As I mentioned in my post, that's the one case where I find value in temporary turning on logging, but having old logs around is awfully costly for the off chance that they help.
> tracing individual resources/requests through a bunch of microservices to better understand some unique situation or behaviour,
That's request tracing, which is different. Just storing a request trace is useful, but that's not what I mean by logging. I'm talking about server application logs.
> identifying whether a particular code path is actually in use and if so how heavily
Again that would be request tracing, not application logging.
> They're particularly valuable for production issues. If something has gone wrong and alerts are pinging around, I appreciate the chance to figure out what it was from what the logs are telling me, rather than having to turn the logs on and try to trigger the same problem again deliberately in a production environment!
If the problem doesn't happen again after you turn on logging, then the incident is over isn't it? The customer is no longer affected. If they are affected, then your logs will keep showing it.