large requests crash mod_jk on HP-UX 11.0

["Ashton, Bruce" <br...@metoffice.com>]

I am really a tomcat user, but I have ended up trolling through the mod_jk
code looking for this bug, so I thought this was the appropriate place to
post my discoveries;


Background:
Our Unix team have built Apache from the source for HP-UX 11.0 with the HP
cc compiler, not gcc.
cpp and ccom versions are A.11.01.00, ld version is B.11.25
mod_jk has been built with the same compiler using apxs and the
build-hpux-cc.sh script.
Apache is version 1.3.22.
The mod_jk source comes from the file
jakarta-tomcat-connectors-4.0.2-01-src.zip from the Jakarta site.


The problem:
Mostly this build works except: certain URL's on our site fail with an
"internal server error" message to the browser.  The request doesn't show up
in Apache's access log.

The following shows up in Apache's error log:

[Mon May 27 14:07:01 2002] [notice] child pid 13574 exit signal Segmentation
fault (11)
[Mon May 27 14:07:01 2002] [notice] child pid 12572 exit signal Segmentation
fault (11)

always two lines.

With JkLogLevel set to debug, the following shows up in the mod_jk log:

[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (447)]: Into
jk_uri_worker_map_t::map_uri_to_worker
[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (464)]: Attempting to map
URI '/openroadTextFC.do'
[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (529)]:
jk_uri_worker_map_t::map_uri_to_worker, Found a suffix match web -> *.do
[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (447)]: Into
jk_uri_worker_map_t::map_uri_to_worker
[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (464)]: Attempting to map
URI '/openroadTextFC.do'
[Mon May 27 14:07:01 2002]  [jk_uri_worker_map.c (529)]:
jk_uri_worker_map_t::map_uri_to_worker, Found a suffix match web -> *.do


I add some extra debugging myself and ascertained that jk_translate runs
through to completion, but no debugging code in jk_handler ever gets run.
If jk_handler is entered at all it must fail at line 1075, "const char
*worker_name = ap_table_get(r->notes, JK_WORKER_ID);"

I've kind of shied away from trying to debug through apache.  I didn't do
the original build anyway.

For any given URL that fails, it is possible to make it work again by making
the URL *SHORTER*.  This is possible if there is extraneous crap in the
query string, which you can cut out.  I have discovered that for a
particular page (any one bad page will fail consistently)  There will be a
particular length of URL: An 86 character URL may work say, but an 87
character URL will fail.  The useable length of URL will vary from page to
page though, and I haven't seen any other pattern in it.
My speculation is that it's the size of the request object or something, of
which the URL is only a component.  A bit woolly I admit, but it's all I've
got.

Part of my problem is that I don't really know in what order Apache calls
methods from mod_jk, so I don't know if any other part of mod_jk is being
run between jk_translate and jk_handler.
I also don't know that it isn't Apache code that is falling over instead of
mod_jk code.
Also IANACP (I Am Not A C Programmer) - not really anyway.  Mostly Java, and
I've just started dabbling in C in incidental ways such as this.
And I'm new to HP-UX as well.

Can anyone maybe shed a little light on the Apache/mod_jk interface for me?
A bit of a rundown on the order in which the methods are called would be
useful.  Also is there any Apache API or mod_jk API documentation online
which I might not be aware of.  And if anyone tells me that I need to start
hacking the Apache code, I'll be very depressed.

Thanks in advance,



Bruce Ashton
Java Developer
Internet Application Development
The Met Office   http://www.metoffice.com/
ext. 4560


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: large requests crash mod_jk on HP-UX 11.0

[co...@covalent.net]

On Mon, 27 May 2002, Ashton, Bruce wrote:

> Apache is version 1.3.22. The mod_jk source comes from the file
> jakarta-tomcat-connectors-4.0.2-01-src.zip from the Jakarta site.

Can you try with the CVS head ? 

There are many bug fixes since 4.0.2.

> [Mon May 27 14:07:01 2002] [notice] child pid 13574 exit signal Segmentation
> fault (11)
> [Mon May 27 14:07:01 2002] [notice] child pid 12572 exit signal Segmentation
> fault (11)

I hope using CVS head will fix this. If not, could you try running:
  gdb httpd 
  > run -X

Then do the request, you should see the sig11 in gdb - then do 
'bt' and send us the result ?

Or configure apache to generate core dumps, and go: 
   gdb .../httpd .../core
   > bt


> I add some extra debugging myself and ascertained that jk_translate runs
> through to completion, but no debugging code in jk_handler ever gets run.
> If jk_handler is entered at all it must fail at line 1075, "const char
> *worker_name = ap_table_get(r->notes, JK_WORKER_ID);"

Weird. Have you added a printf() before and after, and it crashes at that 
line ??? 


> For any given URL that fails, it is possible to make it work again by making
> the URL *SHORTER*.  This is possible if there is extraneous crap in the
> query string, which you can cut out.  I have discovered that for a

Interesting. What's the URL that fails ? 

> Part of my problem is that I don't really know in what order Apache calls
> methods from mod_jk, so I don't know if any other part of mod_jk is being
> run between jk_translate and jk_handler.

None AFAIK. Are you sure jk_handler is never invoked ?


Costin


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>