mailscanner, queue & nfs

[Glenn Steen]

On 17/03/2008, Alessandro Dentella <sandro at e-den.it> wrote:
> On Mon, Mar 17, 2008 at 09:08:22PM +0100, Hugo van der Kooij wrote:
>  > -----BEGIN PGP SIGNED MESSAGE-----
>  > Hash: SHA1
>  >
>  > Alessandro Dentella wrote:
>  > | Hi,
>  > |
>  > |   can I keep the postfix queues on NFS so that 2 concurrent host run
>  > |   mailscanner on the same queue?
>  > |
>  > |   Which kind of problems... am I looking for?
>  >
>  > How are you managing locks?
>  >
>  > If you have two hosts then distribute the mail with SMTP on both hosts.
>  > That way you can lose one system and still get something done.
>  >
>  > That NFS ploy is just looking for more trouble instead of solving an issue.
>
>
> I'm not yet doing this, so I'm not managing locks!
>  This setup would olny be an intermediate way to migrate a system that is
>  always under too hight load to a newer one (same hw, really).
I'm pretty certain it is the "wrong" way to go. Much better to make
the new one act as a GW to the old one, disable MS on the old one,
config/enable MS on the new one... Less risks, less time spent on
solving "the wrong" type of problems.

>  I'like to understand how different blocks are impacting on load so I wanted
>  to isolate mailscanner, while at it I also wandered if it would be possible
>  to go in parallel. (the same box has also imap/pop and we receive around
>  150.000 mail a day)
Separation is good, yes, but having one or two GWs in front of your
mail store is a much simpler thing to handle... If the GWs are too
loaded... And you can't seem to find a reasonable "easyblock", then
... just add more GWs:-).


>
>  And what about haveing (temporarily) postfix on a box and mailscanner on a
>  different box that works on an NFS exported spool directory.
I always shy away from this type of thing... NFS isn't exactly the
epitome of performance (other than compared to other "network
filesystems":-), and it isn't really that likely that your problem is
with PF taking too much resources from MS... More likely that all the
other things you might have there (IMAP software, webmail,
database(s)...) are the problem.
All "solved" if you make the new host a GW to the old one.

>  Would that be again looking for troubles or not (since there should not be
>  concurrency between different MailScanner processes.)
Well, I can't really see that you'd be gaining anything by it.

>  My understanding (please correct me if I'm wrong) is that mailScanner only
>  acts on mails in the queue with HOLD flag and postfix won't touch those
>  files. My guess is that non lock problem shou arise, but I'm /really/ not en
>  expert un this issue.
Quite true, but ... why make it that messy?
Simpler to set things up on the new one (PF and MS in "relay mode":-),
then just futs your MX records to "slide" it in before the old one,
disable MS on the old one and start normal PF (without the HOLD header
check), perhaps set that one to use the GW as a smart host ... and
you're set...
That's probably what I'd do anyway:-).

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