This forum has been moved here:
Helicon Tech Community Forum

Helicon Ape (Forum Locked Forum Locked)
 Helicon Tech : Helicon Ape
Subject Topic: loadbalancing with stickysession
Author
Message |
henknijveld
Newbie


Joined: 19 July 2010
Location: Netherlands
Posts: 5
Posted: 19 July 2010 at 2:03pm

Hi Everyone,

I have a little problem/discusion over loadbalancing with APE. And I hope that you guy's can help me out.

Situation:
We have a IIS6 frontend, and BEA Weblogic backend. Running a backend application(website), and some webservices on that platform. BEA has a own reverse proxy tool (iisproxy.dll) what is running fine right now, but... on our acceptance environment, the functionalty is broken (probably by windows updates), and we want to move our environmnet away from BEA to JBOSS. This project is still running, and the support of oracle has not been extent (because of the migration).

We also have some .NET webservice reverseproxyed with isapi rewrite.

so we think that is is time for us to move to APE? agree?

The problem is, that we have more backend applictions (more than 10), that we cannot touch. How can we solve this.

httpd.conf:
<location /application1/>
 RewriteEngine on
 RewriteMap hosts rnd:routing.txt
 Setenv my-host ${hosts:application1}
 RewriteBase /
 RewriteCond %{HTTP_COOKIE} !balanceid [nc]
 RewriteRule .* - [CO=BALANCEID:balancer.%{my-host}, R, L]
 ProxyPass balancer://cluster/ lbmethod=byrequests stickysession=BALANCEID
 ProxyPassReverse balancer://application1/
</location>

<Proxy balancer://application1>
 BalancerMember http://server1/applicaiton1 route=server1
 BalancerMember http://server2/application1 route=server2 
</Proxy>

routing.txt:
application1 server1|server2

What happens under the hood. The first time a browser is requesting the application1, and there is no cookie been set, the cookie is made of the random values of the textfile. and than proxy'ed. the next time, when the cookie is available, the cluster config takes over. BUT...

The whole trick works if both nodes are online.

ps: I have also a config, that push a 302 to the client to set the cookie, to make sure the balancer is working with the cookie the very first request. (same problem when the node is off-line)

questions
1 Is there a way to "look" at the subnodes and test if this node is online, and try an other one?
2a bea weblogic has somthing called sessionid, but the layout is different than apache accept. "sessionid=<sessionnumber>!<serverid>". maybe we could use this also to route in the balancer if we can change/add a cookie and replace the "!" with a "." to route. question b: if this is possible, is it
3 something else I overlook?

If somebody can help us out with this, I would be very thankfull.

Regards,

Henk

Back to Top
 
Anton
Admin Group


Joined: 30 January 2007
Location: Ukraine
Posts: 10519
Posted: 22 July 2010 at 5:27am

Hello, Henk

Sorry to make you wait for reply for so long.
We were testing the feature you are concerned about.
Unfortunately, currently the ballancer works properly only with both nodes online.
The correct behavior in case one of the nodes is offline will be implemented in the nearest builds within a couple of weeks.
We will inform you as soon as it happens.

__________________
Regards,
Anton
Back to Top
 

Sorry, you can NOT post a reply.
This forum has been locked by a forum administrator.

Printable version Printable version