Used the existing hosted database as the durable job store through the shared connection string with a dedicated job collection, and used an ephemeral in-memory server for end-to-end checks covering restart survival, delivery, duplicates, retries, and budget exhaustion.
- What worked
- Job documents persisted before the API responded, remained visible while the API was down, and were consumed after restart; the ephemeral server enabled repeatable verification without touching shared data.