distributed-tracing

Deploy Jaeger using the Kubernetes Operator with Elasticsearch storage, or run the all-in-one Docker image exposing ports 16686 for the UI and 14268 for the collector. Then instrument your services with OpenTelemetry exporters pointing at the collector endpoint.

Create a TracerProvider with a service name resource, add a BatchSpanProcessor with a JaegerExporter, and call FlaskInstrumentor().instrument_app(app). Use tracer.start_as_current_span to create custom spans with attributes around business logic and database calls.

Jaeger provides a dedicated UI with service dependency graphs and supports Elasticsearch storage. Tempo integrates with Grafana and uses object storage like S3, accepting Jaeger, Zipkin, and OTLP protocols, making it cost-effective for high trace volumes.

Check the collector endpoint configuration, verify network connectivity between the application and Jaeger agent, review sampling configuration since low rates drop most traces, and inspect application logs for exporter errors.

Sample 1-10% of traces in production to keep tracing overhead under 1% CPU impact. Use probabilistic sampling with a param like 0.01, rate-limiting sampling for a fixed traces-per-second cap, or parent-based sampling for deterministic decisions.