I am still at this problem ...
In short: instead of keeping parallel deliveries around maxconcurrency values
qmail "explodes" to the maximum, goes down to about 15-25 deliveries,
stays there for a while and then "explodes" back to the maximum.
On Tue, Oct 05, 1999 at 08:08:37PM +0200, Markus Stumpf wrote:
> To illustrate the behaviour I put a gnuplot plot at
> http://www.lamer.de/maex/creative/software/qmail/deliver-stats.gif
I've spent a few hours debugging qmail-send and doing stress tests
and I *think* I have tracked the problem down:
a) While there are unprocessed messages in the queue, qmail does only few
deliveries (I still don't fully understand why, but I think it's due to
the "management flow").
b) Then comes the time, when all those unprocessed messages are processed,
qmail-send starts as much parallel deliveries as possible.
c) Too bad that few of those deliveries generate bounces. qmail injects
unprocessed messages to the queue and falls back to a)
In my tests I first had problems simulating the behaviour. I'd injected
emails with local "null" deliveries, and qmail always was around maximum
parallel deliveries. Then I did inject messages with two recipients
(one valid null delivery and one address that generated a bounce) and
the problem became evident.
Another test I did was kill -STOP qmail-send with a totally empty queue.
I then injected "good" emails and did a kill -CONT qmail-send. Even
during the "preprocessing phase"qmail used maximum deliveries
immediately and stayed at this value.
So I *think* it's due to the bounce handling.
Does anybody know why qmail behaves like in a) and if there is any way
around it?
Thanks
\Maex
--
SpaceNet GmbH | http://www.Space.Net/ | Yeah, yo mama dresses
Research & Development | mailto:[EMAIL PROTECTED] | you funny and you need
Joseph-Dollinger-Bogen 14 | Tel: +49 (89) 32356-0 | a mouse to delete files
D-80807 Muenchen | Fax: +49 (89) 32356-299 |
