Co-locating workflow state with data in Postgres for exactly-once execution

Category: Engineering post
Status: discussion
Sources: DBOS blog, HN discussion
Summary: A DBOS post argues that keeping durable-workflow checkpoints in the same Postgres database as application data lets the step checkpoint and the data update commit in one transaction. That removes the failure window that forces application-level idempotency bookkeeping, and it collapses the transactional-outbox pattern: either the update commits and the workflow is enqueued or neither happens.
Why it matters: It reframes a common distributed-systems problem (exactly-once execution across a workflow engine and a database) as a single-database transaction for teams already standardized on Postgres.