func (c *PodAffinityChecker) getMatchingAntiAffinityTerms(pod *v1.Pod, allPods []*v1.Pod) ([]matchingPodAntiAffinityTerm, error) {
var result []matchingPodAntiAffinityTerm
for _, existingPod := range allPods {
affinity, err := v1.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) {
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(existingPod, &term)
selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
if err != nil {
return nil, err
}
match := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, namespaces, selector)
if match {
result = append(result, matchingPodAntiAffinityTerm{term: &term, node: existingPodNode})
}
}
}
}
return result, nil
}
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
}
// Checks if scheduling the pod onto this node would break any rules of this pod.
func (c *PodAffinityChecker) satisfiesPodsAffinityAntiAffinity(pod *v1.Pod, node *v1.Node, affinity *v1.Affinity) bool {
allPods, err := c.podLister.List(labels.Everything())
if err != nil {
return false
}
// Check all affinity terms.
for _, term := range getPodAffinityTerms(affinity.PodAffinity) {
termMatches, matchingPodExists, err := c.anyPodMatchesPodAffinityTerm(pod, allPods, node, &term)
if err != nil {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v,because of PodAffinityTerm %v, err: %v",
podName(pod), node.Name, term, err)
return false
}
if !termMatches {
// If the requirement matches a pod's own labels are namespace, and there are
// no other such pods, then disregard the requirement. This is necessary to
// not block forever because the first pod of the collection can't be scheduled.
if matchingPodExists {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v,because of PodAffinityTerm %v, err: %v",
podName(pod), node.Name, term, err)
return false
}
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(pod, &term)
selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
if err != nil {
glog.V(10).Infof("Cannot parse selector on term %v for pod %v. Details %v",
term, podName(pod), err)
return false
}
match := priorityutil.PodMatchesTermsNamespaceAndSelector(pod, namespaces, selector)
if !match {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v,because of PodAffinityTerm %v, err: %v",
podName(pod), node.Name, term, err)
return false
}
}
}
// Check all anti-affinity terms.
for _, term := range getPodAntiAffinityTerms(affinity.PodAntiAffinity) {
termMatches, _, err := c.anyPodMatchesPodAffinityTerm(pod, allPods, node, &term)
if err != nil || termMatches {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v,because of PodAntiAffinityTerm %v, err: %v",
podName(pod), node.Name, term, err)
return false
}
}
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 afinnity/anti-affinity constraints satisfied.",
podName(pod), node.Name)
}
return true
}
// TODO: Share it with predicates by moving to better location.
// TODO: Can we avoid error handling here - this is only a matter of non-parsable selector?
func podMatchesNamespaceAndSelector(pod *api.Pod, affinityPod *api.Pod, term *api.PodAffinityTerm) (bool, error) {
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(affinityPod, *term)
if len(namespaces) != 0 && !namespaces.Has(pod.Namespace) {
return false, nil
}
selector, err := unversioned.LabelSelectorAsSelector(term.LabelSelector)
if err != nil || !selector.Matches(labels.Set(pod.Labels)) {
return false, err
}
return true, nil
}
// Checks whether the given node has pods which satisfy all the required pod affinity scheduling rules.
// If node has pods which satisfy all the required pod affinity scheduling rules then return true.
func (checker *PodAffinityChecker) NodeMatchesHardPodAffinity(pod *api.Pod, allPods []*api.Pod, node *api.Node, podAffinity *api.PodAffinity) bool {
var podAffinityTerms []api.PodAffinityTerm
if len(podAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
podAffinityTerms = podAffinity.RequiredDuringSchedulingIgnoredDuringExecution
}
// TODO: Uncomment this block when implement RequiredDuringSchedulingRequiredDuringExecution.
//if len(podAffinity.RequiredDuringSchedulingRequiredDuringExecution) != 0 {
// podAffinityTerms = append(podAffinityTerms, podAffinity.RequiredDuringSchedulingRequiredDuringExecution...)
