What is Jaeger, and what is a good step-by-step workflow for finding a slow component with it?

Jaeger is an open-source distributed tracing backend and UI; the workflow is to trigger a request, find its trace, open the individual spans, and identify the component that consumed the time.

* The four-step hunt for the slow component. *

1. Trigger a request. Place an order in the shop, so a fresh, known trace exists.
2. Search for the trace. Pick the service (e.g. frontend), optionally the operation, and a time window; filter by attributes like error=true or by minimum duration.
3. Open the trace and its spans. The timeline shows every span as a bar, nested under its parent. Click one to see its attributes, logs and status.
4. Identify the slow component. Look for the longest bar that has no long children of its own: that span's time was spent in that service itself, not waiting on someone else.

Jaeger was built at Uber and is a graduated CNCF project (the same maturity level as Kubernetes and Prometheus). It stores traces and makes them searchable, and it can also show a service map and aggregated per-service metrics.

Go deeper:

• Jaeger documentation — architecture, UI and deployment.