On 17 Dec 1999 08:10:26 -0000 , "Petr Novotny" writes:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 16 Dec 99, at 20:48, [EMAIL PROTECTED] wrote:
> > I would list a few things in sendmail's favor:
> >
> > 1) The ability to rewrite headers "up front" without
> > requiring double delivery (once to a rewriting
> > program, then again to the destination).
>
> Nice - but any program with root privileges doing header rewriting is
> a recipe for disaster.
Well, perhaps -- it does make existing security
holes potentially more damaging.
The inability to do header rewriting without making
two trips through the queue causes a severe performance
hit, if you are doing anything other than low volume.
> It also depends on what you mean by "delivery"; qmail's logic is
> much simpler, but internally sendmail will do "almost" the same,
> without giving you so many chances to inspect and debug.
By delivery, I mean "a call to qmail-inject" (two fsyncs).
> > 1a) The ability to forward mail, up front....
>
> Sorry?
That is forwarding with only one trip through the queue.
> > 2) The ability (even if theoretical) to deliver a
> > message without fsync()ing it into the queue, unless
> > absolutely necessary.
>
> I don't find that a plus... I prefer to stay on the safe side, rather than
> (potentially) losing messages for the sake of speed.
Such an approach is not necessarily unsafe. Unsafe
means that messages can be lost after they are
acknowledged. If the message is successfully
delivered before acknowledgement is given, then
enqueueing the message is unnecessary.
--
Chris Mikkelson | Problems are posed by fools like me;
[EMAIL PROTECTED] | But only Heuristics can search a tree.
