event-driven

The outbox pattern ensures reliable dual-writes by capturing events in a local database outbox table within the same transaction as business state changes, then polling or streaming them to a broker to guarantee delivery without data inconsistency.

Idempotent consumers in Kafka or other brokers typically store processed event identifiers and check them before handling, ensuring duplicate deliveries are safely ignored without re-applying state mutations or triggering side effects.

Event sourcing stores domain changes as an immutable event log, while CQRS read-model projections subscribe to these events to build query-optimized views, enabling state replay and full auditing by reprocessing past events.

A dead-letter queue should be used when messages fail processing after exceeding retry limits, isolating poison pills or unprocessable events for later debugging and reprocessing without blocking the main consumer pipeline.

Safe schema evolution requires defining backward-compatible event envelopes and applying schema versioning strategies, allowing consumers to handle multiple event versions gracefully without breaking downstream processing during transitions.

Yes, the patterns apply to Kafka, RabbitMQ, and cloud brokers, providing guidance on topic and partition topology, delivery semantics, and cross-service coordination tailored to each specific messaging platform's capabilities.