Mailing-List: contact dev-help@httpd.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@httpd.apache.org
Received-SPF: pass (athena.apache.org: domain of jim@jagunet.com designates
 209.133.199.10 as permitted sender)
Message-Id: <C2474143-C29F-48A2-969B-9359E79C6005@jaguNET.com>
From: Jim Jagielski <jim@jaguNET.com>
To: dev@httpd.apache.org
In-Reply-To: <aafb0e350809191424jf904d91nf770ec79a58f7cce@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed; delsp=yes
Content-Transfer-Encoding: quoted-printable
Mime-Version: 1.0 (Apple Message framework v929.2)
Subject: Re: mod_proxy race condition bug #37770
Date: Sat, 20 Sep 2008 09:30:22 -0400
References: <aafb0e350805121052g74102f86v18e527c1a50b35a2@mail.gmail.com>
 <20080512203154.119e4507@grimnir>
 <aafb0e350805121315t1f6c1226l53142a08378a3104@mail.gmail.com>
 <aafb0e350805161421v10417b9dl573fdcf12fd6c8c5@mail.gmail.com>
 <482F30D2.90507@apache.org> <48444C87.8080400@apache.org>
 <aafb0e350806021311j3a266519i93c36d7ef88a746@mail.gmail.com>
 <4844564B.70301@apache.org>
 <aafb0e350809191424jf904d91nf770ec79a58f7cce@mail.gmail.com>


On Sep 19, 2008, at 5:24 PM, Adam Woodworth wrote:

> Now that we're using 2.2.9 in the field, I've noticed a problem with
> r657443, or at least it's a problem for our application.
>
> r657443 added code to shutdown the connection w/o any response with an
> EOC in the case that a second (or later) request on a keepalive
> connection failed to get a response from the backend.
>
> The problem with this is that it will also do this for connections to
> the backend that timed out.  For our application, we want actual
> timeouts to be caught as normal and kicked back through to the client
> as a 50x error.
>
> I made a simple one line patch to our copy of httpd 2.2.9 and it seems
> to work.  I just check the rc code to see if it's a timeout error and
> if so skip the new code from r657443.
>
> I thought I'd raise this to the httpd developers -- do you think it's
> something that should be part of the httpd code?  Having a timeout to
> the backend causing the client to re-request seems not so good.
>

Agreed. Good catch.

>
> *** mod_proxy_http.c.bak	Thu Sep 18 20:42:05 2008
> --- mod_proxy_http.c	Fri Sep 19 15:39:32 2008
> ***************
> *** 1376,1382 ****
>               * seamonkey only display an empty page in this case =20
> and do
>               * not do a retry.
>               */
> !             if (r->proxyreq =3D=3D PROXYREQ_REVERSE && =
c->keepalives) {
>                  apr_bucket *eos;
>
>                  ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r,
> --- 1376,1385 ----
>               * seamonkey only display an empty page in this case =20
> and do
>               * not do a retry.
>               */
> !             if (r->proxyreq =3D=3D PROXYREQ_REVERSE && c->keepalives =
&&
> rc !=3D APR_TIMEUP) {
>                  apr_bucket *eos;
>
>                  ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r,
>
>
>
> On Mon, Jun 2, 2008 at 4:21 PM, Ruediger Pluem <rpluem@apache.org> =20
> wrote:
>>
>>
>> On 06/02/2008 10:11 PM, Adam Woodworth wrote:
>>>>
>>>> =46rom a usage standpoint I have no real preference other than I =20=

>>>> need to
>>>
>>> be able to configure it from httpd.conf on a per-host basis. :)
>>>
>>> What is a worker specific property?
>>
>> All the properties you put behind ProxyPass / balancermember or via =20=

>> proxyset
>> in a block
>> like
>>
>> <Proxy http://www.somebackend.com/blah>
>> </Proxy>
>>
>> Regards
>>
>> R=FCdiger
>>
>>
>