Subject: Re: Postfix
To: Greg A. Woods <>
From: Pete Naylor <>
List: current-users
Date: 08/15/2000 01:36:40
Greg A. Woods wrote...

> But what I really meant was to conclude from what you'd said directly --
> i.e. you said that postfix was still not "full featured", so, therefore
> one can conclude that it's "smaller" and "simpler" in concepts.

Now I understand your point - thanks.  Not sure that I really agree
though, given the number of different components of postfix (as you can
tell, I'm not of the opinion that more daemons equates to more security).

> I.e. it's not cluttered (yet) with a whole lot of complexity that'll
> only be used by a very few people.  This is what's really important to
> me (outside of its own claims to fame, that is).

It is cluttered by a bunch of separate processes though.

I don't believe that sendmail is a good MTA, or even a good piece of
software to use as a basic null client mailer - _but_ it has been present
in the distribution for a long time now (and in other OSs) and many people
are familiar and comfortable with making it do the basic stuff that they
want it to do.  People basically accept the sendmail
source/build/binary/config when they choose NetBSD or a number of other
OSs for their host.  I can't really see replacing it now based on the size
of the binary or the complexity that the software offers if desired.  It
works just fine in the required simplistic null client configuration that
a lot of people want, and I'm sure postfix does too (as well as the other
alternatives which are not being "evaluated" by the developers).  But why
confuse the situation and make people uncomfortable by pulling the
sendmail rug out from under them?

> > > simpler,
> > 
> > No, it's not.  Anything represented by a jumble of a dozen daemons is
> > likely to be daunting to a newbie - it's certainly not going to be easy
>   ^^^^^^
> In your opinion.  and you exaggerate greatly!  :-)

In my opinion, yes.  It's not my intention to cut postfix down or compare
it against other MTAs, because there's no point.  It will suit some
people, and other MTAs will be better for others - I am glad there is a
choice.  However, I do know that sendmail is familiar to many (indeed,
many folks count on it being a part of the OS distribution) and that it
does the basic job required of it in the distribution just fine.  In my
opinion, postfix having so many different components will be odd and
disconcerting to people who are familiar with sendmail, and will look like
a much more complex beast to a newbie than a single binary and associated
config file.

> [[ There are just under two dozen programs in /usr/libexec/postfix to be
> sure, but there are normally only three daemons running.... ]]

How would someone who is familiar with sendmail and has no requirement for
anything different/better know that without spending time learning about
postfix?  And all those people should allocate that time why?  A few
NetBSD developers who happen to like postfix decided that they're right
and everybody else is wrong - that's why.

> Postfix is extremely well documented, and I don't mean that in the sense
> that it too has a 1500 lb. -- errr pg. book to tell you all about it!

So are other MTAs I've used - including sendmail.  I'll be glad of this if
I decide that postfix fits my needs someday, but right now it's irrelevant.

> > Most people don't need to do much sendmail configuration.  For a simple
> > node, it doesn't need much help - and a as distributed would
> > work pretty much out of the box.
> I've only seen one mailer that works ideally for any small leaf node or
> minor MX handler without any configuration tweaking, and it's not
> sendmail or postfix!  ;-)

A suitable sendmail configuration is entirely possible - proven by the
people who built the for various OS distributions.  That's not
ideal, but apparently it works fine for a lot of folks who are happy with
it as it stands.

You may be talking about exim.  If the object of this postfix "evaluation"
is to come up with a small and simple mailer to replace sendmail in the
NetBSD distribution, why aren't you suggesting that Exim should be used
instead of postfix?

The right answer is to make postfix, exim, etc available as packages, and
keep the base distribution basic and familiar to users by leaving sendmail
alone (though making it easily replaced such as Sun have done with recent
Solaris releases is a bonus).

> > Or even postfix - people are free to get it and install if they so desire.

> I think we all knew that right from the start.....   :-)

If everybody is aware of that, and postfix is so obviously wonderful, why
not just make it available optionally and let people choose it as desired
instead of forcing it upon them?

> > Replacing sendmail with postfix in the base distribution is senseless.
> Why? 

The operation and administration of the software distributed as part of
NetBSD shouldn't change without good reason, because it will upset a bunch
of people and cause them discomfort in their day to day use of NetBSD.  So
the onus is upon those who favour replacing sendmail with postfix, to give
good solid reasons as to why that replacement should be forced upon all
NetBSD users.  I suggest that leaving sendmail (and only sendmail) in the
distribution with a simple configuration while making it easy for users to
choose a more advanced (substitute secure, configurable, flexible, fast,
powerful - whatever) MTA as their needs dictate and plug it in to the OS.

