Mailing-List: contact dev-help@httpd.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@httpd.apache.org
Date: Tue, 02 Mar 2004 09:12:43 -0800
From: Justin Erenkrantz <justin@erenkrantz.com>
To: dev@httpd.apache.org
Subject: Re: FreeBSD 4.x and Apache2: worker MPM issue 
Message-ID: <2147483647.1078218763@[10.0.1.96]>
In-Reply-To: <20040302113616.K23774@ganymede.hub.org>
References: <20040228220424.S82441@ganymede.hub.org>
 <2147483647.1078009181@[10.0.1.91]> <20040229122848.A82441@ganymede.hub.org>
 <2147483647.1078045404@[10.0.1.91]> <20040229153545.M82441@ganymede.hub.org>
 <2147483647.1078092202@[10.0.1.92]> <20040301112516.R776@ganymede.hub.org>
 <2147483647.1078180247@localhost> <20040302113616.K23774@ganymede.hub.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

--On Tuesday, March 2, 2004 11:47 AM -0400 "Marc G. Fournier" 
<scrappy@hub.org> wrote:

> Note that *BSD is looking at a 4.10 RSN, and I'm trying to fight for
> trying to get this fixed, if its possible, which is why I'm trying to come
> up with some data to fight with ...
>
> Is there anywhere that there is a summary of this "gnarly stuff"?
> Something you could point me at, that I can use/question?

You'd have to look at the archives for dev@httpd.  I would have been the one 
posting it along with David Reid and Aaron Bannert.  We first looked at it in 
late 2001 and early 2002.  This was coupled with issues with sendfile in 
libc_r that we got Alfred P. to fix.

I'd look inside libc_r and see if someone committed some fixes to the 
scheduling and condition variable implementations recently.  If so, then yah, 
someone might have fixed this.  (Don't have the time to check the CVS history 
myself.)

I also know that I checked this a few months ago on minotaur (ASF's CVS 
server), which is running 4.9-STABLE (from Nov 29 2003).  So, if it got fixed, 
it's probably after that as it was still broken then.

> which is what I would expect ...
>
> now, running http_load with a rate of 2 (simple), I'm still left with
> those three processes ...

Okay.

> Great, so either it did get fixed at some point, and nobody has
> acknowledge it, or I'm really doing something wrong trying to trigger it
> :( that last one, if I read the notes on http_load, would have hit it 500
> times in 10 seconds, which should have simulated a good load, I would have
> thought?

The only sure-fire reproduction case we had was to use two copies of telnet 
against the server.  What happens if you don't use http_load at all?

Yes, it's possible it's been fixed in 4.x; we know it's fixed in 5.2+ with 
libkse and libthr.  But, the libc_r in 5.2 is still broken.  -- justin