Producer Acknowledgments

An order service publishes an event and receives a successful result. What has Kafka actually confirmed? The answer depends on the producer’s acknowledgment setting. Success might mean the client sent bytes without waiting for the broker, the leader appended the record, or the write met a replication requirement.

That distinction matters when a broker fails immediately afterward. It also matters when the producer receives no response: the record may be missing, or Kafka may already have stored it.

In this chapter, we’ll compare the acknowledgment modes, connect them to replication, and examine what an application can conclude from success and failure. The examples use the Apache Kafka Java client 4.3.1 and ordinary, non-transactional writes to a KRaft-based cluster.