an interesting point came up at work today...
the guys here have some Session Facades over an MBean service on JB*ss.
The service was created and started OK, then at a later date got itself into an inconsistant state - it needed redeployment, and might even have been able to signal this to it's container, but there was no mechanism for it.... - requests kept coming in through the facades that could have been balanced elsewhere in the cluster...
I guess I am proposing some mechanism whereby a component can signal it's broken-ness to it's container which can then try e.g. redeploying it, or if that fails, undeploying it and notifying an administrator. Undeploying it would have had a knock on effect down the dependency tree and undeployed it's facades and future client requests would then have been directed to a fully functional node and satisfied - everyone is happy.
I know we are looking to implement mandatory fn-ality ASAP and not get sidetracked into creeping feature-itis, but I would like to just put the germ of an idea at the back of peoples minds so that space might be left for something like this....
Cheers,
Jules
P.S.
Of course it is possible that the component may be so badly broken that it is not aware that it needs killing off - perhaps we could have some sort of periodic test that the container could apply - if a component fails it it gets re-started/deployed...
-- /********************************** * Jules Gosnell * Partner * Core Developers Network (Europe) **********************************/
