Comment by damienkatz

"...if the client wanted true multi-node durability, then the write wouldn't have succeeded (the client would timeout waiting for replicas(s) to receive the update) and the client wouldn't unknowingly lose data."