modcplusplus-user

[Modcplusplus-user] time out during post requests

[Nathan S. <na...@al...>]

Greetings all,

I had emailed John about this earlier, but am posting here to recap what
I've figured out thus far.

Whenever a HTTP POST request is made to a cpp module, apache immediatly
returns whatever output the module provides but then inexplicably (so
far) doesn't close the connection, leading to an around 10 second
timeout before apache (or browser, unsure which) gives up on waiting,
and closes the tcp connection.

This only seems to occur with browsers that attempt a keep alive
connection during posts.  Both Mozilla and IE versions > 4 do this,
however Netscape 4.x does not.

Here's a sniffed request that works:

POST /cpp-handler HTTP/1.1
User-Agent: curl/7.9.2 (i686-pc-linux-gnu) libcurl 7.9.2 (OpenSSL
0.9.6b)
Host: nas.office.allmed.net
Pragma: no-cache
Accept: image/gif, image/x-xbitmap, image/jpeg, image/pjpeg, */*
Content-Length: 17
Content-Type: application/x-www-form-urlencoded

test1=sa&test2=32


And one that does not:


POST /cpp-handler HTTP/1.1
Host: nas.office.allmed.net
Accept: */*
Accept-Language: en
Pragma: no-cache
Connection: Keep-Alive
Referer: http://nas.office.allmed.net/test.html
User-Agent: Mozilla/4.0 (compatible; MSIE 5.12; Mac_PowerPC)
UA-OS: MacOS
UA-CPU: PPC
Content-type: application/x-www-form-urlencoded
Extension: Security/Remote-Passphrase
Content-length: 17

test1=sa&test2=32


As you can see, the content-length and content is the same for both
requests, they are both HTTP/1.1,  however the failing request has the
Connection: Keep-Alive header.

From what I've been able to figure out by crude ap_log statements in the
mod_cpp source, the request does get through mod_cpp just fine, and
control passes back to apache before the time-out occures.  gdb is
useless with the multithreaded nature of apache2, and segfaults before
even getting it running.

The problem to me seems to lie in mod_cpp, as normal cgi's work fine. 
Therefore  perhaps there is something that mod_cpp should perform for
post requests that it's not?

I've confirmed that turning off keep alives off in httpd.conf, makes the
problem go away.

Does anyone else see this behaviour or is it only myself?

Nathan

Re: [Modcplusplus-user] time out during post requests

[<jo...@st...>]

very interesting - 

the content length looks correct.....  but the browser hangs?  I'll try to 
reproduce this.

sterling

On 15 Apr 2002, Nathan Stitt wrote:

> Greetings all,
> 
> I had emailed John about this earlier, but am posting here to recap what
> I've figured out thus far.
> 
> Whenever a HTTP POST request is made to a cpp module, apache immediatly
> returns whatever output the module provides but then inexplicably (so
> far) doesn't close the connection, leading to an around 10 second
> timeout before apache (or browser, unsure which) gives up on waiting,
> and closes the tcp connection.
> 
> This only seems to occur with browsers that attempt a keep alive
> connection during posts.  Both Mozilla and IE versions > 4 do this,
> however Netscape 4.x does not.
> 
> Here's a sniffed request that works:
> 
> POST /cpp-handler HTTP/1.1
> User-Agent: curl/7.9.2 (i686-pc-linux-gnu) libcurl 7.9.2 (OpenSSL
> 0.9.6b)
> Host: nas.office.allmed.net
> Pragma: no-cache
> Accept: image/gif, image/x-xbitmap, image/jpeg, image/pjpeg, */*
> Content-Length: 17
> Content-Type: application/x-www-form-urlencoded
> 
> test1=sa&test2=32
> 
> 
> And one that does not:
> 
> 
> POST /cpp-handler HTTP/1.1
> Host: nas.office.allmed.net
> Accept: */*
> Accept-Language: en
> Pragma: no-cache
> Connection: Keep-Alive
> Referer: http://nas.office.allmed.net/test.html
> User-Agent: Mozilla/4.0 (compatible; MSIE 5.12; Mac_PowerPC)
> UA-OS: MacOS
> UA-CPU: PPC
> Content-type: application/x-www-form-urlencoded
> Extension: Security/Remote-Passphrase
> Content-length: 17
> 
> test1=sa&test2=32
> 
> 
> As you can see, the content-length and content is the same for both
> requests, they are both HTTP/1.1,  however the failing request has the
> Connection: Keep-Alive header.
> 
> >From what I've been able to figure out by crude ap_log statements in the
> mod_cpp source, the request does get through mod_cpp just fine, and
> control passes back to apache before the time-out occures.  gdb is
> useless with the multithreaded nature of apache2, and segfaults before
> even getting it running.
> 
> The problem to me seems to lie in mod_cpp, as normal cgi's work fine. 
> Therefore  perhaps there is something that mod_cpp should perform for
> post requests that it's not?
> 
> I've confirmed that turning off keep alives off in httpd.conf, makes the
> problem go away.
> 
> Does anyone else see this behaviour or is it only myself?
> 
> Nathan
> 
> 
> _______________________________________________
> Modcplusplus-user mailing list
> Mod...@li...
> https://lists.sourceforge.net/lists/listinfo/modcplusplus-user
> 

Re: [Modcplusplus-user] time out during post requests

[<jo...@st...>]

