Notes / Event-driven Java
Event-driven architecture is well documented right up to the point where it meets a real database, a real compliance requirement and a consumer owned by another team. These notes are about that point: replay safety, write-ahead log pressure, schema evolution across organizational boundaries, and audit trails that survive review.
Backward compatible is not the same as safe to deploy. Which compatibility mode to set, why defaults matter more than types, and how to retire a field over quarters.
An inactive replication slot pins the write-ahead log and fills the disk. Why it happens on quiet tables, and the heartbeat that stops it.
At-least-once delivery means a consumer sees some messages twice. Why a deduplication table alone falls short for financial operations, and what to add.