I installed pound on a Fedora Core1 box, everything ran fine in testing, 
but when I changed the NAT rules to direct traffic through the proxy, I 
saw many errors:

On some clients web pages simply wouldn't load
Some clients displayed a web page with the following error:
   libgcc_s.so.1 must be installed for pthread_cancel to work

In the logs, some requests seemed normal but there were also errors such as:
   Dec 14 18:15:44 proxy pound: error read from 66.180.233.5: 
Input/output error
   Dec 14 19:14:22 proxy pound: error flush headers to 24.61.185.71: 
Connection reset by peer
   Dec 15 10:00:39 proxy pound: error copy chunk cont: Connection reset 
by peer

Here is my pound.cfg:

ListenHTTP *,80
User nobody
Group nobody
RootJail /chroot/pound
ExtendedHTTP 0
WebDAV 0
LogLevel 2
Alive 30
Server 0
Client 10
RewriteRedirect 1

UrlGroup "weather"
   BackEnd 192.168.168.51,80,5
   BackEnd 192.168.168.126,80,5
EndGroup

UrlGroup ".*"
   BackEnd 192.168.168.32,80,5
   BackEnd 192.168.168.48,80,5
EndGroup

Any ideas?
[...]

Garrison Hoffman wrote:[...]
Do You have libgcc installed?

Best Regards
Maciej Bogucki

Maciej Bogucki wrote:
[...][...]

Naturally that was the first thing I checked, I should also have 
included this info in my first post.

I have libgcc-3.3.2-1 installed and the library in question is at 
/lib/libgcc_s.so.1
[...]

On Thursday 16 December 2004 02.07, Garrison Hoffman wrote:[...]

What do you mean with NAT? Is there a difference between clients coming in 
from inside the firewall and the ones from the outside?
[...]

I assure you this does NOT come from Pound - there just isn't anything like 
that in the code. Are you sure your NAT rules go to the Pound box?
[...]

These are normal informative messages - timeout on the client or a client that 
closed the socket before the reply could be sent. Nothing to worry about 
here.
[...]

I strongly suggest you remove the RootJail and try again - we have seen quite 
often problems with incorrectly set up environments (such as 
missing /dev/urandom or /dev/syslog, or some missing libraries). If that 
works correctly try adding whatever might be missing until you hit the right 
combination - that depends very much on your particular system and libraries.

As an aside: there's not much point in defining two back-ends with identical 
priorities (5 in your case) - you'll get the same results with priority 1.[...]

Robert Segall wrote:
[...]

There are no LAN side clients; all the servers are on a private subnet, 
so instead of waiting for DNS to propagate I simply changed the NAT to 
feed web server traffic through the proxy.
[...]

OK, thanks.
[...]

Eliminating the chroot solved it, though I'm not sure why it worked (or 
seemed to) on a few test clients, then failed under load after the NAT 
change.  All this may be moot as some problems have been discovered on 
the secondary web server.
[...]

My thinking was that from 5 I can increase or decrease priority, kind of 
a neutral setting.  Out of curiosity, what does pound do when all hosts 
are responding and of equal priority?  Does it default to a round robin 
method, or perhaps attempt to guage load by response time?[...]

On Thursday 16 December 2004 23.42, Garrison Hoffman wrote:[...]

Neither. Pound always distributes requests in a random way (linear uniform), 
except back-ends are weighted according to their priorities.[...]