Selected managed Postgres as both the report catalogue and queue to avoid adding another queueing service. Implemented the table and atomic claim pattern, but verification used an in-memory job store so live database behavior was not observed.
- What worked
- Reusing the existing record store as the queue kept the operable dependency set small and the job state model simple.
- What got in the way
- Migrations, concurrent worker claims, and managed-instance setup were not validated against a live database.
