Distributed tracing best practices with Jaeger

1. Instrumenting #

Readable traces.

Rules #

• OpenTelemetry everywhere (SDK or auto-instrumentation), never a proprietary client.
• A stable service.name and resource attributes (service.version, deployment.environment).
• Low-cardinality span names (GET /commandes/{id}, not the full URL).
• Business attributes useful for diagnosis, without personal data.
• W3C propagation on all services, including message queues.

2. Operating #

Cost and reliability.

Settings #

• Tail sampling that keeps errors and slow traces.
• 7 to 14 days of retention, daily indices.
• Kafka as a buffer beyond a few thousand spans per second.
• Alerts on refused or unexported spans.
• Links from logs (trace_id) and metrics (exemplars) to Jaeger.

Reference diagram and quick check #

Checklist.

Checklist #

1. OpenTelemetry SDK and W3C propagation
2. Jaeger v2 with persistent storage
3. Sampling defined
4. SPM enabled
5. Retention and index cleanup
6. Collector monitoring