Three Real-World Failures That Expose Hidden Flaws in the Outbox Pattern

The outbox pattern — writing events in the same database transaction as the triggering row, then polling and forwarding to a message broker — is widely described as a reliable consistency technique. A developer running it in production between two real services encountered three distinct failure modes that only surface when the system is actually polling a table and talking to a broker. The first flaw involved no retry ceiling: a permanently failing message blocked all rows behind it indefinitely, with no alert or escalation path. The fix required tracking attempt counts and parking stuck rows as FAILED after a configurable limit, so downstream messages could continue processing. Additional issues arose around distributed trace context, which HTTP calls propagate automatically but polled outbox messages do not, highlighting how real operational gaps differ sharply from the pattern's clean theoretical description.