Massive queue buildup

[Glenn Steen]

On 15/11/06, Res <res at ausics.net> wrote:
> On Tue, 14 Nov 2006, Chandler, Jay wrote:
>
> > ov 14 20:07:42 brewer MailScanner[37903]: Batch (30 messages) processed
> > in 328.23 seconds
> > Nov 14 20:07:51 brewer MailScanner[38335]: Batch (30 messages) processed
> > in 670.92 seconds
> > Nov 14 20:08:37 brewer MailScanner[38125]: Batch (30 messages) processed
> > in 643.34 seconds
> >
> > That's not good.
> >
>
> No :)
Have to agree here. More below.

> If you nee dit sorted , disable spam assassin, leave spam checks on, but
> disable SA, you'll find it will clear the queue in no time.
Might not be the culprit after all. Well, it still might bet....:-)

> Also do you RBL in MS or MTA? MTA is far better
In one of Jays earlier responses (to Brent Addis) he mentioned doing
3-4 BLs in MS, which he found to be "nothing insane". Well, he just
might be wrong, taken that those would _serialize_, much like most
MTAs but unlike SA, and with some poor choices made on which lists to
check in MS... Voila, bad performance here we come:-). At least a
theory worth exploring;-).
As you all know, the MTA (for early rejection) or SA (for
parallellism) is the place to do this (and possibly one or two in
MS...:-).

Martin has posted his list of SA RBLs he disables (by setting score to
0) a few times, which might be interesting to you Jay, if you go the
SA route.

>
> > Disabled DCC, Razor, and Pyzor, and I'm still seeing batch times in the
> > same general range.  I've got RAM to burn, so I kicked up the number of
> > children to 30, and I'm still seeing the same batch times, but the queue
> > is decrementing.  Finally, I built up a box that's blazingly fast in
> > processors, and woefully short of RAM, and put that ahead of this one in
> > the mailserver precedence list.  We'll see how it handles tomorrow.
> >
> > Thanks for the assist, folks-- I appreciate it.
-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se