Hi,

I did try searching the archives but no clear information on this problem:

I have a service with 3 backends on which I'd like to do some session failover
testing. Here's the relevant pound configuration:

LogLevel 2
Alive 5
TimeOut 15
Grace 15
Control "/tmp/pound.socket"

ListenHTTP
  Address 172.16.10.140
  Port    80

Service
        Session
                Type    Cookie
                ID      "JSESSIONID"
                TTL     1920
        End

        HeadRequire "Host: .*www.lab.*"

        BackEnd
                Address 10.10.10.1
                Port    9001
                TimeOut 5
        End
        BackEnd
                Address 10.10.10.2
                Port    9001
                TimeOut 5
        End
        BackEnd
                Address 10.10.10.3
                Port    9001
                TimeOut 5
        End
End
End

But when I use kill the backend, sessions that are persisting to that node
continue to send requests that way, even though the node is down and will never
process that request.

Is there some way to configure the session persistence so that if the node a
session is persisting on dies, then pound re-selects a different node from the
cluster to service the session?

If not, how does one get persisting sessions to failover to other cluster nodes
in the event of cluster-node failure? I cannot get pound to simulate this
behavior, so any help is appreciated.

Thanks,
Steve.
Senior SaaS Administrator
HALOGEN SOFTWARE INC.
Employee Performance and
Talent Management Solution of Choice
Tel:  613-270-1011 Ext. 294 | Fax:  613-482-5045
Cell: 613-795-2559

[cid:image001.jpg(at)01C92E01.5773ED70]<http://www.halogensoftware.com/about/testimonials_video.php>

> Is there some way to configure the session persistence so that if the 
node a session is persisting on dies, then pound re-selects a different 
node from the cluster to service the session?

Maybe we need to excute "poundctl -n" manually.

 > If not, how does one get persisting sessions to failover to other 
cluster nodes in the event of cluster-node failure? I cannot get pound 
to simulate this behavior, so any help is appreciated.
 >

How about change the "TTL" in Session directive?
more short.

[...]

On Tue, 2008-10-14 at 13:35 -0400, Stephen Hillier wrote:[...]

You have set the grace period to 15 seconds. Set it to 0 if you want the
sessions to be invalidated immediately.[...]