On Wed, Jun 27, 2012 at 6:53 PM, Rusk, Patrick <Patrick_Rusk@ssga.com> wrote:

We recently upgraded our Infinispan from 4.2 to 5.1.5. A behavior that we encountered afterwards was that some of our important replicated caches would only have partial state after their initial state transfer was complete. The rest of the state never seems to make it over, either.

 

Under 4.2, if any keys became available in the cache, they were all there. We have always considered full state transfer to be appropriate for our app because our caches are small and all of the data is needed for our rules engine calculations.

 

Is there any discussion of this change in behavior in the documentation or forums? Is it driven by parameters that we might be unaware of?

 


Pat, we have NOT introduced partial state transfer. What did make one change in 5.1, to let each existing member of the cluster send a part of the data to the joiner, instead of the coordinator sending everything. But the initial getCache() call on the joiner should not return until it did receive all the data from all the members. Did you see any errors in the logs of any of the servers?

If you're able to reproduce this behaviour, please create an issue in JIRA (https://issues.jboss.org/) and attach your configuration and logs, with TRACE enabled for org.infinispan. Of course, if you could add a small test that reproduces the issue, that would be great.

Cheers
Dan