Michał Matłoka

Apache Kafka is often the backbone of modern event-driven systems, but many teams still underestimate how difficult it is to design reliable disaster recovery and multi-region deployments. In-cluster replication alone does not protect against region failures, network partitions, or operational mistakes. This talk is a practical guide to Kafka disaster recovery architectures used in production. We will compare active-passive, active-active, 3-data-center, and 2.5-DC topologies, discussing their trade-offs in terms of RPO, RTO, latency, operational complexity, and cost. The session will also cover Kafka replication technologies such as MirrorMaker 2 and Cluster Linking, as well as advanced topics like observer replicas, automatic observer promotion, and replica placement constraints. Rather than focusing on theoretical "perfect architectures," this talk emphasizes real-world engineering trade-offs, common pitfalls, and lessons learned from operating Kafka at scale. Attendees will leave with a clear understanding of how to choose the right multi-region Kafka strategy and avoid the most common disaster recovery mistakes.