Do I need 2 clusters for upgrades with fail-over?

[Gal Robert <ro...@gmail.com>]

Hi,

this thread refers to 'Silent runtime replace of a class' thread.

We need to provide application upgrades, where users work won't be 
interrupted. We also need fail-over future for application, meaning, if
one of the appservers goes down, users won't lost their work.

We need the simplest solution, but since we don't have any experience 
with fail-over future, we'd like to find out the simplest solution by 
your help.

We've read some documents and designed solution with these components:

1) front-end software-based load balancer (probably tomcat with the 
balancer webapp).
2) main cluster A with two tomcat instances
3) offline spare cluster B with two tomcat instances
The whole configuration is on a single machine.

The upgrade scenario will look like this:
1) bring up, deploy and test the new application on the spare cluster
2) modify the load balancer policy to redirect new user sessions to the 
spare cluster (with upgraded app)

So all the new sessions will run on the upgraded cluster. After the 
number of active users on the main cluster drops to zero, it's safe to 
stop it; so the roles of cluster will change.
To ensure fail-over future, in-memory session replication will be enabled.

The main reason to divide the system into two cluster is because of the 
fail-over future, more precisiously the session replication.
If there will be only one cluster, the session replication won't work 
correctly between the upgraded/old app. Do you agree?

Do you have any recommendations/ideas for a simpler setup with 
upgrade/fail-over futures? Any ideas, points to articles is welcomme!

--robert




---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org