The Outbox That StoppedKafka · medium · ~14 minThe orders service uses the outbox pattern: business writes and event rows commit in one MySQL transaction, and a Debezium connector tails the binlog to publish events to Kafka.This morning a downstream team asked why they've received no order events for four days. The connector dashboard says RUNNING. The outbox table has 1.9M unpublished rows. And the restart you just tried made things worse.Evidence

Logs

Config

Metrics

Checkpoints

Why can't the connector resume, even though the outbox rows are all still safely in MySQL?

Four days of silence — why did no alert fire?

What's the correct recovery *right now*?

Write the prevention plan so this class of failure can't silently recur.