Too messages in Hold folder when spamassassin is activated in MailScanner

[Glenn Steen]

On 28/11/2007, Michael Mansour <micoots at yahoo.com> wrote:
> Hi,
>
> Glenn Steen <glenn.steen at gmail.com> wrote:
>  On 27/11/2007, Israel Garcia wrote:
> > On Nov 27, 2007 4:36 PM, Glenn Steen wrote:
> >
> > >
> > > On 27/11/2007, Israel Garcia wrote:
>
> > > > On Nov 27, 2007 4:01 PM, UxBoD wrote:
> > > >
> > > > > So when disabling RBLs the problem disappears.
> > > >
> > > > No, the problem disapear when I turn off spamassassin completily in
> > > > mailscanner config file. Sometimes I enable Spamassassin check but
> > disable
> > > > razor2, DCC and pyzor in spam.assassin.prefs.conf and it work better..
> > BUT,
> > > > if I fully enalbe spamassassin with DCC, razor2 and pyzor the load
> > average
> > > > begins to increase and thousands of mails begins comes to HOLD folder
> ..
> > Do
> > > > you think it's time to split the load in two servers? I mean something
> > like
> > > > this:
> > > > Router
> > > > NAT rules to SMTP farm serves.
> > > > I
> > > > I
> > > > ______________________________
> > > > I I
> > > > Server2 SMTP Server1 SMTP
> > > > Mailscanner/sa Mailscanner/sa
> > >
> > > If the bottleneck is network-related/lookups.... this will buy you
> > > nothing, unfortunately.
> > I know, but how am I sure this is the cause of my problem?
> >
> :-) .... when you remove some load and the problem goes away:-):-).
> How much memory/swap is used when these messages get queued and not
> processed?
>
> I'm currently trouble-shooting a problem which does the same thing and have
> tracked it down to the memory running out ie. mostly swap being used while
> CPU's aren't (load average goes to 12 and 13 because of the heavy swap
> usage).
>
> I'm organising more memory for the server at the moment, which should
> resolve the issue.
>
> Regards,
>
> Michael.

Good point. On the other hand, you could also look at a) Why does the
memory run out, and b) what can you do about it.
If your MailScanner children eat up too much, one thing could be to
lower the amount of children. If the childrens workers hang around
waiting for slow responses, you might need look more closely on why
that is... and correct that (in this case more RAM could be a very
good aid, but still a band.aid where an amputation would be
better;-):-).

Anyway, I'm sure you know where to look, what to do:).

Cheers
-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se