Core answer
Start here. This is the direct spoken answer to practice first.
Why this question matters
When a request crosses APIs, databases, and queues, isolated logs are not enough. This drill matters because the production behavior is shaped by platform configuration, code boundaries, and the operational path the team will use during a real incident.
I would use distributed tracing so one user operation can be followed across the API and its dependencies. HTTP calls would carry trace context, logs would include the trace or operation id, and queue messages would include correlation metadata when work becomes asynchronous. I would avoid putting secrets or sensitive personal data into correlation values because those values spread into logs and telemetry.