[Top] [All Lists]

Re: Last Call: draft-klensin-rfc2821bis

2008-03-29 08:11:43
One of the problems I have seen first-hand is "disappearing" mail.  
Example: A webserver sends outbound email directly, but doesn't want  
to receive inbound email. The hostname leaks and mail gets sent to  
that address, based  on the A(AAA) record. The mail is "received", but  
disappears into some never-seen /var file. In that case, the sender  
never suspects that anything is amiss; it would be much better if the  
sender got an immediate "sorry, that domain name doesn't support email  
service" error.

Even if you turn off sendmail, as someone has pointed out earlier, the  
sending MTA will retry for several days until giving up, thus delaying  
error notification that would be immediate otherwise in this  
particular case.

You can obviously turn off sendmail local delivery, but many of the  
standard web hosting (cPanel and kin, but also the standard RH mail  
setup) arrangements don't make it particularly easy to have outbound- 
only sendmail.

Thus, disabling AAAA checking seems to provide much cleaner error  
behavior. The 'MX 0' proposals would achieve some of the same results,  
but removing the AAAA lookup is default-safe, rather than requiring  
operator action. It's too late to change the A behavior, but there  
doesn't seem to be a reason to perpetuate this violation of the  
principle of least surprise.


On Mar 29, 2008, at 10:34 AM, Theodore Tso wrote:
On Sat, Mar 29, 2008 at 10:16:10AM -0400, Keith Moore wrote:
I think it is time to put an end to specious arguments.

These standards get used for decades.  I don't think it's  
appropriate to
cripple them because of some arrangement that happens to exist now  
a few dysfunctional DNS providers.  Providers will get more  
flexible as
the need becomes apparent, and domain owners who have problems with
their DNS providers can change providers.  It's not difficult.

So I must be missing something, probably because I deleted without
reading closely enough one of the earlier messages on this thread.
But please indulge me --- exactly what is the benefit of deprecating
the "A" fallback, and/or not doing a lookup on the AAAA record if the
MX record doesn't exist?  Is it the load on the nameservers that
people would believe would be reduced if we didn't do this?  Is that
really a problem?  Or is it something else?

                                      - Ted
IETF mailing list

IETF mailing list