Mailing-List: contact modperl-help@perl.apache.org; run by ezmlm
Precedence: bulk
Received-SPF: pass (nike.apache.org: local policy)
Date: Tue, 12 Jan 2010 13:52:15 -0500 (EST)
From: Alex Aminoff <aminoff@nber.org>
To: Mark Copper <mcopper@titaninterface.com>
cc: modperl@perl.apache.org, feenberg@nber.org
Subject: Re: intermittent segfaults, ssl?
In-Reply-To: <20100112182417.GF3477@titaninterface.com>
Message-ID: <alpine.LFD.2.00.1001121343160.8353@perlw2.nber.org>
References: <20100112182417.GF3477@titaninterface.com>
User-Agent: Alpine 2.00 (LFD 1167 2008-08-23)
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; format=flowed; charset=US-ASCII


We had a very similar problem. After a lot of effort trying to track down 
the source, we believe we isolated it to a line of perl code that 
incorrectly sets $/ without localizing it.

mod_backtrace showed that the problem seemed to be happening in

  modperl_perl_global_request_save

Here is the thread that suggested the fix that worked for us:

  http://www.gossamer-threads.com/lists/modperl/modperl/100066

Here is the thread we posted, not actually any more informative:

  http://www.gossamer-threads.com/lists/modperl/modperl/100842

The proof is in the pudding. After fixing the line with $/, we went from 
dozens of seg faults per hour to none.

   - Alex Aminoff
     BaseSpace.net
     National Bureau of Economic Research (nber.org)

On Tue, 12 Jan 2010, Mark Copper wrote:

> Hi,
>
> I have a server like this:
>  Server Version: Apache/2.2.9 (Debian) mod_ssl/2.2.9 OpenSSL/0.9.8g
>          mod_apreq2-20051231/2.6.0 mod_perl/2.0.4 Perl/v5.10.0
> I'm also using HTML::Mason
>
> I've been getting intermittent segfaults like this:
>  child pid 10142 exit signal Segmentation fault (11)
> ever since in installed Apache2 in March.
>
> I am going to try to debug this, but I thought I would ask if anyone
> might have a suggestion based on this behavior:
>  - no segfaults occur with 8 hours of an apache restart; the first fault
>    can be 8 to 48 hours after restart, the 2nd may occur within seconds;
>    there have never been more than 5 in a day.
>
>  - I have observed these faults *only* for port 443 (ssl) requests.
>
>  - The simplest case has been when a plain HTML page was served through
>    mod_perl and mason; e.g. no database call.
>
> I know probably shouldn't be imposing on list-readers time without doing
> more work, but I just wonder is it isn't something really elementary
> that I'm missing.
>
> Mark
>