Mark dataupload datadownload status failed when velero pod restart

Signed-off-by: Ming Qiu <mqiu@vmware.com>
This commit is contained in:
Ming Qiu
2023-07-10 07:14:24 +00:00
committed by Ming
parent e54a8af0ad
commit 480fe445b1
6 changed files with 277 additions and 38 deletions
+62 -2
View File
@@ -256,11 +256,15 @@ func (s *nodeAgentServer) run() {
s.logger.WithError(err).Fatal("Unable to create the pod volume restore controller")
}
if err = controller.NewDataUploadReconciler(s.mgr.GetClient(), s.kubeClient, s.csiSnapshotClient.SnapshotV1(), repoEnsurer, clock.RealClock{}, credentialGetter, s.nodeName, s.fileSystem, s.logger).SetupWithManager(s.mgr); err != nil {
dataUploadReconciler := controller.NewDataUploadReconciler(s.mgr.GetClient(), s.kubeClient, s.csiSnapshotClient.SnapshotV1(), repoEnsurer, clock.RealClock{}, credentialGetter, s.nodeName, s.fileSystem, s.logger)
s.markDataUploadsCancel(dataUploadReconciler)
if err = dataUploadReconciler.SetupWithManager(s.mgr); err != nil {
s.logger.WithError(err).Fatal("Unable to create the data upload controller")
}
if err = controller.NewDataDownloadReconciler(s.mgr.GetClient(), s.kubeClient, repoEnsurer, credentialGetter, s.nodeName, s.logger).SetupWithManager(s.mgr); err != nil {
dataDownloadReconciler := controller.NewDataDownloadReconciler(s.mgr.GetClient(), s.kubeClient, repoEnsurer, credentialGetter, s.nodeName, s.logger)
s.markDataDownloadsCancel(dataDownloadReconciler)
if err = dataDownloadReconciler.SetupWithManager(s.mgr); err != nil {
s.logger.WithError(err).Fatal("Unable to create the data download controller")
}
@@ -333,6 +337,62 @@ func (s *nodeAgentServer) markInProgressCRsFailed() {
s.markInProgressPVRsFailed(client)
}
func (s *nodeAgentServer) markDataUploadsCancel(r *controller.DataUploadReconciler) {
// the function is called before starting the controller manager, the embedded client isn't ready to use, so create a new one here
client, err := ctrlclient.New(s.mgr.GetConfig(), ctrlclient.Options{Scheme: s.mgr.GetScheme()})
if err != nil {
s.logger.WithError(errors.WithStack(err)).Error("failed to create client")
return
}
if dataUploads, err := r.FindDataUploads(s.ctx, client, s.namespace); err != nil {
s.logger.WithError(errors.WithStack(err)).Error("failed to find data downloads")
} else {
for i := range dataUploads {
du := dataUploads[i]
if du.Status.Phase == velerov2alpha1api.DataUploadPhaseAccepted ||
du.Status.Phase == velerov2alpha1api.DataUploadPhasePrepared ||
du.Status.Phase == velerov2alpha1api.DataUploadPhaseInProgress {
updated := du.DeepCopy()
updated.Spec.Cancel = true
updated.Status.Message = fmt.Sprintf("found a dataupload with status %q during the node-agent starting, mark it as cancel", du.Status.Phase)
if err := client.Patch(s.ctx, updated, ctrlclient.MergeFrom(&du)); err != nil {
s.logger.WithError(errors.WithStack(err)).Errorf("failed to mark datadownload %q cancel", du.GetName())
continue
}
s.logger.WithField("dataupload", du.GetName()).Warn(du.Status.Message)
}
}
}
}
func (s *nodeAgentServer) markDataDownloadsCancel(r *controller.DataDownloadReconciler) {
// the function is called before starting the controller manager, the embedded client isn't ready to use, so create a new one here
client, err := ctrlclient.New(s.mgr.GetConfig(), ctrlclient.Options{Scheme: s.mgr.GetScheme()})
if err != nil {
s.logger.WithError(errors.WithStack(err)).Error("failed to create client")
return
}
if dataDownloads, err := r.FindDataDownloads(s.ctx, client, s.namespace); err != nil {
s.logger.WithError(errors.WithStack(err)).Error("failed to find data downloads")
} else {
for i := range dataDownloads {
dd := dataDownloads[i]
if dd.Status.Phase == velerov2alpha1api.DataDownloadPhaseAccepted ||
dd.Status.Phase == velerov2alpha1api.DataDownloadPhasePrepared ||
dd.Status.Phase == velerov2alpha1api.DataDownloadPhaseInProgress {
updated := dd.DeepCopy()
updated.Spec.Cancel = true
updated.Status.Message = fmt.Sprintf("found a datadownload with status %q during the node-agent starting, mark it as cancel", dd.Status.Phase)
if err := client.Patch(s.ctx, updated, ctrlclient.MergeFrom(dd)); err != nil {
