Expose FastAPI request metrics (latency / count) via OpenTelemetry middleware

[1fanwang]

Description

Airflow 3.0 moved the public API and UI-backing endpoints from Flask to FastAPI (airflow-core/src/airflow/api_fastapi/). Operators monitoring production deployments need per-route request count, latency, and status distributions to diagnose slow endpoints (Grid TI-summary fan-out, log download, DAG list paging). Today the API server emits process-level metrics but no per-route HTTP timing.

The existing OTel metrics infrastructure (airflow/metrics/otel_logger.py) already supports histograms. Adding a small FastAPI middleware that records request duration with route, method, status attributes would close the gap without bringing in a third-party dependency like prometheus-fastapi-instrumentator.

Use case / motivation

• Operators need to alert on p99 latency for /api/v2/dags/{dag_id}/dagRuns before users notice — today the only signal is gunicorn worker timeouts.
• UI-perf regression detection on the Grid view depends on knowing which back-end endpoints regress, not just the front-end page-load time.
• Capacity planning for the API server (worker count, replica count) needs request-rate histograms grouped by route.

Proposal

Add an opt-in FastAPI middleware that records:

• airflow.api.requests_total (counter, tagged route/method/status)
• airflow.api.request_duration_seconds (histogram, tagged route/method/status)

Wired into airflow/api_fastapi/app.py's create_app() and gated on the existing metrics.statsd_on / OTel-enabled flags so operators not on a metrics backend pay no cost.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct