func hasPodAffinityConstraints(pod *api.Pod) bool {
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil || affinity == nil {
return false
}
return affinity.PodAffinity != nil || affinity.PodAntiAffinity != nil
}
func (c *PodAffinityChecker) getMatchingAntiAffinityTerms(pod *api.Pod, allPods []*api.Pod) ([]matchingPodAntiAffinityTerm, error) {
var result []matchingPodAntiAffinityTerm
for _, existingPod := range allPods {
affinity, err := api.GetAffinityFromPodAnnotations(existingPod.Annotations)
if err != nil {
return nil, err
}
if affinity != nil && affinity.PodAntiAffinity != nil {
existingPodNode, err := c.info.GetNodeInfo(existingPod.Spec.NodeName)
if err != nil {
return nil, err
}
for _, term := range getPodAntiAffinityTerms(affinity.PodAntiAffinity) {
match, err := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, existingPod, &term)
if err != nil {
return nil, err
}
if match {
result = append(result, matchingPodAntiAffinityTerm{term: &term, node: existingPodNode})
}
}
}
}
return result, nil
}
// Admit will deny any pod that defines AntiAffinity topology key other than metav1.LabelHostname i.e. "kubernetes.io/hostname"
// in requiredDuringSchedulingRequiredDuringExecution and requiredDuringSchedulingIgnoredDuringExecution.
func (p *plugin) Admit(attributes admission.Attributes) (err error) {
// Ignore all calls to subresources or resources other than pods.
if len(attributes.GetSubresource()) != 0 || attributes.GetResource().GroupResource() != api.Resource("pods") {
return nil
}
pod, ok := attributes.GetObject().(*api.Pod)
if !ok {
return apierrors.NewBadRequest("Resource was marked with kind Pod but was unable to be converted")
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
glog.V(5).Infof("Invalid Affinity detected, but we will leave handling of this to validation phase")
return nil
}
if affinity != nil && affinity.PodAntiAffinity != nil {
var podAntiAffinityTerms []api.PodAffinityTerm
if len(affinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
podAntiAffinityTerms = affinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(affinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// podAntiAffinityTerms = append(podAntiAffinityTerms, affinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, v := range podAntiAffinityTerms {
if v.TopologyKey != metav1.LabelHostname {
return apierrors.NewForbidden(attributes.GetResource().GroupResource(), pod.Name, fmt.Errorf("affinity.PodAntiAffinity.RequiredDuringScheduling has TopologyKey %v but only key %v is allowed", v.TopologyKey, metav1.LabelHostname))
}
}
}
return nil
}
func (checker *PodAffinityChecker) InterPodAffinityMatches(pod *api.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (bool, error) {
node := nodeInfo.Node()
if node == nil {
return false, fmt.Errorf("node not found")
}
allPods, err := checker.podLister.List(labels.Everything())
if err != nil {
return false, err
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return false, err
}
// Check if the current node match the inter-pod affinity scheduling constraints.
// Hard inter-pod affinity is not symmetric, check only when affinity.PodAffinity exists.
if affinity.PodAffinity != nil {
if !checker.NodeMatchesHardPodAffinity(pod, allPods, node, affinity.PodAffinity) {
return false, ErrPodAffinityNotMatch
}
}
// Hard inter-pod anti-affinity is symmetric, we should always check it.
if !checker.NodeMatchesHardPodAntiAffinity(pod, allPods, node, affinity.PodAntiAffinity) {
return false, ErrPodAffinityNotMatch
}
return true, nil
}
// Admit will deny any pod that defines AntiAffinity topology key other than unversioned.LabelHostname i.e. "kubernetes.io/hostname"
// in requiredDuringSchedulingRequiredDuringExecution and requiredDuringSchedulingIgnoredDuringExecution.
