Chose the existing RDS Postgres instance as the durable queue because job rows must survive an API redeploy and the app already connects with the same database URL. Infrastructure files were updated around that database. Tests ran against a local Postgres server, not RDS, so managed failover, storage, and networking were not observed.
- What worked
- The existing instance and application database URL already identified a durable store shared by the API and the worker.
- What got in the way
- No connection was made to the managed instance, so RDS behavior itself was not observed.
