Date: Thu, 1 Nov 2007 13:14:52 -0400
From: Andrew Sullivan <ajs@crankycanuck.ca>
To: pgsql-www@postgresql.org
Subject: Re: what is up with the PG mailing lists?
Message-ID: <20071101171452.GT27676@crankycanuck.ca>
References: <25716.1193887595@sss.pgh.pa.us>
	<DADF296033D290EF9634F611@ganymede.hub.org>
	<26669.1193891360@sss.pgh.pa.us> <47299585.7030402@hagander.net>
	<47299957.5020605@postgresql.org> <2968.1193919208@sss.pgh.pa.us>
	<20071101080959.49f3087b@scratch>
	<20071101152333.GM27676@crankycanuck.ca>
	<4729F105.30704@hagander.net> <20071101084752.64ac3936@scratch>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20071101084752.64ac3936@scratch>
User-Agent: Mutt/1.5.9i

On Thu, Nov 01, 2007 at 08:47:52AM -0700, Joshua D. Drake wrote:
> Exactly. It is pretty silly to think that a modern, well engineered
> system will take hours to deliver mail. 

You have a strange idea of "well engineered" for very large mail
systems.

I have news for those on this list who do not operate systems with
thousands of simultaneous users: your assumption that a couple
hundred users scales to several thousands by adding hardware is as
wrong in mailing lists and DNS systems as it would be for newbies
implementing PostgreSQL systems.  It don't work that way, folks.

Keep in mind that the mail server relies on DNS servers that are out
there in ISP-land, and not in your-control land.  Since DNS
lookups can result in soft failures, it's not at all impossible that
mail to your specific mailbox will get delayed by annoying DNS
holdups that have more to do with, for instance, random nasty people
trying to prevent resolution of every site under .org.  Or under
.postgresql.org (I'd be interested to see graphs of query volumes
against that domain, BTW.  Because I bet they're all over the place
in unpredictable ways).

This means that, even though the mail has made it into the relay, it
might not make it out anything like as fast as you think, and there's
more than one system that can easily cause the problem.  Now,
multiply by, say, 1000 simultaneous delivery attempts, and you have
significant load issues that no mail system can solve, because the
fundamental Internet infrastructure is kinda broken that way.

So, without clear outlines of _exactly_ where the problem is, which
means logging of months of headers, putting aside only those messages
that hung up, this is going to amount to nothing more than "you did
x" " no I didn't lalalalala" discussion.  You want this to get
better?  Capture your logs, and let's do some analysis.

A

-- 
Andrew Sullivan  | ajs@crankycanuck.ca
When my information changes, I alter my conclusions.  What do you do sir?
		--attr. John Maynard Keynes