func (p *plugin) Admit(attributes admission.Attributes) (err error) {
if attributes.GetResource().GroupResource() != api.Resource("pods") {
return nil
}
pod, ok := attributes.GetObject().(*api.Pod)
if !ok {
return apierrors.NewBadRequest("Resource was marked with kind Pod but was unable to be converted")
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
// this is validated later
return nil
}
if affinity.PodAntiAffinity != nil {
var podAntiAffinityTerms []api.PodAffinityTerm
if len(affinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
podAntiAffinityTerms = affinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(affinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// podAntiAffinityTerms = append(podAntiAffinityTerms, affinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, v := range podAntiAffinityTerms {
if v.TopologyKey != unversioned.LabelHostname {
return apierrors.NewForbidden(attributes.GetResource().GroupResource(), pod.Name, fmt.Errorf("affinity.PodAntiAffinity.RequiredDuringScheduling has TopologyKey %v but only key %v is allowed", v.TopologyKey, unversioned.LabelHostname))
}
}
}
return nil
}
func (c *PodAffinityChecker) InterPodAffinityMatches(pod *api.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (bool, []algorithm.PredicateFailureReason, error) {
node := nodeInfo.Node()
if node == nil {
return false, nil, fmt.Errorf("node not found")
}
if !c.satisfiesExistingPodsAntiAffinity(pod, meta, node) {
return false, []algorithm.PredicateFailureReason{ErrPodAffinityNotMatch}, nil
}
// Now check if <pod> requirements will be satisfied on this node.
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return false, nil, err
}
if affinity == nil || (affinity.PodAffinity == nil && affinity.PodAntiAffinity == nil) {
return true, nil, nil
}
if !c.satisfiesPodsAffinityAntiAffinity(pod, node, affinity) {
return false, []algorithm.PredicateFailureReason{ErrPodAffinityNotMatch}, nil
}
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.Infof("Schedule Pod %+v on Node %+v is allowed, pod (anti)affinity constraints satisfied",
podName(pod), node.Name)
}
return true, nil, nil
}
// CalculateNodeAffinityPriority prioritizes nodes according to node affinity scheduling preferences
// indicated in PreferredDuringSchedulingIgnoredDuringExecution. Each time a node match a preferredSchedulingTerm,
// it will a get an add of preferredSchedulingTerm.Weight. Thus, the more preferredSchedulingTerms
// the node satisfies and the more the preferredSchedulingTerm that is satisfied weights, the higher
// score the node gets.
func CalculateNodeAffinityPriority(pod *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodeLister algorithm.NodeLister) (schedulerapi.HostPriorityList, error) {
nodes, err := nodeLister.List()
if err != nil {
return nil, err
}
var maxCount float64
counts := make(map[string]float64, len(nodes.Items))
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil, err
}
// A nil element of PreferredDuringSchedulingIgnoredDuringExecution matches no objects.
// An element of PreferredDuringSchedulingIgnoredDuringExecution that refers to an
// empty PreferredSchedulingTerm matches all objects.
if affinity.NodeAffinity != nil && affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution != nil {
// Match PreferredDuringSchedulingIgnoredDuringExecution term by term.
for _, preferredSchedulingTerm := range affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
if preferredSchedulingTerm.Weight == 0 {
continue
}
nodeSelector, err := api.NodeSelectorRequirementsAsSelector(preferredSchedulingTerm.Preference.MatchExpressions)
if err != nil {
return nil, err
}
for _, node := range nodes.Items {
if nodeSelector.Matches(labels.Set(node.Labels)) {
counts[node.Name] += float64(preferredSchedulingTerm.Weight)
}
if counts[node.Name] > maxCount {
maxCount = counts[node.Name]
}
}
}
}
result := make(schedulerapi.HostPriorityList, 0, len(nodes.Items))
for i := range nodes.Items {
node := &nodes.Items[i]
if maxCount > 0 {
fScore := 10 * (counts[node.Name] / maxCount)
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.Infof("%v -> %v: NodeAffinityPriority, Score: (%d)", pod.Name, node.Name, int(fScore))
}
} else {
result = append(result, schedulerapi.HostPriority{Host: node.Name, Score: 0})
}
}
return result, nil
}
func getMatchingAntiAffinityTerms(pod *api.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 := api.GetAffinityFromPodAnnotations(existingPod.Annotations)
if err != nil {
catchError(err)
return
}
if affinity == nil {
continue
}
for _, term := range getPodAntiAffinityTerms(affinity.PodAntiAffinity) {
match, err := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, existingPod, &term)
if err != nil {
catchError(err)
return
}
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
}
// CalculateNodeAffinityPriority prioritizes nodes according to node affinity scheduling preferences
// indicated in PreferredDuringSchedulingIgnoredDuringExecution. Each time a node match a preferredSchedulingTerm,
// it will a get an add of preferredSchedulingTerm.Weight. Thus, the more preferredSchedulingTerms
// the node satisfies and the more the preferredSchedulingTerm that is satisfied weights, the higher
// score the node gets.
func (s *NodeAffinity) CalculateNodeAffinityPriority(pod *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodeLister algorithm.NodeLister) (schedulerapi.HostPriorityList, error) {
var maxCount int
counts := map[string]int{}
nodes, err := nodeLister.List()
if err != nil {
return nil, err
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil, err
}
// A nil element of PreferredDuringSchedulingIgnoredDuringExecution matches no objects.
// An element of PreferredDuringSchedulingIgnoredDuringExecution that refers to an
// empty PreferredSchedulingTerm matches all objects.
if affinity.NodeAffinity != nil && affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution != nil {
// Match PreferredDuringSchedulingIgnoredDuringExecution term by term.
for _, preferredSchedulingTerm := range affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
if preferredSchedulingTerm.Weight == 0 {
continue
}
nodeSelector, err := api.NodeSelectorRequirementsAsSelector(preferredSchedulingTerm.Preference.MatchExpressions)
if err != nil {
return nil, err
}
for _, node := range nodes.Items {
if nodeSelector.Matches(labels.Set(node.Labels)) {
counts[node.Name] += preferredSchedulingTerm.Weight
}
if counts[node.Name] > maxCount {
maxCount = counts[node.Name]
}
}
}
}
result := []schedulerapi.HostPriority{}
for _, node := range nodes.Items {
fScore := float64(0)
if maxCount > 0 {
fScore = 10 * (float64(counts[node.Name]) / float64(maxCount))
}
result = append(result, schedulerapi.HostPriority{Host: node.Name, Score: int(fScore)})
glog.V(10).Infof("%v -> %v: NodeAffinityPriority, Score: (%d)", pod.Name, node.Name, int(fScore))
}
return result, nil
}
// The pod can only schedule onto nodes that satisfy requirements in both NodeAffinity and nodeSelector.
func podMatchesNodeLabels(pod *api.Pod, node *api.Node) bool {
// Check if node.Labels match pod.Spec.NodeSelector.
if len(pod.Spec.NodeSelector) > 0 {
selector := labels.SelectorFromSet(pod.Spec.NodeSelector)
if !selector.Matches(labels.Set(node.Labels)) {
return false
}
}
// Parse required node affinity scheduling requirements
// and check if the current node match the requirements.
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
glog.V(10).Infof("Failed to get Affinity from Pod %+v, err: %+v", podName(pod), err)
return false
}
// 1. nil NodeSelector matches all nodes (i.e. does not filter out any nodes)
// 2. nil []NodeSelectorTerm (equivalent to non-nil empty NodeSelector) matches no nodes
// 3. zero-length non-nil []NodeSelectorTerm matches no nodes also, just for simplicity
// 4. nil []NodeSelectorRequirement (equivalent to non-nil empty NodeSelectorTerm) matches no nodes
// 5. zero-length non-nil []NodeSelectorRequirement matches no nodes also, just for simplicity
// 6. non-nil empty NodeSelectorRequirement is not allowed
nodeAffinityMatches := true
if affinity.NodeAffinity != nil {
nodeAffinity := affinity.NodeAffinity
// if no required NodeAffinity requirements, will do no-op, means select all nodes.
