// syncJob will sync the job with the given key if it has had its expectations fulfilled, meaning
// it did not expect to see any more of its pods created or deleted. This function is not meant to be invoked
// concurrently with the same key.
func (jm *JobController) syncJob(key string) error {
startTime := time.Now()
defer func() {
glog.V(4).Infof("Finished syncing job %q (%v)", key, time.Now().Sub(startTime))
}()
if !jm.podStoreSynced() {
// Sleep so we give the pod reflector goroutine a chance to run.
time.Sleep(replicationcontroller.PodStoreSyncedPollPeriod)
glog.V(4).Infof("Waiting for pods controller to sync, requeuing job %v", key)
jm.queue.Add(key)
return nil
}
obj, exists, err := jm.jobStore.Store.GetByKey(key)
if !exists {
glog.V(4).Infof("Job has been deleted: %v", key)
jm.expectations.DeleteExpectations(key)
return nil
}
if err != nil {
glog.Errorf("Unable to retrieve job %v from store: %v", key, err)
jm.queue.Add(key)
return err
}
job := *obj.(*batch.Job)
// Check the expectations of the job before counting active pods, otherwise a new pod can sneak in
// and update the expectations after we've retrieved active pods from the store. If a new pod enters
// the store after we've checked the expectation, the job sync is just deferred till the next relist.
jobKey, err := controller.KeyFunc(&job)
if err != nil {
glog.Errorf("Couldn't get key for job %#v: %v", job, err)
return err
}
jobNeedsSync := jm.expectations.SatisfiedExpectations(jobKey)
selector, _ := unversioned.LabelSelectorAsSelector(job.Spec.Selector)
podList, err := jm.podStore.Pods(job.Namespace).List(selector)
if err != nil {
glog.Errorf("Error getting pods for job %q: %v", key, err)
jm.queue.Add(key)
return err
}
activePods := controller.FilterActivePods(podList.Items)
active := int32(len(activePods))
succeeded, failed := getStatus(podList.Items)
conditions := len(job.Status.Conditions)
if job.Status.StartTime == nil {
now := unversioned.Now()
job.Status.StartTime = &now
}
// if job was finished previously, we don't want to redo the termination
if isJobFinished(&job) {
return nil
}
if pastActiveDeadline(&job) {
// TODO: below code should be replaced with pod termination resulting in
// pod failures, rather than killing pods. Unfortunately none such solution
// exists ATM. There's an open discussion in the topic in
// https://github.com/kubernetes/kubernetes/issues/14602 which might give
// some sort of solution to above problem.
// kill remaining active pods
wait := sync.WaitGroup{}
wait.Add(int(active))
for i := int32(0); i < active; i++ {
go func(ix int32) {
defer wait.Done()
if err := jm.podControl.DeletePod(job.Namespace, activePods[ix].Name, &job); err != nil {
defer utilruntime.HandleError(err)
}
}(i)
}
wait.Wait()
// update status values accordingly
failed += active
active = 0
job.Status.Conditions = append(job.Status.Conditions, newCondition(batch.JobFailed, "DeadlineExceeded", "Job was active longer than specified deadline"))
jm.recorder.Event(&job, api.EventTypeNormal, "DeadlineExceeded", "Job was active longer than specified deadline")
} else {
if jobNeedsSync {
active = jm.manageJob(activePods, succeeded, &job)
}
completions := succeeded
complete := false
if job.Spec.Completions == nil {
// This type of job is complete when any pod exits with success.
// Each pod is capable of
// determining whether or not the entire Job is done. Subsequent pods are
// not expected to fail, but if they do, the failure is ignored. Once any
// pod succeeds, the controller waits for remaining pods to finish, and
// then the job is complete.
if succeeded > 0 && active == 0 {
complete = true
}
} else {
// Job specifies a number of completions. This type of job signals
// success by having that number of successes. Since we do not
// start more pods than there are remaining completions, there should
// not be any remaining active pods once this count is reached.
