github/kubernetes
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

interpodaffinity: scheduler queueing hints

Browse Source
This commit is contained in:
nayihz
2023-12-25 10:09:05 +08:00
parent fe9e4698a3
commit 0cfe4438e9
3 changed files with 293 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

88
pkg/scheduler/framework/plugins/interpodaffinity/plugin.go
View File

@@ -20,6 +20,7 @@ import (
"context"
"fmt"
v1 "k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/runtime"
listersv1 "k8s.io/client-go/listers/core/v1"
@@ -29,6 +30,7 @@ import (
"k8s.io/kubernetes/pkg/scheduler/framework"
"k8s.io/kubernetes/pkg/scheduler/framework/parallelize"
"k8s.io/kubernetes/pkg/scheduler/framework/plugins/names"
"k8s.io/kubernetes/pkg/scheduler/util"
)
// Name is the name of the plugin used in the plugin registry and configurations.
@@ -73,8 +75,8 @@ func (pl *InterPodAffinity) EventsToRegister() []framework.ClusterEventWithHint
// As a workaround, we add UpdateNodeTaint event to catch the case.
// We can remove UpdateNodeTaint when we remove the preCheck feature.
// See: https://github.com/kubernetes/kubernetes/issues/110175
{Event: framework.ClusterEvent{Resource: framework.Pod, ActionType: framework.All}},
{Event: framework.ClusterEvent{Resource: framework.Node, ActionType: framework.Add | framework.UpdateNodeLabel | framework.UpdateNodeTaint}},
{Event: framework.ClusterEvent{Resource: framework.Pod, ActionType: framework.All}, QueueingHintFn: pl.isSchedulableAfterPodChange},
{Event: framework.ClusterEvent{Resource: framework.Node, ActionType: framework.Add | framework.UpdateNodeLabel | framework.UpdateNodeTaint}, QueueingHintFn: pl.isSchedulableAfterNodeChange},
}
}
@@ -141,3 +143,85 @@ func GetNamespaceLabelsSnapshot(logger klog.Logger, ns string, nsLister listersv
logger.V(3).Info("getting namespace, assuming empty set of namespace labels", "namespace", ns, "err", err)
return
}
func (pl *InterPodAffinity) isSchedulableAfterPodChange(logger klog.Logger, pod *v1.Pod, oldObj, newObj interface{}) (framework.QueueingHint, error) {
originalPod, modifiedPod, err := util.As[*v1.Pod](oldObj, newObj)
if err != nil {
return framework.Queue, err
}
if (modifiedPod != nil && modifiedPod.Spec.NodeName == "") || (originalPod != nil && originalPod.Spec.NodeName == "") {
logger.V(5).Info("the added/updated/deleted pod is unscheduled, so it doesn't make the target pod schedulable",
"pod", klog.KObj(pod), "originalPod", klog.KObj(originalPod), "modifiedPod", klog.KObj(modifiedPod))
return framework.QueueSkip, nil
}
terms, err := framework.GetAffinityTerms(pod, framework.GetPodAffinityTerms(pod.Spec.Affinity))
if err != nil {
return framework.Queue, err
}
antiTerms, err := framework.GetAffinityTerms(pod, framework.GetPodAntiAffinityTerms(pod.Spec.Affinity))
if err != nil {
return framework.Queue, err
}
// Pod is updated. Return Queue when the updated pod matching the target pod's affinity or not matching anti-affinity.
if modifiedPod != nil && originalPod != nil {
if !podMatchesAllAffinityTerms(terms, originalPod) && podMatchesAllAffinityTerms(terms, modifiedPod) {
logger.V(5).Info("a scheduled pod was updated to match the target pod's affinity, and the pod may be schedulable now",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.Queue, nil
}
if podMatchesAllAffinityTerms(antiTerms, originalPod) && !podMatchesAllAffinityTerms(antiTerms, modifiedPod) {
logger.V(5).Info("a scheduled pod was updated not to match the target pod's anti affinity, and the pod may be schedulable now",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.Queue, nil
}
logger.V(5).Info("a scheduled pod was updated but it doesn't match the target pod's affinity or does match the target pod's anti-affinity",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.QueueSkip, nil
}
// Pod is added. Return Queue when the added pod matching the target pod's affinity.
if modifiedPod != nil {
if podMatchesAllAffinityTerms(terms, modifiedPod) {
logger.V(5).Info("a scheduled pod was added and it matches the target pod's affinity",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.Queue, nil
}
logger.V(5).Info("a scheduled pod was added and it doesn't match the target pod's affinity",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.QueueSkip, nil
}
// Pod is deleted. Return Queue when the deleted pod matching the target pod's anti-affinity.
if podMatchesAllAffinityTerms(antiTerms, originalPod) {
logger.V(5).Info("a scheduled pod was deteled but it matches the target pod's anti-affinity",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.QueueSkip, nil
}
logger.V(5).Info("a scheduled pod was deleted but it doesn't match the target pod's anti-affinity, and the pod may be schedulable now",
"pod", klog.KObj(pod), "modifiedPod", klog.KObj(modifiedPod))
return framework.Queue, nil
}
func (pl *InterPodAffinity) isSchedulableAfterNodeChange(logger klog.Logger, pod *v1.Pod, oldObj, newObj interface{}) (framework.QueueingHint, error) {
_, modifiedNode, err := util.As[*v1.Node](oldObj, newObj)
if err != nil {
return framework.Queue, err
}
terms, err := framework.GetAffinityTerms(pod, framework.GetPodAffinityTerms(pod.Spec.Affinity))
if err != nil {
return framework.Queue, err
}
for _, term := range terms {
if _, ok := modifiedNode.Labels[term.TopologyKey]; ok {
logger.V(5).Info("a node with topologyKey was added/updated and it may make pod schedulable", "pod", klog.KObj(pod), "node", klog.KObj(modifiedNode))
return framework.Queue, err
}
}
logger.V(5).Info("a node is added/updated but doesn't have any topologyKey of pod affinity", "pod", klog.KObj(pod), "node", klog.KObj(modifiedNode))
return framework.QueueSkip, nil
}