For services managed by systemd, logging to stdout/stderr is captured in the 'journal', a kind of log database which captures relevant metadata like timestamps and process IDs along with the log messages. Logs are compressed, automatically rotated, and older data automatically discarded to limit the space used. It also means that uncaught exceptions, including errors during startup, are recorded in the same place as explicit calls to log.info() etc.
I'm thinking of one day using this instead of our text based log files. For now, this just directs the logging to the journal (via stderr) as well as the familiar log files, so we can see how it works for us.
Deployed on xcaltst, logs viewed with cal-logs alias.