Monitoring

Health check

curl http://localhost:5000/health

Returns {"status":"healthy"} with HTTP 200 when the app is serving.

Docker health status

The container ships with a health check configured:

docker inspect bakwas-app | grep -A 10 Health

Docker marks the container healthy after three successful checks (about 90 seconds after start).

Logs

The container logs to stdout/stderr:

docker logs -f bakwas-app
# or
docker-compose logs -f bakwas

Both Gunicorn access logs and Flask application logs appear in the same stream.

Memory footprint

Out of the box:

• Idle: ~60 MB resident. LiteLLM and its provider SDKs are deferred until a summarize request arrives.
• After first summary: ~200 MB. Once loaded, those modules stay resident for the life of the worker.
• Growth: flat afterwards. Python caches modules but your summaries themselves don't leak memory.

If you see continuous growth over time, open an issue with docker stats output.

Verbose LLM debugging

Set DEBUG=True in .env to enable verbose LiteLLM logging. Useful when diagnosing provider-specific issues — the logs will show the exact request/response shapes. Debug mode also disables rate limiting.