Maybe I've missed something, but the good practice is to forward stdout and stderr to the log files instead of explicitly using only files. It's not flexible, we can't use an aggregation, and it's not effective in development mode. Regarding Docker context in this diff, as I know we are switching to nix and should consider this when making such changes.
It's a good idea to make those discussions in Linear so everyone on the team can read the context about this. Personally, I think we should use a flag or parameter in the configuration file to use the file logging like:

log-files-dir=/logs

And if this parameter is defined we are logging in to the files instead of stdout and stderr through glog.

Agree with @max on most points. Some thoughts:

1. I think we should be using stdout / stderr for logs because it integrates with other things well. For instance, to monitor keyserver logs I use docker compose logs --follow --timestamps to see the logs from stdout / stderr
2. I wouldn't worry about Nix for production deployments quite yet
3. Agree Linear is a good place for discussion, probably a bit better than chat since it's easier to search and reference later
4. If we do decide to have a directory like this, I agree it would be good to make it runtime configurable as @max suggested
5. Last piece, I think generally /var/log is the place to put logs on a UNIX-y system