Basically the setup consists of two tomcat6 and two memcached nodes. memcached session manager version - 1.8.3 (latest release).
I'm testing the following scenario:
- Client logs in into the application using load balancer url
- Load balancer is dispatching the request to one of the tomcat nodes.
- Tomcat saves the session information into the both memcached nodes.
- One of the nodes is the primary node for that session.
- The primary node goes down to test the failover of the memcached nodes.
- The session is restored from secondary node.
- The former primary node goes back up.
After it I'm expecting that the next user request will trigger the session back up to former primary node and this node will become secondary. But this is not what actually happens as tomcat didn't recognise that former primary node is back to normal. Session ocasionally gets backed up but some time has to pass before it will make a health check (ping check) and tomcat will get the information that node is up again.
The question is where I can configure the value of property controlling the timeout so the nodes that are got back to up state will be recognised without delay or with minimal one?
