Mailing-List: contact test-dev-help@httpd.apache.org; run by ezmlm
Precedence: bulk
Reply-To: test-dev@httpd.apache.org
Message-ID: <3F313027.7050809@apache.org>
Date: Wed, 06 Aug 2003 12:43:19 -0400
From: Greg Ames <gregames@apache.org>
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US;
 rv:1.4) Gecko/20030624
MIME-Version: 1.0
To: test-dev@httpd.apache.org
Subject: Re: [PATCH] mod_specweb99.c: thread-safe lock
References: <BB55BA78.30A5F%sctemme@covalent.net>
In-Reply-To: <BB55BA78.30A5F%sctemme@covalent.net>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Sander Temme wrote:

>All,
>
>Months ago, Madhu posted a patch to put thread-safe locks around the Post
>logging in mod_specweb99... here's an alternative using apr_global_mutex
>locks:
>  
>
excellent!  I was running specweb99 earlier this week and was seeing 
post log bogosity with worker on Linux PPC.  Maybe it's due to the 
threads.  Plus, I thought Madhu said that our current implementation is 
slow on HP/UX too.  Hopefully the APR default mutex will be a good 
performer on our major platforms.

>This approach is also used in mod_rewrite. I know apr_global_mutexes are
>advertised as potentially expensive, but in a hybrid situation you're going
>to end up doing something like this, and apr_global_mutexes turn into
>process locks when APR is thread-free or (not tested) on Win32.
>  
>
I think this approach is the Right Thing for now.  Longer term, I would 
like to use atomics a shared memory variable.

>Tested on Darwin under light load and Solaris under moderate load. Thoughts,
>opinions, comments? Or shall I just commit the thing?
>
I'll try it on the Linux PPC machine where I saw the post log flakyness and report back.  I wouldn't have a problem if you just committed it but I should be able to beat it up heavily and report back this afternoon or tomorrow.

Thanks!
Greg