Date: 12 Dec 1996 00:59:14 -0000
Message-ID: <19961212005914.16351.qmail@koobera.math.uic.edu>
From: "D. J. Bernstein" <djb@koobera.math.uic.edu>
To: drums@cs.utk.edu
Subject: Re: Large Site MX document

> qmail may have survived this particular instance of the problem, but
> if there had been multiple simultaneous large site outages, could
> qmail not have tied up all of its SMTP sending threads?

Unlikely. Even a site with as many MXes as AOL takes up only one or two
threads. (Each IP address takes 1 minute to try and is then ignored for
an average of 80 minutes, so 60 dead MXes means an average concurrency
of just 0.75. This grows a bit for very large queues, but it's nowhere
near qmail's default concurrency limit of 20.)

Of course, every thread will be stuck if the entire net is down.

> Such a reading of 1123 is a bit of a stretch.

No. RFC 1123 says explicitly that ``the sender-SMTP MUST be able to try
(and retry) each of the addresses in this list in order, until a
delivery attempt succeeds,'' possibly subject to a configurable limit on
the total number of attempts.

Does ``each ... in order'' allow arbitrary subsequences, or only
prefixes? If a host tries just the last address, is that ``each of the
addresses in order''? How about the last two? How about the first and
last? First, fifth, and twelfth?

There's also the painful old question of what ``delivery attempt
succeeds'' means. There are at least eight obvious possibilities, and
implementors realize immediately that the most obvious possibility is
downright stupid, so MTAs do lots of different things here.

> because "that's the way AOL does it".

Right. As I said, this is pointless if you can't convince Knowles.

---Dan
Put an end to unauthorized mail relaying. http://pobox.com/~djb/qmail.html