general/4447: Apache 1.3.6 w/ Jrun Pro and multiple jsms returning invalid header information intermittently

[Ryan Clifton <rc...@inetbiz.com>]

>Number:         4447
>Category:       general
>Synopsis:       Apache 1.3.6 w/ Jrun Pro and multiple jsms returning invalid header information intermittently
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    apache
>State:          open
>Class:          sw-bug
>Submitter-Id:   apache
>Arrival-Date:   Thu May 20 13:30:01 PDT 1999
>Last-Modified:
>Originator:     rclifton@inetbiz.com
>Organization:
apache
>Release:        1.3.6
>Environment:
Solaris 2.6, current patch level.  compiler=gcc

uname -a
SunOS ultra2.inetbiz.com 5.6 Generic sun4u sparc SUNW,Ultra-2
>Description:
Using Apache 1.3.6 and the new ability to put srm.conf informtion into the 
<Virtual Host> tags, we have one httpd server running multiple Jrun Pro java vms through proxy hosts (each proxy host is on a different port on the same machine).
Every once in a while, but enough that it can be reproduced within about 5 - 10 minutes, the web server will return invalid header information, aparently without ever going to the proxy connector.
Here is a copy of the bad header information:

ncoded COOKIE: jrunsessionid=92713774134783398 IF_MODIFIED_SINCE: REFERER:
<http://www.markethome.com/webj/superadmin?func=C_modifyCat&cid=42&pid=LLBea
nBuyers>
USER_AGENT: Mozilla/4.08 [en] (Win95; I ;Nav) QUERY_STRING: REQUEST_METHOD:
POST
SCRIPT_NAME: /webj/webj/superadmin AUTH_TYPE: PATH_INFO: REMOTE_USER:
REMOTE_HOST: 206.111.144.155 SERVER_NAME: www.markethome.com
<http://www.markethome.com> SERVER_PORT: 80
SERVER_PORT_SECURE: SERVER_PROTOCOL: HTTP/1.0 REMOTE_ADDR: 206.111.144.155

We have logs coming out of Jrun that track every hit, jrun is never receiving a request when we get this bug.
So.. while this looks like a bug with Apache 1.3.6 (we were previously using 1.3.4 with only once jsm per httpd root process and never saw this), there is also a possibility that it could be problem with the interface between 1.3.6 and the jrun connector proxy.

What I do know is that we were running Apache 1.3.4 with Jrun Pro v 2.2 in production, under a heavy load and never saw this.  Now with Apache 1.3.6, the same version of Jrun but with multple java vms per Apache root process, this problem shows up quite often.

This problem appears on our production, staging, and dev servers, meaning that it is not localized to one machine.

This is a critical problem, as most hosting sites are going to need to set up a similar environment in order to run Java servlets with Apache.
>How-To-Repeat:
Set up Apache 1.3.6 on Solaris.  Enable at least two jsms, on two different proxy ports.
Bang at them for a while.  For us, just hitting reload on a browser can reproduce the problem.

Alternatively, go to www.markethome.com/webj/subscriber?func=S_survey&c=1&p=1920
and hit reload a bunch of times (hopefully murphys law won't kick in and never show the bug).
>Fix:

>Audit-Trail:
>Unformatted:
[In order for any reply to be added to the PR database, ]
[you need to include <ap...@Apache.Org> in the Cc line ]
[and leave the subject line UNCHANGED.  This is not done]
[automatically because of the potential for mail loops. ]
[If you do not include this Cc, your reply may be ig-   ]
[nored unless you are responding to an explicit request ]
[from a developer.                                      ]
[Reply only with text; DO NOT SEND ATTACHMENTS!         ]