A service needs end-to-end tracing across an API, asynchronous queue processing, and downstream calls. Which design is appropriate?

1. Use one request timestamp as proof that every downstream call belongs to it.
2. Propagate supported trace context and instrument the participating components using appropriate tracing integrations.
3. Disable queue correlation because asynchronous work cannot be observed.
4. Enable only CPU metrics and assume they contain complete causal request traces.