Gunnar Morling

A community-driven exploration of strategies for handling Kafka broker unavailability in producer applications. The post synthesizes a Twitter discussion covering options like in-memory buffering, local disk persistence, embedded brokers (ActiveMQ/Artemis), cloud storage fallbacks (DynamoDB/S3), and the transactional outbox pattern with CDC tools like Debezium. Key trade-offs discussed include consistency guarantees, failure domains, latency, and operational complexity. The conclusion emphasizes that the right approach depends on business requirements and acceptable risk, and that outages are inevitable so explicit policies must be defined upfront.

O Kafka, Where Art Thou?