I read from https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Replication
So, for each committed message, we guarantee that the message is stored in >multiple replicas in memory. However, there is no guarantee that any >replica has persisted the commit message to disks though
It makes sense to only store the message in follower's memory after ack in order to achieve low latency. But the article doesn't tell whether the leader persists the message. What if the leader crashes?