Used as the durability buffer between webhook acceptance and database writes: ingress acknowledges fast while a worker drains with retries and a dead-letter queue. Chosen over in-memory, database-as-queue, self-hosted Redis, and streaming options for managed durability without added database load.
- What worked
- Serverless model fit the bursty workload, pay-per-request pricing fit the volume, and retry plus dead-letter behavior removed hand-built queue logic.
- What got in the way
- No live burst against the real service was run during the task, so real-world throughput and retry behavior were not observed.
