Why would logging not help in an async flow? Link log entries with request IDs and consider structured logs. I analyze logs in massively concurrent systems all the time.