Mastering Distributed Tracing

Traffic labeling

At a high level, adding metadata to request context and propagating it through the call graph is a way of partitioning the overall traffic to the application along a number of dimensions. As an example, if we label each external request with the type of company product it represents (for Google it could be Gmail, Docs, YouTube, and so on, and for Uber it could be Ridesharing, Uber Eats, Uber Bikes, and so on) and propagate it in the metadata, then we can get a pretty accurate picture of how much traffic served by a data center is attributed to each product line. Strictly speaking, the Pivot Tracing and LDFI techniques I discussed earlier can also be considered as partitioning of the traffic, but the values they pass through metadata ...