Monitor the service

GET /metrics exposes Prometheus metrics. Scrape it from Prometheus and chart it in Grafana.

scrape_configs:
  - job_name: mes
    static_configs:
      - targets: ["mes:8000"]

Warning

The endpoint is unauthenticated. Make it reachable only by your Prometheus scraper, not through a public gateway.

Metrics

Metric	Type	Labels
mes_extract_requests_total	Counter	status: success, partial, failed, rejected, error
mes_extract_request_duration_seconds	Histogram
mes_jobs_waiting	Gauge
mes_jobs_waiting_bytes	Gauge
mes_batch_requests_total	Counter	status: success, partial, failed, rejected, accepted
mes_batch_files	Histogram
mes_uploaded_file_bytes	Histogram
mes_uploaded_bytes_total	Counter
mes_tool_runs_total	Counter	tool, status
mes_tool_duration_seconds	Histogram	tool

Notes

• Asynchronous jobs are counted in mes_extract_requests_total (and the duration histogram) once a worker has run them, never while they wait or if they expire first. Each file of a batch counts as one request.
• mes_jobs_waiting and mes_jobs_waiting_bytes are read from the job store once per scrape, so they are true totals with any number of workers (with Redis they cover all workers). If the job store cannot be read, these two are left out of that scrape.
• Prometheus uses base units. For MB, use mes_jobs_waiting_bytes / 1e6. Accumulated GB uploaded per day: increase(mes_uploaded_bytes_total[1d]) / 1e9.
• In the container, gunicorn workers share metrics through PROMETHEUS_MULTIPROC_DIR.