// TODO: Replace next line with subsequent commented-out line when implement RequiredDuringSchedulingRequiredDuringExecution.
if nodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution == nil {
// if nodeAffinity.RequiredDuringSchedulingRequiredDuringExecution == nil && nodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution == nil {
return true
}
// Match node selector for requiredDuringSchedulingRequiredDuringExecution.
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
// if nodeAffinity.RequiredDuringSchedulingRequiredDuringExecution != nil {
// nodeSelectorTerms := nodeAffinity.RequiredDuringSchedulingRequiredDuringExecution.NodeSelectorTerms
// glog.V(10).Infof("Match for RequiredDuringSchedulingRequiredDuringExecution node selector terms %+v", nodeSelectorTerms)
// nodeAffinityMatches = nodeMatchesNodeSelectorTerms(node, nodeSelectorTerms)
// }
// Match node selector for requiredDuringSchedulingIgnoredDuringExecution.
if nodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution != nil {
nodeSelectorTerms := nodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution.NodeSelectorTerms
glog.V(10).Infof("Match for RequiredDuringSchedulingIgnoredDuringExecution node selector terms %+v", nodeSelectorTerms)
nodeAffinityMatches = nodeAffinityMatches && nodeMatchesNodeSelectorTerms(node, nodeSelectorTerms)
}
}
return nodeAffinityMatches
}
// Checks whether the given node has pods which satisfy all the
// required pod anti-affinity scheduling rules.
// Also checks whether putting the pod onto the node would break
// any anti-affinity scheduling rules indicated by existing pods.
// If node has pods which satisfy all the required pod anti-affinity
// scheduling rules and scheduling the pod onto the node won't
// break any existing pods' anti-affinity rules, then return true.
func (checker *PodAffinityChecker) NodeMatchesHardPodAntiAffinity(pod *api.Pod, allPods []*api.Pod, node *api.Node, podAntiAffinity *api.PodAntiAffinity) bool {
// foreach element podAntiAffinityTerm of podAntiAffinityTerms
// if the pod matches the term (breaks the anti-affinity),
// don't schedule the pod onto this node.
for _, podAntiAffinityTerm := range getPodAntiAffinityTerms(podAntiAffinity) {
podAntiAffinityTermMatches, _, err := checker.AnyPodMatchesPodAffinityTerm(pod, allPods, node, podAntiAffinityTerm)
if err != nil || podAntiAffinityTermMatches {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v, because not all the existing pods on this node satisfy the PodAntiAffinityTerm %v, err: %v",
podName(pod), node.Name, podAntiAffinityTerm, err)
return false
}
}
// Check if scheduling the pod onto this node would break
// any anti-affinity rules indicated by the existing pods on the node.
// If it would break, system should not schedule pod onto this node.
for _, ep := range allPods {
epAffinity, err := api.GetAffinityFromPodAnnotations(ep.Annotations)
if err != nil {
glog.V(10).Infof("Failed to get Affinity from Pod %+v, err: %+v", podName(pod), err)
return false
}
if epAffinity == nil {
continue
}
epNode, err := checker.info.GetNodeInfo(ep.Spec.NodeName)
if err != nil {
glog.V(10).Infof("Failed to get node from Pod %+v, err: %+v", podName(ep), err)
return false
}
for _, epAntiAffinityTerm := range getPodAntiAffinityTerms(epAffinity.PodAntiAffinity) {
match, err := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, ep, &epAntiAffinityTerm)
if err != nil {
glog.V(10).Infof("Failed to get label selector from anti-affinityterm %+v of existing pod %+v, err: %+v", epAntiAffinityTerm, podName(pod), err)
return false
}
if match && checker.failureDomains.NodesHaveSameTopologyKey(node, epNode, epAntiAffinityTerm.TopologyKey) {
glog.V(10).Infof("Cannot schedule Pod %+v, onto node %v because the pod would break the PodAntiAffinityTerm %+v, of existing pod %+v, err: %v",
podName(pod), node.Name, epAntiAffinityTerm, podName(ep), err)
return false
}
}
}
// all the required pod anti-affinity scheduling rules are satisfied
glog.V(10).Infof("Can schedule Pod %+v, on node %v because all the required pod anti-affinity scheduling rules are satisfied", podName(pod), node.Name)
return true
}
// CalculateNodeAffinityPriority prioritizes nodes according to node affinity scheduling preferences
// indicated in PreferredDuringSchedulingIgnoredDuringExecution. Each time a node match a preferredSchedulingTerm,
// it will a get an add of preferredSchedulingTerm.Weight. Thus, the more preferredSchedulingTerms
// the node satisfies and the more the preferredSchedulingTerm that is satisfied weights, the higher
// score the node gets.
