Hi,

I have a WebSpher cluster with several app. servers. I'd like to know what exactly happens when I stop one of the app. servers.

I've noticed that as soon as I try to stop the app. server, the ports asociated with them go down.
Does the server wait for current transactions/sessions to finish?
Does the server failover the current transaction/sessions to another app. server in the cluster?

thank you very much!

when you stop the application , if you dont have session management, all the session which are on that app server will go off , plus there is list of port which comes up when server is up , so when it stop all the ports go down where the application listen

I've set memory-to-memory replication at web web container level. I guees sessions are failovered to another server... but what about transactions? it waits for these to finish before stop completely the app. server?
Anyone knows about any IBM document that involves this kind of behaviours when an app. server in a cluster is being stoped?

The way it works is if you have replication and when the server is stopped , plugin marsk it down and then the request is sent to new server and if you have replication then it the transaction is picked up from there. Each server have cline id when plugin sends the request, and if it is down then the request is fowarded to another one

my 2 cents 

• When you give stopServer , it sends out a message that this JVM is going down to rest of the cluster member. This is handled by DCS. 


• Any active session will be transferred to another available member [when you have session replication enabled]


• Now all the members of the cluster and mapped web servers remove this member from their view. Hence no more new requests. 


• The transaction manager will wait before it ends, for any inflight transaction to complete. In case of an failure/abrupt stop .. it mark the transaction for recovery when the server starts again. 


• now it stops all the applications running on that JVM

Notes:

Before you give stopServer command, try these :

• do a tail -f on nodeagent and servers's SystemOut logs

1. You can see what DCS is doing in the nodeAgent's systemout log. This shows dcs removing member from the view.


2. You can see the sequence of server stop, session handover and any transactions marked for recovery in the JVM's systemout log. 

 

Perfect! Thank you very much!