|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
Re: Flushing mailq
From: Victor Duchovni (Victor.Duchovni
morganstanley.com)
Date: Tue Sep 02 2003 - 00:16:40 CDT
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Vivek Khera wrote:
> DS> Well, yes, it's only so much at a time...
>
> No, I mean it finishes the queue, but there are still old messages
> there. Wait a while, type postfix flush again, and they sit there.
> Eventually they go out. Postfix flush runs the queue -- it doens't
> force it to process every message in the queue.
On busy systems there is a race between the flush daemon moving messages
into the incoming queue and the cleanup daemon waking up the queue manager
(no FLUSH_DEAD trigger, so dead hosts and deferred transports are obeyed).
This means that multiple "ETRN/sendmail -qSsite/postqueue -s site" or
"sendmail -q/postqueue -f/postfix flush" commands may be needed to flush
the queue if new mail is arriving. If one first shuts down "smtpd" and
"pickup" and disables the periodic queue manager wakeup trigger (edit
master.cf and reload), flush becomes a single pass operation. In practice
the required master.cf edits are a pain...
Wietse has received a patch that implements a more precise flush protocol.
The patch still needs some internal polish before it can see the light of
day, but the basic mechanism is IMHO reasonably sound. Hopefully the patch
will be integrated into 2.1 or (more likely?) an early post 2.1 snapshot.
On the other hand Wietse might decide to do a larger queue re-architecture
which would make the patch obsolete, this would naturally take longer :-(
Bottom line:
- Flush (full or site) is reliable when there is no other mail entering
the incoming queue (no smtpd, qmqpd or pickup) and there is no periodic
wakeup trigger for the queue manager in master.cf
- Flush (full or site) can be partial otherwise when new incoming mail or
the periodic wakeup trigger from the master daemon cause an incoming queue
scan in the middle of the flush. In this case some flushed messages to
dead hosts or deferred transports may be re-deferred without entering the
active queue.
--
Viktor.
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]