func CalculateNodeAffinityPriorityMap(pod *api.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (schedulerapi.HostPriority, error) {
node := nodeInfo.Node()
if node == nil {
return schedulerapi.HostPriority{}, fmt.Errorf("node not found")
}
var affinity *api.Affinity
if priorityMeta, ok := meta.(*priorityMetadata); ok {
affinity = priorityMeta.affinity
} else {
// We couldn't parse metadata - fallback to computing it.
var err error
affinity, err = api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return schedulerapi.HostPriority{}, err
}
}
var count int32
// A nil element of PreferredDuringSchedulingIgnoredDuringExecution matches no objects.
// An element of PreferredDuringSchedulingIgnoredDuringExecution that refers to an
// empty PreferredSchedulingTerm matches all objects.
if affinity != nil && affinity.NodeAffinity != nil && affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution != nil {
// Match PreferredDuringSchedulingIgnoredDuringExecution term by term.
for i := range affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
preferredSchedulingTerm := &affinity.NodeAffinity.PreferredDuringSchedulingIgnoredDuringExecution[i]
if preferredSchedulingTerm.Weight == 0 {
continue
}
// TODO: Avoid computing it for all nodes if this becomes a performance problem.
nodeSelector, err := api.NodeSelectorRequirementsAsSelector(preferredSchedulingTerm.Preference.MatchExpressions)
if err != nil {
return schedulerapi.HostPriority{}, err
}
if nodeSelector.Matches(labels.Set(node.Labels)) {
count += preferredSchedulingTerm.Weight
}
}
}
return schedulerapi.HostPriority{
Host: node.Name,
Score: int(count),
}, nil
}
// NodeMatchPodAffinityAntiAffinity checks if the node matches
// the requiredDuringScheduling affinity/anti-affinity rules indicated by the pod.
func (checker *PodAffinityChecker) NodeMatchPodAffinityAntiAffinity(pod *api.Pod, allPods []*api.Pod, node *api.Node) bool {
// Parse required affinity scheduling rules.
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
glog.V(10).Infof("Failed to get Affinity from Pod %+v, err: %+v", podName(pod), err)
return false
}
// check if the current node match the inter-pod affinity scheduling rules.
// hard inter-pod affinity is not symmetric, check only when affinity.PodAffinity is not nil.
if affinity.PodAffinity != nil {
if !checker.NodeMatchesHardPodAffinity(pod, allPods, node, affinity.PodAffinity) {
return false
}
}
// hard inter-pod anti-affinity is symmetric, check both when affinity.PodAntiAffinity is nil and not nil.
return checker.NodeMatchesHardPodAntiAffinity(pod, allPods, node, affinity.PodAntiAffinity)
}
func PriorityMetadata(pod *api.Pod) interface{} {
// If we cannot compute metadata, just return nil
if pod == nil {
return nil
}
tolerations, err := getTolerationListFromPod(pod)
if err != nil {
return nil
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil
}
return &priorityMetadata{
nonZeroRequest: getNonZeroRequests(pod),
podTolerations: tolerations,
affinity: affinity,
}
}
// Checks whether the given node has pods which satisfy all the
// required pod anti-affinity scheduling rules.
