Don't Panic: Jaeger

Jaeger is where the scattered bits of one request are persuaded to sit at the same table. In a distributed system, a request can visit a gateway, a catalog service, a payment service, and a database. Logs report their individual gossip. A trace joins the timed pieces into one operation path.

The important unit is a span, which records one logical piece of work. Related spans form a trace. Trace context carries the trace identity across service boundaries, rather like a luggage label that must survive several enthusiastic baggage handlers. If that propagation breaks, the downstream work can appear in a different trace. Jaeger cannot recreate a relationship that was never recorded.

This means Jaeger is not an all-seeing machine. Applications need instrumentation to create spans, and the course uses OpenTelemetry for that job. Jaeger receives the resulting trace data, stores it, and makes it available through query services and the UI. The useful mental model is: instrumentation creates evidence; Jaeger receives, retains, and exposes it.

The backend has roles because the evidence has somewhere to travel. The collector receives trace data and writes storage. The query role reads storage and serves the UI. An all-in-one process is useful for development and testing, but its in-memory data disappears at restart. A larger deployment can split collector and query roles, use external storage, and scale reading separately from writing. Kafka can buffer collection before ingesters write storage, which buys decoupling and also buys another thing to operate. The universe does enjoy invoices.

A trace is the start of an investigation, not its verdict. Find a representative retained request. Check whether it is complete. Follow parent-child timing, errors, attributes, and events. A long span shows observed elapsed time, not root cause. One sampled trace does not show how often something happens. Metrics, logs, profiles, dependencies, and application behavior still have jobs to do.

Read the Intro for the architecture and tradeoffs. Use Slides for the data path and the configuration choices. Keep the Cheatsheet nearby when investigating traces, sampling, storage, security, and evidence boundaries. Then use the practice session to generate a trace and see how the pieces behave when they are not merely diagrams.

Don't Panic: Jaeger

Relevant careers