Victor Duchovni:
> On Mon, Apr 13, 2009 at 05:04:33PM -0400, Wietse Venema wrote:
>
> > r...@yingshen.org:
> > > Hi there,
> > >
> > > Recently I saw some dumped core files from the qmgr process. According
> > > to the backtrace log, it was qmgr_job_free() complaining about a
> > > "non-zero recipient count" error.
> > >
> > > My question are what are the possible reasons that caused this? Is
> > > this something related to configuration? My box is running postfix
> > > 2.3.4 on FreeBSD 6.4-RELEASE-p3.
> >
> > Since you have core dumps and executables with symbol tables,
> > perhaps you can upload a gzipped version of these, as that will
> > avoid the need to figure out how to reproduce this. Contact me
> > off-list for this.
> >
> > If you have thrashed the core dumps, just ping me when you have
> > another one.
> >
> > Perhaps it is worthwhile to also include some logfile context
> > concerning the message that qmgr was cleaning up for (the
> > message->queue_id structure member).
>
> It would be especially interesting to know how many recipients it
> had, and how many were processed between the time the message most
> recently entered the active queue and the core dump. Previous delivery
> history may also be relevant.
>
> If this is a bug, and not hardware issues, it is most likely an adverse
> interaction between "refill" (reading more recipients into memory for
> large messages) and "throttling" (moving all mail for a dead destination
> to the "retry" queue). Such an interaction would happen for large messages
> when refill is required just as a delivery failure triggers a destination
> to be throttled.
The "retry" queue was introduced with Postfix 2.4.0, so can't be
part of the problem here. Once we know the queue ID it would be
worthwhile to see all the logging for that message, from the time
it arrives in the queue. Off-list, of course.
Wietse
