Date: Tue, 08 Jul 1997 10:10:24 -0400 (EDT)
From: John C Klensin <klensin@mci.net>
Subject: Re: MX handling and 5xx errors at low priority relays
To: =?ISO-8859-1?Q?Claus_Andr=E9_F=E4rber?= <mail-std-list@faerber.muc.de>
Cc: drums@cs.utk.edu
Reply-to: John C Klensin <klensin@mci.net>
Message-id: <SIMEON.9707081024.G@tp7.Jck.com>
MIME-version: 1.0
Content-type: TEXT/PLAIN; CHARSET=US-ASCII
Priority: NORMAL


On Mon, 07 Jul 1997 18:31:00 +0200 
=?ISO-8859-1?Q?Claus_Andr=E9_F=E4rber?= 
<mail-std-list@faerber.muc.de> wrote:

> Peter Koch <pk@TechFak.Uni-Bielefeld.DE> wrote:
> >
> > domain.org	MX 10	mail.domain.org
> > 		MX 20	mail.relay.org
> > 		MX 30	mail.fallback.org
> >
> > Does a 5xx reply mean permanent failure everywhere or permanent failure
> > at the SMTP server issuing this reply? The latter would imply solution (b).
> > Neither in RFC974 nor in draft-ietf-drums-smtpupd-05.txt I could find
> > an unambiguous answer.
> 
> This really needs clarification then. I think, it should really be  
> treated as permanent failure at ALL responsible hosts, to avoid going  
> through all servers if a mailbox address has become invalid.
>... 
> That's a configuration error. Mail hosts not accepting mail for a  
> certain domain SHOULD NOT be in the MX record for that domain.

I'd like to hear differing opinions on this, if there are 
any, but, in their absence, the doc will be clarified to 
make it clear that "permanent error" (i.e., a 5yz code) 
means "permanent error".  In essence, an MX record appoints 
a mail receiving agent for the target domain.  If the agent 
returns a 5yz code, it is assumed to have sufficient 
knowledge to do that.  Otherwise, it should return 
something else (and "what else" may need further discussion 
-- I haven't analysed all of the cases).

This has a possible side-effect benefit.  One of the little 
abuses floating around the network these days involves 
listing hosts in the DNS as low-priority MX backups without 
asking them (and sometimes contrary to their instructions). 
If the effect of their noticing this and issuing a 5yz code 
is to definitively prevent the target / DNS-controlling 
site from receiving mail, that would tend, I think, to have 
desirable effects while a "just try some other MX host" 
strategy creates one of those "can't hurt to try and see if 
it works" situations.

IMO, for all of its tentative tone and convoluted language, 
RFC 974 is pretty clear about this.   The retry strategy it 
describes has to do with SMTP-TCP connections that don't 
succeed, not with getting 3/4 of the way into a 
successfully-connected SMTP session and then getting a 
failure code.  Anyone object seriously to my writing it 
that way?

    john