Re: [exim] Retry on 550 errors

On Wed, Dec 06, 2006 at 02:34:27PM +0100, Jakob Hirsch wrote:
> Quoting Chris Lightfoot:
>
> > Yes, it would be nice if a 5xx message could always be
> > taken to mean ``there is no point in retrying
> > automatically'', but that's not a description of the world
> > as it currently is.
>
> True, but what do you thing Exim should do about it?
> Treat 5xx always as temporary? I doubt that's what you want.
> Treat 5xx as temporary whenever the word "quota" appears in the reply
> text? This is feasible, but do we really want to start interpreting
> reply texts which are explicitely intended to be read by humans (and do
> not necessarily contain useful information)? I doubt that, too.

I certainly wouldn't advocate treating all 550s as
transient errors! The issue came up because of Marc
Perkel's question about how to treat 550s as temporary in
*specific* cases which can somehow be identified be the
MTA that receives the error.

In the over-quota case I'm describing it (conveniently)
turned out that the sites that showed the problematic
behaviour also sent RFC1894 structured error codes, so it
was easy to discover when an over-quota condition resulted
in a permanent error condition with small risk of false
positives (though I did this by handling bounces rather
than at SMTP time -- obviously that's not an option if
you're relaying rather than originating the messages).

Considered in general this area is obviously a complete
nightmare (hence my enthusiasm for *not* advertising
temporary error conditions as permanent).

--
``... and the crowning example, the Mongols! You don't get much
scruffier than that. They didn't even live in houses, and they
conquered half the world.'' (Anthony Mayer)