Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
Subject: Re: ETRN too fast!
From: Patrick Bihan-Faou (patrickmindstep.com)
Date: Fri Sep 29 2000 - 17:44:44 CDT

> > OK so you DON'T start a new session, just use the existent one
> > to do your sends with.
> With a system like postfix, that uses a completely different set
> of commands and programs to handle inbound as opposed to outbound,
> that is completely and totally impossible by any mechanism I know of.
> Fundamentally, ATRN simply *DOES NOT WORK* with non-monolithic
> MTAs. If you disagree, then I want to see some code from you that
> proves your point.
> If you don't have or can't create code to demonstrate the point,
> then at the very least you don't have any grounds to be arguing this
> matter, and you may well have proven my counter-point.

One way you could do it is by splitting the part accepting the connection
and the actual SMTP server. The "connection manager" that terminates the TCP
connection could just forward the request to a SMTP server using pipes or
other IPC mechanism. Then as the SMTP server realizes that it is not capable
of handling the request (as in ATRN), just instruct the connection manager
to use another daemon to handle the rest of the job.

Now I am not saying that this is easy and/or possible with the current
architecture of postfix. I simply don't know enough of postfix to make such
a claim. It is just to illustrate the fact that not having a monolithic
server does not mean that you can not implement something like ATRN, where
you need to share a single TCP connection with potentially several
independent pieces of code. Also such an approach, which relies on the fact
that ultimately on unix, a fd is a fd is a fd may not be portable to other
environments like windows.