// Also checks whether putting the pod onto the node would break
// any anti-affinity scheduling rules indicated by existing pods.
// If node has pods which satisfy all the required pod anti-affinity
// scheduling rules and scheduling the pod onto the node won't
// break any existing pods' anti-affinity rules, then return true.
func (checker *PodAffinityChecker) NodeMatchesHardPodAntiAffinity(pod *api.Pod, allPods []*api.Pod, node *api.Node, podAntiAffinity *api.PodAntiAffinity) bool {
var podAntiAffinityTerms []api.PodAffinityTerm
if len(podAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
podAntiAffinityTerms = podAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(podAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// podAntiAffinityTerms = append(podAntiAffinityTerms, podAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
// foreach element podAntiAffinityTerm of podAntiAffinityTerms
// if the pod matches the term (breaks the anti-affinity),
// don't schedule the pod onto this node.
for _, podAntiAffinityTerm := range podAntiAffinityTerms {
podAntiAffinityTermMatches, err := checker.AnyPodMatchesPodAffinityTerm(pod, allPods, node, podAntiAffinityTerm)
if err != nil || podAntiAffinityTermMatches == true {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v, because not all the existing pods on this node satisfy the PodAntiAffinityTerm %v, err: %v",
podName(pod), node.Name, podAntiAffinityTerm, err)
return false
}
}
// Check if scheduling the pod onto this node would break
// any anti-affinity rules indicated by the existing pods on the node.
// If it would break, system should not schedule pod onto this node.
for _, ep := range allPods {
epAffinity, err := api.GetAffinityFromPodAnnotations(ep.Annotations)
if err != nil {
glog.V(10).Infof("Failed to get Affinity from Pod %+v, err: %+v", podName(pod), err)
return false
}
if epAffinity.PodAntiAffinity != nil {
var epAntiAffinityTerms []api.PodAffinityTerm
if len(epAffinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
epAntiAffinityTerms = epAffinity.PodAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(epAffinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// epAntiAffinityTerms = append(epAntiAffinityTerms, epAffinity.PodAntiAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, epAntiAffinityTerm := range epAntiAffinityTerms {
labelSelector, err := unversioned.LabelSelectorAsSelector(epAntiAffinityTerm.LabelSelector)
if err != nil {
glog.V(10).Infof("Failed to get label selector from anti-affinityterm %+v of existing pod %+v, err: %+v", epAntiAffinityTerm, podName(pod), err)
return false
}
names := priorityutil.GetNamespacesFromPodAffinityTerm(ep, epAntiAffinityTerm)
if (len(names) == 0 || names.Has(pod.Namespace)) && labelSelector.Matches(labels.Set(pod.Labels)) {
epNode, err := checker.info.GetNodeInfo(ep.Spec.NodeName)
if err != nil || checker.failureDomains.NodesHaveSameTopologyKey(node, epNode, epAntiAffinityTerm.TopologyKey) {
glog.V(10).Infof("Cannot schedule Pod %+v, onto node %v because the pod would break the PodAntiAffinityTerm %+v, of existing pod %+v, err: %v",
podName(pod), node.Name, epAntiAffinityTerm, podName(ep), err)
return false
}
}
}
}
}
// all the required pod anti-affinity scheduling rules are satisfied
glog.V(10).Infof("Can schedule Pod %+v, on node %v because all the required pod anti-affinity scheduling rules are satisfied", podName(pod), node.Name)
return true
}
// 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 *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodeLister algorithm.NodeLister) (schedulerapi.HostPriorityList, error) {
nodes, err := nodeLister.List()
if err != nil {
return nil, err
}
allPods, err := ipa.podLister.List(labels.Everything())
if err != nil {
return nil, err
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil, err
}
// convert the topology key based weights to the node name based weights
var maxCount int
var minCount int
counts := map[string]int{}
for _, node := range nodes.Items {
totalCount := 0
// count weights for the weighted pod affinity
if affinity.PodAffinity != nil {
for _, weightedTerm := range affinity.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
weightedCount, err := ipa.CountWeightByPodMatchAffinityTerm(pod, allPods, weightedTerm.Weight, weightedTerm.PodAffinityTerm, &node)
if err != nil {
return nil, err
}
totalCount += weightedCount
}
}
// count weights for the weighted pod anti-affinity
if affinity.PodAntiAffinity != nil {
for _, weightedTerm := range affinity.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
weightedCount, err := ipa.CountWeightByPodMatchAffinityTerm(pod, allPods, (0 - weightedTerm.Weight), weightedTerm.PodAffinityTerm, &node)
if err != nil {
return nil, err
}
totalCount += weightedCount
}
}
// reverse direction checking: count weights for the inter-pod affinity/anti-affinity rules
// that are indicated by existing pods on the node.
