How to Design a Kafka-Based Payment Event Pipeline: Key Architecture Decisions

A fintech engineer with production experience building Kafka-based payment pipelines in Brazil outlines the critical design decisions teams often overlook. Before choosing Kafka, teams must define correctness requirements around ordering, durability, and latency, since Kafka guarantees ordering only within a partition, not across them. The architecture recommends separate topics for each payment lifecycle stage — such as initiated, authorized, settled, and failed — rather than a single topic filtered by event type, enabling independent scaling and retention policies. Every consumer must be built to be idempotent, using database-level checks and the transactional outbox pattern to ensure atomicity between state changes and downstream events. Dead letter topics should be set up for every consumer topic, with DLT queue depth monitored as a critical incident signal to prevent poison messages from stalling entire partitions.