Replicate on Write (ROW) ensures that two replicas of your content exist on two nodes before the UUID and successful completion code are returned to your client. This features guards against a node that accepts a write and fails prior to being able to replicate the object to another node. This is most often used in environments with high, sustained writes, or very sensitive data. As of Swarm 8.0, this is the default mode of operation.
If you wish to change this, please add this to your cluster's configuration file: