Another power failure, that took over four hours to be restored. This time, I could handle the situation myself – but not before 20:00. This time, because syslog did run, I could pinpoint the time of failure to 11:27. On boot, I could now locate the cause of syslog’s failure to start: I simply had a typo in the startup procedure. Now that is settled – and I found it make take a few minutes before lines are actually written to disk. It may be a RMS buffer that is flushed when full, SYSLOGD may buffer itself. This means that the last message before power failure may be written a few minutes before the actual event.
Not that this matters a lot. But in case something is terribly wrong, a direct write might be useful.