> You've given no solid technical arguments yet that I can recall,

They're not necessary, because I'm not trying to promote postfix or cut
down sendmail.  There are many other MTAs and I'm glad of that because
every one of them has strengths in particular environments.  I have no
doubt that sometime I'll stumble on an environment where I'd like to
deploy postfix, for example.  It's not possible to choose any "one true
mailer" and be sure that it'll keep all NetBSD users happy.  There has to
be a mailer included somehow, so I contend that it's better to leave
what's there as a "default" that won't upset or disappoint anybody, and
allow people to replace it with the software they prefer if they wish.

> and the only non-technical argument you've given that I see any merit in
> is that many admins will already know how to work with sendmail because
> it's the de facto industry standard.

That is my point exactly.  Nobody's going to agree to one single mailer
being "best" for NetBSD, so please, leave sendmail in its defacto
position, and let people choose for themselves.

> However that last argument seems contrary to the TNF pubilshed goal of
> technical excellence -- at least so long as you will admit that postfix
> is likely technically superior in at least its design and perhaps even
> in the current extent of the implementation.

I don't agree to that at all - I'm not making a comparison.  Personally, I
don't like the many-component approach of qmail, postfix, PMDF and their
ilk.  Those who like sendmail's approach and are accustomed to its
configuration probably feel that it is an excellent implementation - and
there are a lot of rabid sendmail fans around.

> > It provides no advantage at all, and it confuses a lot of people without
> > justification.  Exim would probably be a better choice, actually, but I
> > still wouldn't want to see it replace sendmail in the base distribution.
> I've not yet seen any actual documented cases of confusion -- just
> complaints about apparent bloat, wasted resources, etc.....

I think that including two MTAs represents bloat and wasted resources.
Since I now understand that the goal is to replace sendmail with postfix,
I anticipate confusion on the part of those people who like sendmail and
expect to find it in the NetBSD distribution, just as it has been for a
long time.

> > No, it's absolutely not.  sendmail as included in Unix OS distributions
> > for many a year has fulfilled the purpose required of it just fine,
> Anything that requries a 1500-page guide to describe its idiosyncracies 
> isn't likely very "basic". 

sendmail's level of complexity depends on the requirements of the
environment - the same is true of the alternatives.  A default
as included in the distribution is fine for many hosts and learning to
administer that simple configuration is unquestionably as "basic" as
Internet mail exchange gets.  The number of pages in the O'Reilly book is
hardly relevant.

> > Anything that includes a very hard-to-
> > program programmable finite state machine as it's core concept is not
> > "basic".

That would depend on whether one has the need to meddle with that
programming or merely needs to use the tool and manage the service on a
day to day basis.

> > and
> > nobody has presented any justification for replacing it with postfix
> > beyond "but I like postfix and I think it's better than sendmail".
> I guess we thought that was obvious. 

Frankly, I don't think you thought about NetBSD users much at all.

> > I'd suggest reading the code if
> > you really need convincing, but in general the overview and anatomy docs
> > available on should be reasonably convincing.  There's
> > quite a pile of positive testimony from Postfix users too....

Convincing of what?  That some people like postfix?  I could guess that
without knowing anything about it.  A lot of people like exim, or
sendmail, or qmail, etc.

> > If there was a compelling
> > reason for replacing sendmail with postfix I wouldn't be complaining, but
> > there is just absolutely no justification at all.  Why fix something
> > that's not broken?
> many people, even some of its fans I think, consider sendmail "broken by
> design"

What a strange statement.  sendmail has active development and a lot of
happy users - no doubt they think there's improvements to be made,
otherwise they'd stop working on it right?  Same is true of postfix,
qmail, exim, etc.  sendmail shouldn't be singled out as "broken" at all.
This only lends strength to my suspicion that the proponents of postfix
replacing sendmail in NetBSD are just fans blinded by their liking for
postfix and who refuse to believe that any NetBSD could want something

> However the mere fact that there are many good choices of MTA other than
> sendmail should be an indication that something needs fixing, broken or
> not.

Eh?!  You really believe that there should be "one true" MTA?  That there
should be no choice and that all alternatives should be stamped out in
favour of postfix?  *sheesh* Anything that isn't in line with the one true
way is "broken" I suppose.

> >  Seems like it's just a personal crusade by a few
> > postfix fans :(
> Basically what I'm hearing from you is that sendmail is a sacred cow and
> it must not be touched.

Pretty much, yes.  Do not mess with the sacred cow, and allow folks to
choose their own replacement MTA if they wish - as that's a deeply
religious decision.  There are heaps of other tools distributed with
NetBSD which should be considered part of the sacred herd, because a _lot_
of users depend on them being in the distribution and working in the
familiar manner. postfix does not cleanly or adequately replace sendmail
for all user requirements, and there is no justification for trying.

Pete Naylor