On Mon, Nov 25, 2002 at 06:41:10PM +0100, Claudio Jeker wrote:
> On Mon, Nov 25, 2002 at 12:15:08PM -0500, Gene Parks wrote:
> > Claudio,
> > 
> > I get the same thing over here.  Not sure what causes it as I have not
> > investigated it that much.  We don't reload all the often.  But I will
> > see what I can come up with.
> > 
> 
> Could please all people that have a problem with shutting qmail-ldap down
> send the following information:
> OS (uname -a):
> libc version: e.g glibc-2.2.35
> qmail-ldap patch release: (if possible use 20021101)
> 
> I can not reproduce the problem on my test machines:
> OpenBSD 3.1  sparc
> OpenBSD 3.2  i386
> 
> qmail-send starts and stops without a prob via svc -d && svc -u
> 
> I will try to generate a high load and try it then again.
> 

OK, I did some test. I loaded the queue with 5000 messages (not
preprocessed) and delivered to users with "|sleep 20" as their .qmail
file.

As soon as I started qmail (svc -u /var/qmail/service/qmail)
all local concurrency slots where filled in no time.
After a while is stoped qmail (svc -d ...) and got the fameous "qmail-todo
exit asap" and ps ax showed me qmail-todo and his qmail-clean as zombies.
qmail-send was still running becaus the local concurrency use was not yet
0 (e.g. status: local 68/80 remote 0/20 exitasap) as soon as it dropped 0
qmail-send exited and cleand all up.

status: local 0/80 remote 0/20 exitasap
end msg 40386
status: exiting


I now modifed the code so that qmail-todo does not exit but stops
processing of messages. So the zombies disapear and the same behaviour as
with qmail-[rl]spawn applies. Wait until all current deliveries are done
and then exit gracefully.


-- 
:wq Claudio

Reply via email to