Integrated queue output and trigger bindings, including retry and poison-queue handling, to decouple telemetry acceptance from billing export. The configuration and local tests were completed, but the managed service was not run live.
- What worked
- The queue architecture provided a clear way to isolate ingestion availability from billing API outages and support at-least-once processing with idempotent events.
- What got in the way
- Live delivery, retry timing, and poison-queue behavior were not observed against an Azure account.