[slaboure]

I've just tested your code under a fresh Branch_3_2 checkout (i.e. jboss 3.2.8RC) and it works correctly: one of the node says:
isMasterReplica = true replicants= [new_value, new_value]


And the other one:
isMasterReplica = false replicants= [new_value, new_value]

Which version do you use? Can you re-verify your setup, etc.?

Cheers,


sacha

[slaboure]

My files if other want to try

[weiqingh]

Sacha, thanks a lot for trying it out.

i initally tried jboss-3.2.0RC5. i just did a cvs checkout by using -r Branch_3_2 jboss-3.2 and got version 3.2.2RC1. it still failed in this build.

how do i get the 3.2.8RC build? i am not familiar with the branch/label scheme in jboss.

[ivelin.ivanov]

I had similar problems when running clustering on Windows. Things got better when I flipped the loopback parameter to true, as advised in the clustering book. The parameter is in clustering-service.xml

[weiqingh]

thanx for the advice. my loopback value has always been set to true. in fact, the default config value is true after i installed jboss.

[ivelin.ivanov]

Just checked in code in the JBoss_3_2 branch that makes the Schedule Providers cluster aware.
If run within a cluster, all providers with the same mbean name and same partition will coordinate among themselves and make sure that only one of them schedules.

As an effect if you have a schedule provider build for a standalone server and you drop it in a clustered environment, it will continue to work the same way as before. It will be deployed on each node, but only one node will provide schedule to its local manager.

There are two mbean attributes: HASingleton (boolean) and PartitionName (string) which can change the default behaviour.

[slaboure]

Great, thank you Ivelin.

Cheers,


sacha

[weiqingh]

is there any "central database" used for synchronization (aka distributed locking) or it's based on node discovery?

basically how much guarantee it has in terms of only one scheduler is run at a time, in cases like there is a network partition.

thanx.

[slaboure]

Network partition management (i.e. disabling members in a partition) is for 4.0