Topics, partitions, replication — leaders, ISR, acks, durability

acks=1 means that the leader has persisted the record. Suppose the leader then crashes before any follower replicates the record. Then the record is lost, even though the producer got an ack.

Source: https://kafka.apache.org/documentation/#producerconfigs_acks

With acks=all, the leader requires at least min.insync.replicas members in the ISR. ISR=1 is less than 2. So the leader refuses the write. It does not acknowledge a write that lives on only one broker.

Source: https://kafka.apache.org/documentation/#topicconfigs_min.insync.replicas

One broker down means the ISR goes from 3 to 2, which is >= misr. Writes continue on >=2 brokers. A second broker down means the ISR goes from 2 to 1, which is < misr. Kafka refuses writes. So you never ack a write that would survive on only one disk. EOS is a separate topic (KFK-05).

Source: https://kafka.apache.org/documentation/#replication

Unclean election promotes an out-of-sync replica to restore availability when all in-sync replicas are gone. The cost is that you lose the records that only the dead in-sync replicas had. The default is off. Keep it off for completeness-critical data.

Source: https://kafka.apache.org/documentation/#brokerconfigs_unclean.leader.election.enable

The ISR is dynamic. It shrinks when replicas lag or die. It grows when they catch up. The leader is always in the ISR. Clean election chooses from the ISR. The ISR is not fixed, so D is false.

Source: https://kafka.apache.org/documentation/#replication

misr must be <= RF. With RF=1, the ISR is at most 1. This is below misr=2, so acks=all writes always fail. Set min.insync.replicas to match the replication factor (RF=3 -> misr=2).

Source: https://kafka.apache.org/documentation/#topicconfigs_min.insync.replicas

Kafka 4.x runs KRaft only. The controller quorum keeps cluster metadata (including leadership) in a replicated log. ZooKeeper is gone.

Source: https://kafka.apache.org/documentation/#kraft