Just curious - do you have the example filters loaded too in your config 
file?  maybe the problem is in the filters......  i'm not quite able to 
reproduce it yet....


have you modified the exmpample cpp handler?


thanks.

sterling

Re: [Modcplusplus-user] time out during post requests

[Nathan S. <na...@al...>]

On Mon, 2002-04-15 at 13:22, jo...@st... wrote:
> Just curious - do you have the example filters loaded too in your config 
> file?  maybe the problem is in the filters......  i'm not quite able to 
> reproduce it yet....
> 

No, I'm running a completly stock Apache 2.0.35 GA, configured with: 
./configure --enable-info --with-mpm=worker

and the following added in the httpd.conf:

LoadFile /usr/lib/libstdc++.so.3.0.2
LoadModule cplusplus_module modules/libmod_cplusplus.so.0.0.0
LoadCPPHandler test_handler modules/libtest_handler.so

AddCPPConnectionInputFilter test_input_filter
<Location /cpp-handler>
        SetHandler cpp-handler
        AddCPPHandler test_handler
</Location>


> have you modified the exmpample cpp handler?
> 
> 
I'm not sure which handler your refering to.  I've been testing with the
example/handler/test_handler.cpp.  The only changes I've made is to add
mHits = 0 in the constructor, and then when I saw it wasn't handling
posts correctly, added a bunch of ap_log_error(...) calls.

I'm testing with some of the other examples now to see if they exibit
the same behaviour.

Will let you know the results.
Nathan

Re: [Modcplusplus-user] time out during post requests

[sterling <ste...@co...>]

On 15 Apr 2002, Nathan Stitt wrote:

> On Mon, 2002-04-15 at 13:22, jo...@st... wrote:
> > have you modified the exmpample cpp handler?
> > 
> > 
> I'm not sure which handler your refering to.  I've been testing with the
> example/handler/test_handler.cpp.  The only changes I've made is to add
> mHits = 0 in the constructor, and then when I saw it wasn't handling
> posts correctly, added a bunch of ap_log_error(...) calls.

cool - i think i just found the problem.  if your handler expects POSTs it 
either need to suck the content out of the bucket, or tell the request to 
discard it before returning.....  i just committed it to cvs - are you 
building from cvs? or using the tarball up on sourceforge?  i'll update 
the tarball with the change.


sterling

Re: [Modcplusplus-user] time out during post requests

[Nathan S. <na...@al...>]

On Mon, 2002-04-15 at 13:59, sterling wrote:
> On 15 Apr 2002, Nathan Stitt wrote:
> 
> > On Mon, 2002-04-15 at 13:22, jo...@st... wrote:
> > > have you modified the exmpample cpp handler?
> > > 
> > > 
> > I'm not sure which handler your refering to.  I've been testing with the
> > example/handler/test_handler.cpp.  The only changes I've made is to add
> > mHits = 0 in the constructor, and then when I saw it wasn't handling
> > posts correctly, added a bunch of ap_log_error(...) calls.
> 
> cool - i think i just found the problem.  if your handler expects POSTs it 
> either need to suck the content out of the bucket, or tell the request to 
> discard it before returning.....  i just committed it to cvs - are you 
> building from cvs? or using the tarball up on sourceforge?  i'll update 
> the tarball with the change.
> 

Cool!  Was wondering if it was something like that. I'm running cvs,
I'll resync tommorow and give it a shot.

Thanks for all your help on this!

On a side note, I've been looking into moddifing the gnu c++ cgi
library  http://www.gnu.org/software/cgicc/cgicc.html for use with
modcplusplus apache.  Right now it looks like a rather easy change, but
will know more once I get started on it.  It's a bit more heavy that I
would like, as it has lots of classes to generate html tables, forms,
etc. I probably won't use them, but looks like they aren't constructed
unless used, and someone else would probably appreciate them.

Later,
Nathan

Re: [Modcplusplus-user] time out during post requests

[Nathan S. <na...@al...>]

On Mon, 2002-04-15 at 13:59, sterling wrote:

> cool - i think i just found the problem.  if your handler expects POSTs it 
> either need to suck the content out of the bucket, or tell the request to 
> discard it before returning.....  i just committed it to cvs - are you 
> building from cvs? or using the tarball up on sourceforge?  i'll update 
> the tarball with the change.
> 

It looks as though that fixed it!  So all I was missing reading the
content?  I was mistakenly ignoring it for the purpose of my simple
tests, I guess apache didn't like that.

Thanks for all your help on this.  Now I can start focusing on getting
an actually library together for easily accessing the get and post
values.

Later,
nathan

Re: [Modcplusplus-user] time out during post requests

[<jo...@st...>]

On 16 Apr 2002, Nathan Stitt wrote:

> On Mon, 2002-04-15 at 13:59, sterling wrote:
> 
> > cool - i think i just found the problem.  if your handler expects POSTs it 
> > either need to suck the content out of the bucket, or tell the request to 
> > discard it before returning.....  i just committed it to cvs - are you 
> > building from cvs? or using the tarball up on sourceforge?  i'll update 
> > the tarball with the change.
> > 
> 
> It looks as though that fixed it!  So all I was missing reading the
> content?  I was mistakenly ignoring it for the purpose of my simple
> tests, I guess apache didn't like that.

yup - if you are planning on not reading the content, apache needs to be 
told via pRequest->discard_request_body() method.

sterling