Re: virus scanning
On Mon, Feb 16, 2004 at 03:01:41PM +1100, Russell Coker wrote:
> > > > Hey, I know all about it. Even when people don't do it entirely
> > > > wrongly, they can still screw up. Various yahoo.tld mail servers don't
> > > > reject invalid recipients at RCPT stage but at DATA. (I highly doubt
> > >
> > > Unless you're doing call-outs then rejecting at the DATA stage is OK.
> > I am doing callouts, but they'd still be getting stuck in my queues even
> > if I wasn't. They'd be getting stuck with all the other stuff that callouts
> > avoid, though.
> If a 550 is sent then it should not be stuck in your queues, it should be
No, normal spam junk gets stuck with double bounces... the mail passes the
callout in the RCPT stage here, my server bounces the message on to a faked
address over there, and then their server responds with a bounce which gets
stuck in my queue as a double bounce. :/
> > > > > If so why not just have a SMTP proxy on the MX secondary which passes
> > > > > all data through to the primary if it's available, and sends it
> > > > > locally for queuing otherwise?
> > > >
> > > > Ahm. That's basically what a secondary MX usually does, you know. :)
> > >
> > > No. Secondary MX's usually receive the mail, queue it, and then send it
> > > on if possible. They don't just open a TCP connection to the primary and
> > > pipe the data through unchanged.
> > There's not much negative difference (a small delay, a small resource
> > usage), and the advantage of not bothering the primary MX is lost, so I
> > don't see much reason to prefer that.
> The negative difference is that it forces you to have the same
> configuration in terms of valid recipients and virus/spam filters for both
> the primary and secondary
Which is really not much of a problem. The only difference my MXs usually
have is the differently auto-trained Bayesian classifier in SA.
2. That which causes joy or happiness.