if completions >= *job.Spec.Completions {
complete = true
if active > 0 {
jm.recorder.Event(&job, api.EventTypeWarning, "TooManyActivePods", "Too many active pods running after completion count reached")
}
if completions > *job.Spec.Completions {
jm.recorder.Event(&job, api.EventTypeWarning, "TooManySucceededPods", "Too many succeeded pods running after completion count reached")
}
}
}
if complete {
job.Status.Conditions = append(job.Status.Conditions, newCondition(batch.JobComplete, "", ""))
now := unversioned.Now()
job.Status.CompletionTime = &now
}
}
// no need to update the job if the status hasn't changed since last time
if job.Status.Active != active || job.Status.Succeeded != succeeded || job.Status.Failed != failed || len(job.Status.Conditions) != conditions {
job.Status.Active = active
job.Status.Succeeded = succeeded
job.Status.Failed = failed
if err := jm.updateHandler(&job); err != nil {
glog.Errorf("Failed to update job %v, requeuing. Error: %v", job.Name, err)
jm.enqueueController(&job)
}
}
return nil
}
// syncReplicaSet will sync the ReplicaSet with the given key if it has had its expectations fulfilled,
// meaning it did not expect to see any more of its pods created or deleted. This function is not meant to be
// invoked concurrently with the same key.
func (rsc *ReplicaSetController) syncReplicaSet(key string) error {
startTime := time.Now()
defer func() {
glog.V(4).Infof("Finished syncing replica set %q (%v)", key, time.Now().Sub(startTime))
}()
if !rsc.podStoreSynced() {
// Sleep so we give the pod reflector goroutine a chance to run.
time.Sleep(PodStoreSyncedPollPeriod)
glog.Infof("Waiting for pods controller to sync, requeuing ReplicaSet %v", key)
rsc.queue.Add(key)
return nil
}
obj, exists, err := rsc.rsStore.Store.GetByKey(key)
if !exists {
glog.Infof("ReplicaSet has been deleted %v", key)
rsc.expectations.DeleteExpectations(key)
return nil
}
if err != nil {
glog.Infof("Unable to retrieve ReplicaSet %v from store: %v", key, err)
rsc.queue.Add(key)
return err
}
rs := *obj.(*extensions.ReplicaSet)
// Check the expectations of the ReplicaSet before counting active pods, otherwise a new pod can sneak
// in and update the expectations after we've retrieved active pods from the store. If a new pod enters
// the store after we've checked the expectation, the ReplicaSet sync is just deferred till the next
// relist.
rsKey, err := controller.KeyFunc(&rs)
if err != nil {
glog.Errorf("Couldn't get key for ReplicaSet %#v: %v", rs, err)
return err
}
rsNeedsSync := rsc.expectations.SatisfiedExpectations(rsKey)
selector, err := unversioned.LabelSelectorAsSelector(rs.Spec.Selector)
if err != nil {
glog.Errorf("Error converting pod selector to selector: %v", err)
return err
}
podList, err := rsc.podStore.Pods(rs.Namespace).List(selector)
if err != nil {
glog.Errorf("Error getting pods for ReplicaSet %q: %v", key, err)
rsc.queue.Add(key)
return err
}
// TODO: Do this in a single pass, or use an index.
filteredPods := controller.FilterActivePods(podList.Items)
if rsNeedsSync {
rsc.manageReplicas(filteredPods, &rs)
}
// Count the number of pods that have labels matching the labels of the pod
// template of the replicaSet, the matching pods may have more labels than
// are in the template. Because the label of podTemplateSpec is a superset
// of the selector of the replicaset, so the possible matching pods must be
// part of the filteredPods.
fullyLabeledReplicasCount := 0
templateLabel := labels.Set(rs.Spec.Template.Labels).AsSelector()
for _, pod := range filteredPods {
if templateLabel.Matches(labels.Set(pod.Labels)) {
fullyLabeledReplicasCount++
}
}
// Always updates status as pods come up or die.
if err := updateReplicaCount(rsc.kubeClient.Extensions().ReplicaSets(rs.Namespace), rs, len(filteredPods), fullyLabeledReplicasCount); err != nil {
// Multiple things could lead to this update failing. Requeuing the replica set ensures
// we retry with some fairness.
glog.V(2).Infof("Failed to update replica count for controller %v/%v; requeuing; error: %v", rs.Namespace, rs.Name, err)
rsc.enqueueReplicaSet(&rs)
}
return nil
}
