Namespace != nodePod.Namespace {
continue
}
// When we are replacing a failed pod, we often see the previous
// deleted version while scheduling the replacement.
// Ignore the previous deleted version for spreading purposes
// (it can still be considered for resource restrictions etc.)
if nodePod.DeletionTimestamp != nil {
glog.V(4).Infof("skipping pending-deleted pod: %s/%s", nodePod.Namespace, nodePod.Name)
continue
}
matches := false
for _, selector := range selectors {
if selector.Matches(labels.Set(nodePod.ObjectMeta.Labels)) {
matches = true
break
}
}
if matches {
count++
}
}
zoneId := utilnode.GetZoneKey(nodes[i])
countsByNodeNameLock.Lock()
defer countsByNodeNameLock.Unlock()
countsByNodeName[nodeName] = count
if count > maxCountByNodeName {
maxCountByNodeName = count
}
if zoneId != "" {
countsByZone[zoneId] += count
}
}
workqueue.Parallelize(16, len(nodes), processNodeFunc)
}
// Aggregate by-zone information
// Compute the maximum number of pods hosted in any zone
haveZones := len(countsByZone) != 0
maxCountByZone := float32(0)
for _, count := range countsByZone {
if count > maxCountByZone {
maxCountByZone = count
}
}
result := make(schedulerapi.HostPriorityList, 0, len(nodes))
//score int - scale of 0-maxPriority
// 0 being the lowest priority and maxPriority being the highest
for _, node := range nodes {
// initializing to the default/max node score of maxPriority
fScore := maxPriority
if maxCountByNodeName > 0 {
fScore = maxPriority * ((maxCountByNodeName - countsByNodeName[node.Name]) / maxCountByNodeName)
}
// If there is zone information present, incorporate it
if haveZones {
zoneId := utilnode.GetZoneKey(node)
if zoneId != "" {
zoneScore := maxPriority * ((maxCountByZone - countsByZone[zoneId]) / maxCountByZone)
fScore = (fScore * (1.0 - zoneWeighting)) + (zoneWeighting * zoneScore)
}
}
result = append(result, schedulerapi.HostPriority{Host: node.Name, Score: int(fScore)})
if glog.V(10) {
// We explicitly don't do glog.V(10).Infof() to avoid computing all the parameters if this is
// not logged. There is visible performance gain from it.
glog.V(10).Infof(
"%v -> %v: SelectorSpreadPriority, Score: (%d)", pod.Name, node.Name, int(fScore),
)
}
}
return result, nil
}
// Prioritizes the nodes by running the individual priority functions in parallel.
// Each priority function is expected to set a score of 0-10
// 0 is the lowest priority score (least preferred node) and 10 is the highest
// Each priority function can also have its own weight
// The node scores returned by the priority function are multiplied by the weights to get weighted scores
// All scores are finally combined (added) to get the total weighted scores of all nodes
func PrioritizeNodes(
pod *api.Pod,
nodeNameToInfo map[string]*schedulercache.NodeInfo,
meta interface{},
priorityConfigs []algorithm.PriorityConfig,
nodes []*api.Node,
extenders []algorithm.SchedulerExtender,
) (schedulerapi.HostPriorityList, error) {
// If no priority configs are provided, then the EqualPriority function is applied
// This is required to generate the priority list in the required format
if len(priorityConfigs) == 0 && len(extenders) == 0 {
return EqualPriority(pod, nodeNameToInfo, nodes)
}
var (
mu = sync.Mutex{}
wg = sync.WaitGroup{}
errs []error
)
appendError := func(err error) {
mu.Lock()
defer mu.Unlock()
errs = append(errs, err)
}
results := make([]schedulerapi.HostPriorityList, 0, len(priorityConfigs))
for range priorityConfigs {
results = append(results, nil)
}
for i, priorityConfig := range priorityConfigs {
if priorityConfig.Function != nil {
// DEPRECATED
wg.Add(1)
go func(index int, config algorithm.PriorityConfig) {
defer wg.Done()
var err error
results[index], err = config.Function(pod, nodeNameToInfo, nodes)
if err != nil {
appendError(err)
}
}(i, priorityConfig)
} else {
results[i] = make(schedulerapi.HostPriorityList, len(nodes))
}
}
processNode := func(index int) {
nodeInfo := nodeNameToInfo[nodes[index].Name]
var err error
for i := range priorityConfigs {
if priorityConfigs[i].Function != nil {
continue
}
results[i][index], err = priorityConfigs[i].Map(pod, meta, nodeInfo)
if err != nil {
appendError(err)
return
}
}
}
workqueue.Parallelize(16, len(nodes), processNode)
for i, priorityConfig := range priorityConfigs {
if priorityConfig.Reduce == nil {
continue
}
wg.Add(1)
go func(index int, config algorithm.PriorityConfig) {
defer wg.Done()
if err := config.Reduce(pod, results[index]); err != nil {
appendError(err)
}
}(i, priorityConfig)
}
// Wait for all computations to be finished.
wg.Wait()
if len(errs) != 0 {
return schedulerapi.HostPriorityList{}, errors.NewAggregate(errs)
}
// Summarize all scores.
result := make(schedulerapi.HostPriorityList, 0, len(nodes))
// TODO: Consider parallelizing it.
for i := range nodes {
result = append(result, schedulerapi.HostPriority{Host: nodes[i].Name, Score: 0})
for j := range priorityConfigs {
result[i].Score += results[j][i].Score * priorityConfigs[j].Weight
}
}
if len(extenders) != 0 && nodes != nil {
combinedScores := make(map[string]int, len(nodeNameToInfo))
for _, extender := range extenders {
wg.Add(1)
go func(ext algorithm.SchedulerExtender) {
defer wg.Done()
prioritizedList, weight, err := ext.Prioritize(pod, nodes)
if err != nil {
// Prioritization errors from extender can be ignored, let k8s/other extenders determine the priorities
return
}
mu.Lock()
for i := range *prioritizedList {
host, score := (*prioritizedList)[i].Host, (*prioritizedList)[i].Score
combinedScores[host] += score * weight
}
mu.Unlock()
}(extender)
}
// wait for all go routines to finish
wg.Wait()
for i := range result {
result[i].Score += combinedScores[result[i].Host]
}
}
if glog.V(10) {
for i := range result {
glog.V(10).Infof("Host %s => Score %d", result[i].Host, result[i].Score)
}
}
return result, nil
}
