Observability with Grafana

Reference dashboards for traces, metrics and logs

Demo

Grafana Observability

Reference implementation

The reference implementation of Grafana observability lives in the otel/grafana/ directory: the OpenTelemetry Collector configuration (otel-config.yml), Prometheus scrape config (prometheus.yml), Tempo trace storage (tempo.yml) and Grafana datasources (datasources.yml).

These settings are consumed by the compose-grafana.yml file, which starts the application with the full stack — OpenTelemetry Collector (OTLP on :4317/:4318), Prometheus, Tempo, Loki and Grafana.

Run it and open the dashboards (see Running the Application):

docker compose -f compose-grafana.yml up --build

Grafana UI: http://localhost:3000 — default credentials admin / admin
The app's API: http://localhost:18000/docs

What to look at: per-request HTTP spans, nested LangChain LLM calls with token usage, SQLAlchemy statements, and application logs in Loki. What gets instrumented and why: the Observability guide.

Note: this implementation is just a reference. For a real production scenario, take privacy and security settings into account — authentication, authorization and encryption.