Chose managed Postgres as the shared store so web and worker hosts can both reach bookings and jobs after replacement. Wired the app to a connection URL and documented attach, but never provisioned or tested the managed cluster.
- What worked
- The attach-and-shared-URL model was clear enough to split web and worker process groups against one database. That matches the need for a store that is not tied to either machine’s disk.
- What got in the way
- There was no live account or cluster in this environment, so setup, secrets, failover, and attach behavior were not exercised. Local verification used a substitute server instead.
