add flag --enable_online_fs_expansion: control whether pods that refe…

…rence the resized volume need to be restarted.

Kubernetes v1.11 also introduces an alpha feature called online file system expansion. This feature enables file system expansion while a volume is still in-use by a pod. Because this feature is alpha, it requires enabling the feature gate, ExpandInUsePersistentVolumes. It is supported by the in-tree volume plugins GCE-PD, AWS-EBS, Cinder, and Ceph RBD. When this feature is enabled, pod referencing the resized volume do not need to be restarted. Instead, the file system will automatically be resized while in use as part of volume expansion. File system expansion does not happen until a pod references the resized volume, so if no pods referencing the volume are running file system expansion will not happen.

Signed-off-by: zhiyong.huang <[email protected]>

pkg/controller/vitessshard/reconcile_disk.go

@@ -105,7 +105,9 @@ func (r *ReconcileVitessShard) reconcileDisk(ctx context.Context, vts *planetsca
 
 	// If disk size has changed and the changes are all ready, mark the shard as ready to cascade. Otherwise, skip this.
 	if anythingChanged {
-		rollout.Cascade(vts)
+		if !*onlineFileSystemExpansion {
+			rollout.Cascade(vts)
+		}
 		err := r.client.Update(ctx, vts)
 		if err != nil {
 			return resultBuilder.Error(err)

pkg/controller/vitessshard/reconcile_tablets.go

@@ -182,7 +182,9 @@ func (r *ReconcileVitessShard) reconcileTablets(ctx context.Context, vts *planet
 		UpdateRollingRecreate: func(key client.ObjectKey, obj runtime.Object) {
 			newObj := obj.(*corev1.Pod)
 			tablet := tabletMap[key]
-			r.updatePVCFilesystemResizeAnnotation(ctx, tablet, newObj)
+			if !*onlineFileSystemExpansion {
+				r.updatePVCFilesystemResizeAnnotation(ctx, tablet, newObj)
+			}
 			vttablet.UpdatePod(newObj, tablet)
 		},
 		Status: func(key client.ObjectKey, obj runtime.Object) {

pkg/controller/vitessshard/vitessshard_controller.go

@@ -50,6 +50,7 @@ const (
 var (
 	maxConcurrentReconciles = flag.Int("vitessshard_concurrent_reconciles", 10, "the maximum number of different vitessshards to reconcile concurrently")
 	resyncPeriod            = flag.Duration("vitessshard_resync_period", 30*time.Second, "reconcile vitessshards with this period even if no Kubernetes events occur")
+	onlineFileSystemExpansion = flag.Bool("enable_online_fs_expansion", true, "if true, pod referencing the resized volume do not need to be restarted, but provided that the volume plug-in supports, such as GCE-PD, AWS-EBS, Cinder, and Ceph RBD")
 )
 
 var log = logrus.WithField("controller", "VitessShard")