We begin testing Pound, sending requests to one back-end server. Pound is
listening for HTTP and HTTPS traffic. It is being managed by daemontools, so it
was compiled with backgrounding set to off. It was also compiled without
syslog, so that multilog could be used. It works fine for 90% of the people
visiting the website. The remaining 10% are sending no data for Form POST data.
I can't explain it! It is quite strange... Can anyone offer any assistance. I'd
be glad to provide further diagnostic information. Here are some config files:

# cat /service/pound/run

#!/bin/sh
echo Starting Pound With Supervise
exec /usr/local/sbin/pound -f /etc/pound/pound.cfg 2>&1

# cat /etc/pound/pound.cfg

# BEGIN pound.cfg
User            nobody
Group           nobody
ExtendedHTTP    0
WebDAV          0
LogLevel        2
Alive           10
ListenHTTP 192.168.1.1,80
ListenHTTPS 192.168.1.1,443 /etc/pound/web2.pem

UrlGroup "/security.*"
        HeadRequire Host ".*:443"
        BackEnd 192.168.1.2,80,1
EndGroup

UrlGroup "/repgen.*"
        HeadRequire Host ".*:443"
        BackEnd 192.168.1.2,80,1
EndGroup

UrlGroup ".*"
        BackEnd 127.0.0.1,81,4
EndGroup

# END pound.cfg

P.S. -- When pound is compiled without the syslogger, it seems like the pound
log entries are pooled and written in chunks. That's normally not the fault of
mutlilog. Is this done by design? If so, can it be changed? If it is being
caused by user-error, what can be done to fix it?

CG



	
		[...]

Chris,

I don't know if you are seeing this problem or not, but if these are IE users,
you may want to tell them to run Windows update.  Aparently Microsoft released
a buggy updage back a few months that they recently released another patch for.
We were seeing a VERY similar problem with lots of our customers, but once they
update, all is good.

Just thought I would share, I don't know for sure this is the problem you are
seeing.

Here is a link to this issue on Microsoft's website:
http://support.microsoft.com/default.aspx?scid=kb;en-us;831167

John D.

********** Original Email *********
** To:   pound(at)apsis.ch
** From: Chris Gamache <cgg007(at)yahoo.com>
** Date: Fri, 7 May 2004 14:08:48
**********

We begin testing Pound, sending requests to one back-end server. Pound is
listening for HTTP and HTTPS traffic. It is being managed by daemontools, so it
was compiled with backgrounding set to off. It was also compiled without
syslog, so that multilog could be used. It works fine for 90% of the people
visiting the website. The remaining 10% are sending no data for Form POST data.
I can't explain it! It is quite strange... Can anyone offer any assistance. I'd
be glad to provide further diagnostic information. Here are some config files:

# cat /service/pound/run

#!/bin/sh
echo Starting Pound With Supervise
exec /usr/local/sbin/pound -f /etc/pound/pound.cfg 2>&1

# cat /etc/pound/pound.cfg

# BEGIN pound.cfg
User            nobody
Group           nobody
ExtendedHTTP    0
WebDAV          0
LogLevel        2
Alive           10
ListenHTTP 192.168.1.1,80
ListenHTTPS 192.168.1.1,443 /etc/pound/web2.pem

UrlGroup "/security.*"
        HeadRequire Host ".*:443"
        BackEnd 192.168.1.2,80,1
EndGroup

UrlGroup "/repgen.*"
        HeadRequire Host ".*:443"
        BackEnd 192.168.1.2,80,1
EndGroup

UrlGroup ".*"
        BackEnd 127.0.0.1,81,4
EndGroup

# END pound.cfg

P.S. -- When pound is compiled without the syslogger, it seems like the pound
log entries are pooled and written in chunks. That's normally not the fault of
mutlilog. Is this done by design? If so, can it be changed? If it is being
caused by user-error, what can be done to fix it?

CG



	
		[...]

Thanks for the heads-up. I vaguely remember this problem. It didn't seem to
affect our users when we were pre-Pound.

Could Pound be exacerbating the problem?

Our set-up was Firewall (HTTP Proxy) --> Pound (Load Balancing Proxy) -->
IIS
when the problem was occurring. We removed Pound from the mix, and everything
went back to normal. No more complaints. Does Pound regularly reset HTTP
connections on Form Post? 

CG

--- John D <jwdavid(at)ibizvision.com> wrote:[...]




	
		[...]

On Monday 10 May 2004 15.21, Chris Gamache wrote:[...]

Pound should not affect POST data - it is treated no different from any other 
request. I would be interested in seeing a bit of traffic before I can say 
more.

BTW, is there a reason for you using another proxy in front of Pound? That is 
exactly what Pound was designed for (and how our clients use it)...[...]

I don't think Pound its self has anything to do with it. I noticed the problem
got really bad when we upgraded our backend servers to something much faster. I
think somehow the speed of the response plays a role with this particular bug.
Least wise that is the only explination I could come up with.

John D.

********** Original Email *********
** To:   John D <jwdavid(at)ibizvision.com>, pound(at)apsis.ch
** From: Chris Gamache <cgg007(at)yahoo.com>
** Date: Mon, 10 May 2004 06:21:51
**********

Thanks for the heads-up. I vaguely remember this problem. It didn't seem to
affect our users when we were pre-Pound.

Could Pound be exacerbating the problem?

Our set-up was Firewall (HTTP Proxy) --> Pound (Load Balancing Proxy) -->
IIS
when the problem was occurring. We removed Pound from the mix, and everything
went back to normal. No more complaints. Does Pound regularly reset HTTP
connections on Form Post? 

CG

--- John D <jwdavid(at)ibizvision.com> wrote:[...]




	
		[...]

--- Robert Segall <roseg(at)apsis.ch> wrote:[...]

How would you like your traffic? :) I'd be glad to post a tcpdump for you to
(personally) download. Let me know what parameters would be most beneficial for
analysis.
[...]

The firewall proxy in front of Pound is a Symantec Enterprise Firewall. It is
our first line of defense. All servers exist behind the firewall. It offers
/very/ granular access control, which we depend on heavily. However, it does
not offer the one thing that we need pound for : intelligent load balancing. We
could do balancing with round-robin DNS, but that doesn't offer high
availability. Pound was so light-weight and specialized it seemed like the
perfect compliment to our network infrastructure. 

I truly appreciate your assistance!

CG


	
		[...]

On Monday 10 May 2004 20.02, Chris Gamache wrote:[...]

My favorite is tcpwatch - stick one instance between the client (or in your 
case the proxy) and Pound, a second between Pound and the relevant
back-end.[...]