I read around, and it seems that the consensus is it should only be locked
during a multi-thread/multi-process update. However I encountered the error in
a python script (single proc/single thread) that was the only reader/writer to
the database. It seems that I forgot to con.commit() in a loop that was doing
thousands of update statements. It seems to me that there is some limit to how
many pending statements there can be before encountering this error. If that is
true, and not a bug, then maybe add it to the documentation somewhere? I didn't
find any mention of "too many pending statements" as a cause anywhere. Adding a
commit() periodically in the loop made it go away.
