kafka-event-driven-design

# Kafka Event-Driven Design Review ## Quick Reference §1 Scope · §2 Mandatory Gates · §3 Depth & Reference Loading · §4 Degradation Modes · §5 Design Checklist · §6 Partition Design · §7 Anti-Examples · §8 Scorecard · §9 Output Contract **Before scoring a version-sensitive configuration item, read `references/version-client-matrix.md`.** Kafka defaults changed materially at 3.0 and again at 4.0, and the four major client libraries disagree with each other. Which items are version-sensitive — and which fact each one needs — is §2's *Minimum context per scored item*; most of the checklist needs no version at all. --- ## §1 Scope **In scope** — Kafka event-driven architecture for production backend services: - Topic design (naming, partition count, replication factor, retention) - Partition key strategy (ordering guarantees, hot partition avoidance) - Event schema design (Avro/Protobuf/JSON Schema, schema evolution, compatibility) - Producer configuration (acks, retries, idempotence, transactional producers) - Consumer group design (assignment strategy, rebalance handling, commit strategy) - Idempotent consumption (deduplication, idempotency keys, exactly-once semantics) - Dead letter queue (DLQ) and retry patterns - Backpressure and consumer lag management - Schema Registry integration and compatibility modes **Out of scope** — say so and redirect rather than answering partially: | Out of scope | Why it is not this skill | |--------------|--------------------------| | ...