For the raft algorithm to be correct Fsync is required on a majority of node otherwise you are technically not implementing Raft. The reason is that in Raft if

For the raft algorithm to be correct Fsync is required on a majority of node otherwise you are technically not implementing Raft. The reason is that in Raft if a node acknowledges to the leader that it wrote something to the log it must not later accept a different write in the same log position. This mean if for some reason server rebooted with dirty buffered writes that could not be flushed in time. it’s supposed to forgot everything it know and rejoin the cluster using a brand new node id.