From: "David Harris" <dharris@drh.net>
To: <new-httpd@apache.org>
Subject: RE: standard virtual hosting.. how good?
Date: Tue, 29 Jun 1999 10:33:15 -0400
Message-ID: <000201bec23c$5276db60$0500a8c0@delf>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
In-Reply-To: <199906241747.NAA03755@devsys.jaguNET.com>
Importance: Normal
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org


Jim Jagielski wrote:
> But not having -t do a basic config test as checking DocumentRoot
> is totally bogus. So it's braindead to have -t not do the stat.
> And my point wasn't so much that people won't notice the run time
> errors. They will, eventually. My point was that previous versions
> reported the error immediately when Apache was started.
>
> I hope you really aren't proposing removing the stat with -t simply
> because it means a delay when someone is using apachectl to
> restart the server? If so then we're saying "instead of rec'ing
> instant feedback that a DocumentRoot is bogus when you start
> Apache and instead of running apache -t to do so to check the
> DocumentRoot isn't bogus, the new procedure is to simply run
> Apache and then immediately look at error_log". That's not tough
> for them, thats stupid on our part.
>
> Bogus DocumentRoots are a config-time error. It's easy to check
> at config time, we've been checking it at config time and people
> are most likely using it to check at config time. And we want
> to stop that for what reason again?? Because they should be instead
> looking in error_log and we don't _want_ to provide a real config-time
> test??

If the stat in the -t check takes a long time, then I'll don't want to do
it. The reason is this: I'm _virtually certain_ that every document root
will exist because all my virtual hosts are automatically generated, so the
delay in the -t check is not worth it to me. I expect others running
thousands of virtual hosts also have an automatic configuration setup they
trust. I don't know if they consider the trade off "worth it" or not.

However, you are right that removing the stat from the -t command does break
the strength of the config test. But even without the stat, I'm glad to use
the -t check because it catches catastrophic configurations mess ups that
would cause the whole server to fail.

How about a three way option for the stat check: off, on,
only_in_config_check? (And double the amount of descriptive text required
for this obscure directive...)

But don't worry about me too much. I'll just add a patch to my RPM file if I
can't do what I want.

 - David Harris
   Principal Engineer, DRH Internet Services