The logging code keeps records in memory until a slot is closed. If there is limited activity in a database slots aren't closed regularly, so the log server periodically flushes the active slot.

The flush currently happens once per second (along with log archiving and preallocation).

That can lead to losing up to a second worth of log records on an unclean shutdown. I think it would be better if that period was shorter. Perhaps the loop could be restructured so that the active slot is flushed 10 times per second, but prealloc and archive still happen once per second.