for _, ep := range allPods {
epAffinity, err := api.GetAffinityFromPodAnnotations(ep.Annotations)
if err != nil {
return nil, err
}
if epAffinity.PodAffinity != nil {
// count the implicit weight for the hard pod affinity indicated by the existing pod.
if ipa.hardPodAffinityWeight > 0 {
var podAffinityTerms []api.PodAffinityTerm
if len(epAffinity.PodAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
podAffinityTerms = epAffinity.PodAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(affinity.PodAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// podAffinityTerms = append(podAffinityTerms, affinity.PodAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, epAffinityTerm := range podAffinityTerms {
match, err := ipa.failureDomains.CheckIfPodMatchPodAffinityTerm(pod, ep, epAffinityTerm,
func(pod *api.Pod) (*api.Node, error) { return &node, nil },
func(ep *api.Pod) (*api.Node, error) { return ipa.info.GetNodeInfo(ep.Spec.NodeName) },
)
if err != nil {
return nil, err
}
if match {
totalCount += ipa.hardPodAffinityWeight
}
}
}
// count weight for the weighted pod affinity indicated by the existing pod.
for _, epWeightedTerm := range epAffinity.PodAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
match, err := ipa.failureDomains.CheckIfPodMatchPodAffinityTerm(pod, ep, epWeightedTerm.PodAffinityTerm,
func(pod *api.Pod) (*api.Node, error) { return &node, nil },
func(ep *api.Pod) (*api.Node, error) { return ipa.info.GetNodeInfo(ep.Spec.NodeName) },
)
if err != nil {
return nil, err
}
if match {
totalCount += epWeightedTerm.Weight
}
}
}
// count weight for the weighted pod anti-affinity indicated by the existing pod.
