diff --git a/changelogs/unreleased/10263-Lyndon-Li b/changelogs/unreleased/10263-Lyndon-Li new file mode 100644 index 000000000..14af1a9c0 --- /dev/null +++ b/changelogs/unreleased/10263-Lyndon-Li @@ -0,0 +1 @@ +Fix a potential deadlock when resultsLock is held by the informer but blocked on resChan because the early quit of RestorePodVolumes diff --git a/pkg/podvolume/restorer.go b/pkg/podvolume/restorer.go index b52558a2f..589c431c2 100644 --- a/pkg/podvolume/restorer.go +++ b/pkg/podvolume/restorer.go @@ -106,9 +106,9 @@ func newRestorer( if pvr.Status.Phase == velerov1api.PodVolumeRestorePhaseCompleted || pvr.Status.Phase == velerov1api.PodVolumeRestorePhaseFailed || pvr.Status.Phase == velerov1api.PodVolumeRestorePhaseCanceled { r.resultsLock.Lock() - defer r.resultsLock.Unlock() - resChan, ok := r.results[resultsKey(pvr.Spec.Pod.Namespace, pvr.Spec.Pod.Name)] + r.resultsLock.Unlock() + if !ok { log.Errorf("No results channel found for pod %s/%s to send pod volume restore %s/%s on", pvr.Spec.Pod.Namespace, pvr.Spec.Pod.Name, pvr.Namespace, pvr.Name) return @@ -147,7 +147,7 @@ func (r *restorer) RestorePodVolumes(data RestoreData, tracker *volume.RestoreVo r.repoLocker.Lock(repo.Name) defer r.repoLocker.Unlock(repo.Name) - resultsChan := make(chan *velerov1api.PodVolumeRestore) + resultsChan := make(chan *velerov1api.PodVolumeRestore, len(volumesToRestore)) r.resultsLock.Lock() r.results[resultsKey(data.Pod.Namespace, data.Pod.Name)] = resultsChan