What is the transactional behaviour with multiple JDG caches in an embedded clustered environment?

Issue

• The XML configuration contains more transactions modes than the programatic API, how can I set the different modes for the transation?
• We found inconsistent results if a transaction failed and multiple caches and instances are involved, how this can be avoided?
• The transaction flag useSynchronization has a direct effect whether Exceptions are hidden if multiple instances are involved.