From NDS-842, all baselines terminate gracefully except for rm3.
Perhaps something about the length of the queue is being misread, or perhaps length can be zero, and the queue can still be somehow not empty?
Currently, I've hacked a workaround such that 10 minutes after the last command has successfully returned. This is probably fragile, as it assumes a success at least once every 10 minutes.
More investigation is needed into this, if it is deemed a priority.
This ticket is complete when rm3 terminates gracefully without the hack described above, and the hack has been removed.