I, and my peers, look at historical logs daily. Our customer response team is often looking at them on a minute-by-minute basis, and sets up alerts on them. Ou

I, and my peers, look at historical logs daily. Our customer response team is often looking at them on a minute-by-minute basis, and sets up alerts on them. Our logs tell me the state of the application when the error occurred without having to reproduce it. They also tell me how many customers are having the issue, so we can set the severity of the incident at the appropriate level and reach out to those customers proactively if needed. Sure, we might be able to do this with metrics or tracing instead, but that's often just duplicating information we're already logging. For example, by adding a correlation ID in my log metadata, I suddenly have end-to-end request tracing. It's not as pretty as what New Relic can produce, but it definitely works.