s.logger.WithError(errors.WithStack(err)).Errorf("failed to mark datadownload %q cancel", dd.GetName())
continue
}
s.logger.WithField("datadownload", dd.GetName()).Warn(dd.Status.Message)
}
}
}
}
func (s *nodeAgentServer) markInProgressPVBsFailed(client ctrlclient.Client) {
pvbs := &velerov1api.PodVolumeBackupList{}
if err := client.List(s.ctx, pvbs, &ctrlclient.MatchingFields{"metadata.namespace": s.namespace}); err != nil {
+54 -2
View File
@@ -1065,7 +1065,7 @@ func markInProgressBackupsFailed(ctx context.Context, client ctrlclient.Client,
}
for i, backup := range backups.Items {
if backup.Status.Phase != velerov1api.BackupPhaseInProgress {
if backup.Status.Phase != velerov1api.BackupPhaseInProgress && backup.Status.Phase != velerov1api.BackupPhaseWaitingForPluginOperations {
log.Debugf("the status of backup %q is %q, skip", backup.GetName(), backup.Status.Phase)
continue
}
@@ -1078,6 +1078,7 @@ func markInProgressBackupsFailed(ctx context.Context, client ctrlclient.Client,
continue
}
log.WithField("backup", backup.GetName()).Warn(updated.Status.FailureReason)
markDataUploadsCancel(ctx, client, backup, log)
}
}
@@ -1088,7 +1089,7 @@ func markInProgressRestoresFailed(ctx context.Context, client ctrlclient.Client,
return
}
for i, restore := range restores.Items {
if restore.Status.Phase != velerov1api.RestorePhaseInProgress {
if restore.Status.Phase != velerov1api.RestorePhaseInProgress && restore.Status.Phase != velerov1api.RestorePhaseWaitingForPluginOperations {
log.Debugf("the status of restore %q is %q, skip", restore.GetName(), restore.Status.Phase)
continue
}
@@ -1101,5 +1102,56 @@ func markInProgressRestoresFailed(ctx context.Context, client ctrlclient.Client,
continue
}
log.WithField("restore", restore.GetName()).Warn(updated.Status.FailureReason)
markDataDownloadsCancel(ctx, client, restore, log)
}
}
func markDataUploadsCancel(ctx context.Context, client ctrlclient.Client, backup velerov1api.Backup, log logrus.FieldLogger) {
dataUploads := &velerov2alpha1api.DataUploadList{}
if err := client.List(ctx, dataUploads, &ctrlclient.MatchingFields{"metadata.namespace": backup.GetNamespace()}, &ctrlclient.MatchingLabels{velerov1api.BackupUIDLabel: string(backup.GetUID())}); err != nil {
log.WithError(errors.WithStack(err)).Error("failed to list dataUploads")
return
}
for i := range dataUploads.Items {
du := dataUploads.Items[i]
if du.Status.Phase == velerov2alpha1api.DataUploadPhaseAccepted ||
du.Status.Phase == velerov2alpha1api.DataUploadPhasePrepared ||
du.Status.Phase == velerov2alpha1api.DataUploadPhaseInProgress {
updated := du.DeepCopy()
updated.Spec.Cancel = true
updated.Status.Message = fmt.Sprintf("found a dataupload with status %q during the velero server starting, mark it as cancel", du.Status.Phase)
if err := client.Patch(ctx, updated, ctrlclient.MergeFrom(&du)); err != nil {
log.WithError(errors.WithStack(err)).Errorf("failed to mark dataupload %q cancel", du.GetName())
continue
}
log.WithField("dataupload", du.GetName()).Warn(du.Status.Message)
}
}
}
func markDataDownloadsCancel(ctx context.Context, client ctrlclient.Client, restore velerov1api.Restore, log logrus.FieldLogger) {
dataDownloads := &velerov2alpha1api.DataDownloadList{}
if err := client.List(ctx, dataDownloads, &ctrlclient.MatchingFields{"metadata.namespace": restore.GetNamespace()}, &ctrlclient.MatchingLabels{velerov1api.RestoreUIDLabel: string(restore.GetUID())}); err != nil {
log.WithError(errors.WithStack(err)).Error("failed to list dataDownloads")
return
}
for i := range dataDownloads.Items {
dd := dataDownloads.Items[i]
if dd.Status.Phase == velerov2alpha1api.DataDownloadPhaseAccepted ||
dd.Status.Phase == velerov2alpha1api.DataDownloadPhasePrepared ||
dd.Status.Phase == velerov2alpha1api.DataDownloadPhaseInProgress {
updated := dd.DeepCopy()
updated.Spec.Cancel = true
updated.Status.Message = fmt.Sprintf("found a datadownload with status %q during the velero server starting, mark it as cancel", dd.Status.Phase)
if err := client.Patch(ctx, updated, ctrlclient.MergeFrom(&dd)); err != nil {
log.WithError(errors.WithStack(err)).Errorf("failed to mark dataupload %q cancel", dd.GetName())
continue
}
log.WithField("datadownload", dd.GetName()).Warn(dd.Status.Message)
}
}
}