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

Cleanup the Job controller handling of terminating pods

Browse Source
This commit is contained in:
Michal Wozniak 2023-05-16 08:45:19 +02:00
parent 70033bf843
commit 0fe27a06f9
1 changed files with 5 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
pkg/controller/job/job_controller.go
View File

@ -989,23 +989,8 @@ func (jm *Controller) trackJobStatusAndRemoveFinalizers(ctx context.Context, job
// This pod was processed in a previous sync.
continue
}
// Terminating pods are counted as failed. This guarantees that orphan Pods
// count as failures.
// Active pods are terminated when the job has completed, thus they count as
// failures as well.
considerTerminated := pod.DeletionTimestamp != nil || finishedCond != nil
if feature.DefaultFeatureGate.Enabled(features.PodDisruptionConditions) && feature.DefaultFeatureGate.Enabled(features.JobPodFailurePolicy) && job.Spec.PodFailurePolicy != nil {
// TODO(#113855): Stop limiting this behavior to Jobs with podFailurePolicy.
// For now, we do so to avoid affecting all running Jobs without the
// avaibility to opt-out into the old behavior.
// We can also simplify the check to remove finalizers to:
// considerTerminated || job.DeletionTimestamp != nil
considerTerminated = podutil.IsPodTerminal(pod) ||
finishedCond != nil || // The Job is terminating. Any running Pod is considered failed.
isPodFailed(pod, job)
}
if podutil.IsPodTerminal(pod) || considerTerminated || job.DeletionTimestamp != nil {
considerPodFailed := isPodFailed(pod, job)
if podutil.IsPodTerminal(pod) || considerPodFailed || finishedCond != nil || job.DeletionTimestamp != nil {
podsToRemoveFinalizer = append(podsToRemoveFinalizer, pod)
}
if pod.Status.Phase == v1.PodSucceeded && !uncounted.failed.Has(string(pod.UID)) {
@ -1021,7 +1006,8 @@ func (jm *Controller) trackJobStatusAndRemoveFinalizers(ctx context.Context, job
needsFlush = true
uncountedStatus.Succeeded = append(uncountedStatus.Succeeded, pod.UID)
}
} else if pod.Status.Phase == v1.PodFailed || considerTerminated {
} else if considerPodFailed || finishedCond != nil {
// When the job is considered finished, every non-terminated pod is considered failed
ix := getCompletionIndex(pod.Annotations)
if !uncounted.failed.Has(string(pod.UID)) && (!isIndexed || (ix != unknownCompletionIndex && ix < int(*job.Spec.Completions))) {
if feature.DefaultFeatureGate.Enabled(features.JobPodFailurePolicy) && job.Spec.PodFailurePolicy != nil {
@ -1702,7 +1688,7 @@ func isPodFailed(p *v1.Pod, job *batch.Job) bool {
// terminating Pods are marked as Failed. So we only need to check the phase.
// TODO(#113855): Stop limiting this behavior to Jobs with podFailurePolicy.
// For now, we do so to avoid affecting all running Jobs without the
// avaibility to opt-out into the old behavior.
// availability to opt-out into the old behavior.
return p.Status.Phase == v1.PodFailed
}
if p.Status.Phase == v1.PodFailed {