Although this should never happen in the ideal world, it happens from time to time. If a (slightly) different application is deployed to one of the cluster nodes, the current replication mechanism does not work any longer. The failure is not obvious and difficult to debug. We should consider adding some extra information (e.g. a hash of BeanIdentifierIndexes) to serialized messages and fail fast if this situation is detected. We should seek a good compromise between the size of the extra information and the probability it detects a problem.
Problems like these should be detected by the following test:
org.jboss.weld.tests.unit.cluster.NaiveClusterTest.testVariableBeanDeploymentStructure() although due to a different error it does not happen at the moment.