Skip to content

Inbox and Outbox - Professional

Debezium Outbox Event Router publishes committed outbox rows through WAL; Kafka transactions atomically bind Kafka records and offsets but not arbitrary databases; PostgreSQL unique indexes enforce inbox identity. At scale, outbox table/index bloat, relay lag, dedup storage, and hot aggregate ordering dominate. Dashboard intent-to-publish age, duplicates, cleanup lag, inbox conflicts, and end-to-end reconciliation.

Best practices

  • Define stable message identity and aggregate ordering key.
  • Partition and clean tables without outrunning replay requirements.
  • Make reconciliation continuous, not incident-only.
  • Load-test relay catch-up after broker outage. 
    outbox closes DB -> broker loss gap
inbox closes broker -> DB duplicate-effect gap

Test yourself

  1. When can Kafka transactions replace an inbox?
  2. How would you partition a billion-row dedup store?
  3. What proves cleanup is safe?

Further reading

  • Richardson, Transactional Outbox pattern.
  • Debezium Outbox Event Router documentation.
  • Kafka transactions and exactly-once semantics documentation.