Multi-Geo Replication in Apache Kafka

This post details the evolution of multi-geo replication for Apache Kafka, covering stretched clusters (synchronous replication across datacenters) and connected clusters (asynchronous replication using Kafka Connect or Cluster Linking). It explains Kafka's replication mechanisms (ISR, leader epochs, acks), introduces Confluent Server's observer replicas with automatic promotion for enhanced availability, and elaborates on Cluster Linking's ability to mirror topics, synchronize metadata (consumer offsets, ACLs), and maintain offset parity across clusters. It also touches upon Fetch From Followers (KIP-392) for consumer read optimization.