//}
for _, podAffinityTerm := range podAffinityTerms {
podAffinityTermMatches, err := checker.AnyPodMatchesPodAffinityTerm(pod, allPods, node, podAffinityTerm)
if err != nil {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v, an error ocurred when checking existing pods on the node for PodAffinityTerm %v err: %v",
podName(pod), node.Name, podAffinityTerm, err)
return false
}
if !podAffinityTermMatches {
// TODO: Think about whether this can be simplified once we have controllerRef
// Check if it is in special case that the requiredDuringScheduling affinity requirement can be disregarded.
// If the requiredDuringScheduling affinity requirement matches a pod's own labels and namespace, and there are no other such pods
// anywhere, then disregard the requirement.
// This allows rules like "schedule all of the pods of this collection to the same zone" to not block forever
// because the first pod of the collection can't be scheduled.
names := priorityutil.GetNamespacesFromPodAffinityTerm(pod, podAffinityTerm)
labelSelector, err := unversioned.LabelSelectorAsSelector(podAffinityTerm.LabelSelector)
if err != nil || !names.Has(pod.Namespace) || !labelSelector.Matches(labels.Set(pod.Labels)) {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v, because none of the existing pods on this node satisfy the PodAffinityTerm %v, err: %+v",
podName(pod), node.Name, podAffinityTerm, err)
return false
}
// the affinity is to put the pod together with other pods from its same service or controller
filteredPods := priorityutil.FilterPodsByNameSpaces(names, allPods)
for _, filteredPod := range filteredPods {
// if found an existing pod from same service or RC,
// the affinity scheduling rules cannot be disregarded.
if labelSelector.Matches(labels.Set(filteredPod.Labels)) {
glog.V(10).Infof("Cannot schedule pod %+v onto node %v, because none of the existing pods on this node satisfy the PodAffinityTerm %v",
podName(pod), node.Name, podAffinityTerm)
return false
}
}
}
}
// all the required pod affinity scheduling rules satisfied
glog.V(10).Infof("All the required pod affinity scheduling rules are satisfied for Pod %+v, on node %v", podName(pod), node.Name)
return true
}
func (p *podAffinityPriorityMap) processTerm(term *v1.PodAffinityTerm, podDefiningAffinityTerm, podToCheck *v1.Pod, fixedNode *v1.Node, weight float64) {
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(podDefiningAffinityTerm, term)
selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
if err != nil {
p.setError(err)
return
}
match := priorityutil.PodMatchesTermsNamespaceAndSelector(podToCheck, namespaces, selector)
if match {
func() {
p.Lock()
defer p.Unlock()
for _, node := range p.nodes {
if p.failureDomains.NodesHaveSameTopologyKey(node, fixedNode, term.TopologyKey) {
p.counts[node.Name] += weight
}
}
}()
}
}
// AnyPodMatchesPodAffinityTerm checks if any of given pods can match the specific podAffinityTerm.
// First return value indicates whether a matching pod exists on a node that matches the topology key,
// while the second return value indicates whether a matching pod exists anywhere.
// TODO: Do we really need any pod matching, or all pods matching? I think the latter.
func (c *PodAffinityChecker) anyPodMatchesPodAffinityTerm(pod *v1.Pod, allPods []*v1.Pod, node *v1.Node, term *v1.PodAffinityTerm) (bool, bool, error) {
matchingPodExists := false
namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(pod, term)
selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
if err != nil {
return false, false, err
}
for _, existingPod := range allPods {
match := priorityutil.PodMatchesTermsNamespaceAndSelector(existingPod, namespaces, selector)
if match {
matchingPodExists = true
existingPodNode, err := c.info.GetNodeInfo(existingPod.Spec.NodeName)
if err != nil {
return false, matchingPodExists, err
}
if c.failureDomains.NodesHaveSameTopologyKey(node, existingPodNode, term.TopologyKey) {
return true, matchingPodExists, nil
}
}
}
return false, matchingPodExists, nil
}
// 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
}
