Replication Factor

If a broker holding order events fails, another copy can keep those events available. Kafka’s replication factor controls how many broker copies each partition has. Choosing that number affects storage cost, replication traffic, and the failures a topic can withstand.

The number alone does not tell you whether every acknowledged record is safe or whether new writes can continue. Those outcomes also depend on which copies are caught up and what the producer requires before accepting a write as successful.

We’ll work through those relationships using a KRaft-based Apache Kafka 4.3 cluster. The examples assume local broker storage and non-transactional records. Any failure example also states the replica and controller conditions needed for its outcome.