if epAffinity.PodAntiAffinity != nil {
for _, epWeightedTerm := range epAffinity.PodAntiAffinity.PreferredDuringSchedulingIgnoredDuringExecution {
match, err := ipa.failureDomains.CheckIfPodMatchPodAffinityTerm(pod, ep, epWeightedTerm.PodAffinityTerm,
func(pod *api.Pod) (*api.Node, error) { return &node, nil },
func(ep *api.Pod) (*api.Node, error) { return ipa.info.GetNodeInfo(ep.Spec.NodeName) },
)
if err != nil {
return nil, err
}
if match {
totalCount -= epWeightedTerm.Weight
}
}
}
}
counts[node.Name] = totalCount
if counts[node.Name] > maxCount {
maxCount = counts[node.Name]
}
if counts[node.Name] < minCount {
minCount = counts[node.Name]
}
}
// calculate final priority score for each node
result := []schedulerapi.HostPriority{}
for _, node := range nodes.Items {
fScore := float64(0)
if (maxCount - minCount) > 0 {
fScore = 10 * (float64(counts[node.Name]-minCount) / float64(maxCount-minCount))
}
result = append(result, schedulerapi.HostPriority{Host: node.Name, Score: int(fScore)})
glog.V(10).Infof(
"%v -> %v: InterPodAffinityPriority, Score: (%d)", pod.Name, node.Name, int(fScore),
)
}
return result, nil
}
// 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 *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodes []*api.Node) (schedulerapi.HostPriorityList, error) {
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil, err
}
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 *api.Pod) error {
existingPodNode, err := ipa.info.GetNodeInfo(existingPod.Spec.NodeName)
if err != nil {
return err
}
existingPodAffinity, err := api.GetAffinityFromPodAnnotations(existingPod.Annotations)
if err != nil {
return err
}
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
}
// 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 *api.Pod, nodeNameToInfo map[string]*schedulercache.NodeInfo, nodes []*api.Node) (schedulerapi.HostPriorityList, error) {
allPods, err := ipa.podLister.List(labels.Everything())
if err != nil {
return nil, err
}
affinity, err := api.GetAffinityFromPodAnnotations(pod.Annotations)
if err != nil {
return nil, err
}
// convert the topology key based weights to the node name based weights
var maxCount float64
var minCount float64
// counts store the mapping from node name to so-far computed score of
// the node.
counts := make(map[string]float64, len(nodes))
processTerm := func(term *api.PodAffinityTerm, affinityPod, podToCheck *api.Pod, fixedNode *api.Node, weight float64) error {
match, err := podMatchesNamespaceAndSelector(podToCheck, affinityPod, term)
if err != nil {
return err
}
if match {
for _, node := range nodes {
if ipa.failureDomains.NodesHaveSameTopologyKey(node, fixedNode, term.TopologyKey) {
counts[node.Name] += weight
}
}
}
return nil
}
processTerms := func(terms []api.WeightedPodAffinityTerm, affinityPod, podToCheck *api.Pod, fixedNode *api.Node, multiplier int) error {
for _, weightedTerm := range terms {
if err := processTerm(&weightedTerm.PodAffinityTerm, affinityPod, podToCheck, fixedNode, float64(weightedTerm.Weight*multiplier)); err != nil {
return err
}
}
return nil
}
for _, existingPod := range allPods {
existingPodNode, err := ipa.info.GetNodeInfo(existingPod.Spec.NodeName)
if err != nil {
return nil, err
}
existingPodAffinity, err := api.GetAffinityFromPodAnnotations(existingPod.Annotations)
if err != nil {
return nil, err
}
if affinity.PodAffinity != nil {
// For every soft pod affinity term of <pod>, if <existingPod> matches the term,
// increment <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
if err := processTerms(terms, pod, existingPod, existingPodNode, 1); err != nil {
return nil, err
}
}
if affinity.PodAntiAffinity != nil {
// For every soft pod anti-affinity term of <pod>, if <existingPod> matches the term,
// decrement <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
if err := processTerms(terms, pod, existingPod, existingPodNode, -1); err != nil {
return nil, err
}
}
if existingPodAffinity.PodAffinity != nil {
// For every hard pod affinity term of <existingPod>, if <pod> matches the term,
// increment <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 {
if err := processTerm(&term, existingPod, pod, existingPodNode, float64(ipa.hardPodAffinityWeight)); err != nil {
return nil, err
}
}
}
// For every soft pod affinity term of <existingPod>, if <pod> matches the term,
// increment <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
if err := processTerms(terms, existingPod, pod, existingPodNode, 1); err != nil {
return nil, err
}
}
if existingPodAffinity.PodAntiAffinity != nil {
// For every soft pod anti-affinity term of <existingPod>, if <pod> matches the term,
// decrement <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
if err := processTerms(terms, existingPod, pod, existingPodNode, -1); err != nil {
return nil, err
}
}
}
for _, node := range nodes {
if counts[node.Name] > maxCount {
maxCount = counts[node.Name]
}
if counts[node.Name] < minCount {
minCount = 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 * ((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
}