There was discussion some while back about splitting up the handling of
queue runs when the queue was large, and split into multiple
directories. I've just been looking at this, and it would be quite
straightforward to make the following change:
If split_spool_directory is set, a queue runner process proceeds by
scanning each sub-directory and dealing with all its messages before
proceeding to the next sub-directory. The order in which
sub-directories are handled is undefined.
This would spread out the timing of the sub-directory scanning,
I see no reason not to do this as standard - I can't think why anyone
would want to turn it off.
BUT: There are two problems.
(1) Exim allows you to start a queue run from a specific message, e.g.
exim -q 12X0I6-0000i3-0
(and indeed to stop it at a specific message).
I wonder if anybody has ever actually used this facility? It obviously
only applies to manually started queue runs. I can't remember why it
got implemented - it's been in Exim for a very long time.
I would like to abolish this feature. Are there any objections?
(2) There is an option queue_run_in_order, which runs the queue in
lexical message id order (order of arrival). This option was requested
at some point, so I guess it is useful for some people, but this must
be a minority.
Proposal: queue_run_in_order applies only within a single directory of
messages when split_spool_directory is set. If people really do want
to run the queue in absolute order, then they have to do without
split_spool_directory.
Any objections?
--
Philip Hazel University of Cambridge Computing Service,
ph10@??? Cambridge, England. Phone: +44 1223 334714.
