httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Martin Langhoff <martin.langh...@gmail.com>
Subject Re: debugging dropped output on v2.2.15
Date Sat, 19 Oct 2013 17:45:04 GMT
On Sat, Oct 19, 2013 at 9:45 AM, Jeff Trawick <trawick@gmail.com> wrote:
> I'm not sure what you mean by parent/child, or there's some confusion on your part.

Hi Jeff,

thanks for your reply. I'll admit there might be some confusion on my
part. My knowledge of apache internals is... imperfect.

The facts I observe:

 - Across all children in this configuration, fd 12 is the HTTP
stream. The HTTP request is read() from fd 12, and response appears as
writev() to fd 12. When idle, children poll fd 12 until they read a
new HTTP request.

 - In one specific case, a complex page served by mod_php, mod_php has
to do quite a bit of work before it can reply, this delay can be
measured in minutes. In those cases, we sometimes (~50%) see a strange
behavior: the reply _is_ written to fd 12 but is _not_ seen on the
http stream (packet capture with Wireshark, run from an intermediate
host).

My (erroneous?) understanding of apache internals was that the socket
I see in fd 12 was a back to the httpd parent process, which in turn
passes it back to the TCP socket, so I was wondering about a possible
comm breakdown there.

If the fd is directly to the TCP stream, I perhaps need to move the
packet capture to happen on the same host as where apache is running.

cheers,


m
-- 
 martin.langhoff@gmail.com
 -  ask interesting questions
 - don't get distracted with shiny stuff  - working code first
 ~ http://docs.moodle.org/en/User:Martin_Langhoff

Mime
View raw message