Start the telemetry in create_app, which gunicorn runs in every worker
after fork because preload_app is off, and in the relay and worker
processes with their own service names, and stop it after the rest of
the shutdown within TELEMETRY_SHUTDOWN_SECONDS. The API gets server
spans and the http.server.request.duration histogram in seconds from
the FastAPI instrumentation with stable HTTP semantic conventions,
without /health probes and without the internal send and receive
spans, and the request pipeline puts X-Request-Id on the server span
as messenger.request_id. The worker broker gets the FastStream
RabbitTelemetryMiddleware, so every delivery runs inside a consumer
span.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>