Transactions in Apache Kafka | Confluent

This post dives deeper into Kafka transactions, explaining their primary use case in read-process-write stream processing applications. It details how transactions enable atomic multi-partition writes and zombie fencing to achieve exactly-once processing. The post outlines transactional semantics, the Java client API for transactions, and the underlying components like the Transaction Coordinator and Transaction Log. It clarifies that while Kafka guarantees committed transactional messages will be delivered and aborted/open ones withheld, it does not provide atomic reads.