I think there's room for both. Tracing is usually sampled, or much more time limited, and aggregation tooling isn't nearly as good compared to logs with most p

I think there's room for both. Tracing is usually sampled, or much more time limited, and aggregation tooling isn't nearly as good compared to logs with most providers. Much better for deep dives into a narrow context. For a single request drilldown or aggregated performance/error metrics, tracing every time. Structured logging tends to give you rich search against broader contexts (e.g. long spans of time, many requests in aggregate, random non performance/error attributes you may care about) without the same concerns of data cardinality. I felt nowhere near the same power with tracing tooling for random broad context as I've had with e.g. Splunk for logs. That's not to say you can't somehow have the best of both worlds, but I haven't seen it in the ecosystem just yet.