ietf-smtp
[Top] [All Lists]

Re: Proposed change for RFC0974

1998-11-06 08:53:04
At 10:02 AM 11/6/98 -0500, Valdis(_dot_)Kletnieks(_at_)vt(_dot_)edu wrote:
On Fri, 06 Nov 1998 22:18:45 +1100, you said:
Now, it just so happens that many of these offending sites don't HAVE
MX entries, so we attempt delivery to their A record.  I think the point
that was being made was that if they *had* MX entries with a backup at
their ISP, that they couldn't use ETRN to suck down their mail unless my
machine properly handled rollover to the backup MX entry, because their
machine wouldn't know to ETRN to mine, only to the ISP's server.

Are there, in fact, any systems with significant market share that manage
to get MX record handling wrong?

My take on this was entirely different.  I think the points being raised
are:

1) According to Alexey's reading (and mine) of the rfc, you MUST read
the first item of a MX record, and MAY go to the other MX records, in
listed order of preference if a given MX record fails.

2) There are quite a few sites out there that have an A record and no
MX record.

3) DJB feels that having a [static] A record for a site that is only ever
intermittantly connected is a waste.  [For the record, I think this is a
bit silly too.  Using DHCP from a dynamic pool would be better if a
proper solution such as ETRN or ATRN is provided.]

From my POV, if you have a list of 500 people, and 5% of those addresses
are hosts that have tenuous connections to the net, that means for 25
connections, you will try and connect to the first MX record or A record.  
In the standard case for both, you will fail the connection.  For the
MX record case, you may decide to go to MX record #2, which if it is the
ISP, has a far better chance of being reachable.

So, my take on this entire thing is that Alexey thinks it would be a real
good idea if we did something about this.  The A record case is one that
we really can't do anything more about than hand-slapping ("you should
really set up MX records for your mail server!").  For the MX record case,
perhaps all Alexey is asking for is stronger wording to suggest:

that if a given host will normally not be connected to the net that it
not appear in the MX records, but have a method for contacting the node
listed as the "most preferable" host by MX record, and getting the mail
sent from there.

ideas?

regards,
Jack
-------------------------------------------------
Jack De Winter - Wildbear Consulting, Inc.
(519) 884-4498          http://www.wildbear.on.ca/homepages/jack/

<Prev in Thread] Current Thread [Next in Thread>