Hi there,

I'm trying to get Pound to forward requests to 5 zeo clients.  Each
client takes quite a while to become responsive after a restart and
during this time Pound seems to forward to the rest of the clients,
but eventually throws a 500 internal error and then Squid, for some
reason, caches the error.  Squid would wait for  the client became
responsive, but even with weighting it didn't work well as a load
balancer. The site itself is serving about 600000 pages and 1000000
hits per day. Two of the zeo clients live on one server with the zeo
server and the other server has 3 zeo clients.
My pound config is as follows:

User        "zope"
Group       "zope"

LogFacility local0
LogLevel    3
Alive       30

ListenHTTP
        Address 10.0.0.1
        Port    10800

        ## allow PUT and DELETE also (by default only GET, POST and HEAD)?:
        xHTTP           0

        Service
                BackEnd
                        Address 10.0.0.11
                        Port    8011
                        Timeout 180
                End
                BackEnd
                        Address 10.0.0.11
                        Port    8012
                        Timeout 180
                End
                BackEnd
                        Address 10.0.0.12
                        Port    8011
                        Timeout 180
                End
                BackEnd
                        Address 10.0.0.12
                        Port    8012
                        Timeout 180
                End
                BackEnd
                        Address 10.0.0.12
                        Port    8013
                        Timeout 180
                End
        Session
            Type    COOKIE
            ID      "__ac"
            TTL     300
        End

        End
End


Thanks,
Gareth

Hi Gareth,

--On 06 October 2009 11:34 +0200 Gareth Allen <gallen(at)openworld.co.za>

wrote:
[...]

Are you using the patch suggested in 
<https://bugs.launchpad.net/zope2/+bug/143422>
or the new fast-listen 
option? We found that helpful, so that pound wasn't trying to send requests 
to an unresponsive Zope.

I *think* the 500 caching is controlled by the negative_ttl config option 
(<http://www.squid-cache.org/Doc/config/negative_ttl/>)
- the default used 
to be 5 seconds.

Cheers,

Dominic