Kafka Levers

Setting min.insync.replicas to a majority quorum (e.g. $(replication factor / 2) + 1)$ and $acks$ to $all$ would allow one to enforce stricter durability guarantees, while also achieving good availability. Let’s assume $replication factor = 5$, so there are $5$ replicas per partition and $min.insync.replicas = 3$. This would mean up to $2$ node failures can be tolerated with zero data loss, and the cluster is still available for writes and reads.

Setting min.insync.replicas equal to $replication factor$ and $acks$ to $all$ would provide even stronger durability guarantees at the expense of lower availability. In our previous example of $replication factor = 5$, this would mean that up to $4$ node failures can now be tolerated with zero data loss. However, a single node failure makes the cluster unavailable for writes.

Setting $acks$ to $1$ can provide better performance at the expense of weaker durability and consistency guarantees. For example, records will be considered committed and acknowledged as soon as the leader has stored them locally without waiting for any of the followers to catch up. However, in case of a leader’s failure and election of a new leader, records that the previous leader acknowledged but did not make it to the new leader yet will be lost.