Message-Id: <199510060535.BAA03217@wilma.cs.utk.edu>
From: Keith Moore <moore@cs.utk.edu>
To: Robert Elz <kre@munnari.oz.au>
cc: Keith Moore <moore@cs.utk.edu>, Harald.T.Alvestrand@uninett.no,
        drums@cs.utk.edu
Subject: Re: support for Postmaster address 
In-reply-to: Your message of "Fri, 06 Oct 1995 14:32:57 +1000."
             <22867.812953977@munnari.OZ.AU> 
Date: Fri, 06 Oct 1995 01:35:43 -0400
Sender: moore@cs.utk.edu

>     + for any SMTP host that receives or relays mail, 
>     <postmaster@fqdn.for.that.host> and
>     <postmaster@[any-ip-address.of.that.host]> must
>     be valid and reachable addresses,
> 
> No, that, while it might be nice, isn't practical really either.
> 
> Consider a stateless mail relay (no stable storage), that has
> an internet (ethernet, anything) on one side, and x.25 (or
> something) on the other side, and whose whole purpose in life
> is to take smtp mail and relay it to X.400, (and vice versa).
> For now assume there is some point to doing this...
[...] 
> Does this box really have to have a postmaster?

I say yes.  *especially* if its a gateway.

(I think there really are boxes like you describe, that sit 
on the outside of a firewall, filter out things that look 
like security threats, and route email messages to hosts
inside the firewall, but are stateless.)

This really isn't a problem.  Any box that can route based on 
what's on the right-hand side of an @-sign should be able to 
recognize its own address there and if it matches, route on 
the local-part if the local-part is postmaster.  There's a 
small bit of configuration -- it has to know where to forward 
postmaster mail -- but I don't think this adds much to the 
existing burden of having to know a DNS server, a default router, 
IP address, netmask, etc.

> Before we decide on rules for having postmaster support on
> everything in the known universe, we should probably decide on
> just what benefit we expect to achieve from that - ie: make a
> statement on why postmaster support is required, beyond "the rules
> say so".  Knowing just what we are achieving will make it much
> easier to decide exactly what we have to do to achieve it.

Okay.  In my case I want Postmaster to work when I'm trying to
track down the source of a mail problem.  After a point, all
I can do is send a copy of the failed message to the Postmaster
at each of the sites that look like they might be responsible
and say "would you please look at this?".  Often when I do that
I find that the Postmaster address doesn't work at about
half of the sites that claim to have handled the mail.

> While we're doing that, do bear in mind that as mail people, we
> tend to assume that postmaster is the beginning and end of all
> problem resolutions - it's not.  Depending on the problem, assuming
> some postmaster somewhere can help is not necessarily wise.
> Eg: if a host isn't properly handling mail it is being sent
> because of an MX aimed at it, asking its postmaster might be
> useful, or might not.  The correct person to ask there is the
> person who made the MX point that way - they will know if the
> MX is corect, in which case they can get the responsible person
> to fix the broken mailer, or if the MX was just an accident.
> That person's addr is found in the the SOA record of the DNS zone
> in which the MX record was found.   That's where the initial
> enquiry should be directed in this case.

Personally, I do contact the DNS administrator if it looks like
the problem is due to either a broken MX record or an MX record
that's out-of-sync with the MTA it points to.  But quite often the
problem is obviously not a DNS problem -- someone's got his 
mail in a forwarding loop, an address isn't being rewritten
correctly, an MTA doesn't know its FQDN, a disk is full, 
a delivery program core-dumps, etc.  For those cases the Postmaster
address is the most likely way to reach someone who can fix the
problem.

Even if MTAs were better at detecting and reporting local errors,
we would still need Postmaster for exceptional cases.

This is how I came up with the "any host that originates or 
relays mail" criteria -- because only the hosts that originate 
or relay (or gateway) a message will need to be involved when 
tracking down a problem.

Keith