Container-managed / bean-managed transaction demarcation

Question

I am trying to make sure I understand container-managed and bean-managed transaction demarcation and in particular where you have one bean calling another bean. What happens where one of the beans has container-managed transaction demarcation and the other bean-managed transaction demarcation. In fact the initial question to ask is, is this allowed?
Lets use an application scenario to illustrate the issue. The application has a payment transaction. Payments can be received in one of two ways:
1. As a payment at a branch where the individual payment is processed on a client application and resulting in the processing of a single payment transaction.
2. As a batch of payments received from a bank containing, potentially, thousands of payment transactions.
The proposed implementation for this uses two session beans. The first is a Payment session bean that implements the business logic as appropriate calling entity beans to persist the change. The second is a BatchPayment session bean. This processes the batch of payment transactions received from the bank. The BatchPayment reads through the batch of payments from a bank calling the Payment session bean for each payment transaction.
Lets look at the transactional properties of both session beans. In order to support the client application the Payment session bean can implicitly enforce transactional integrity and is therefore set to container-managed transaction demarcation. However the BatchPayment session bean will want to explicitly specify transaction demarcation for performance reasons. The transactional "commit" process is relatively expensive. When processing a large batch of transactions rather than performing a commit after every transaction is processed we want to perform the commit after a number of transactions have been processed. For example, we may decide that after every 100 transactions have been processed we commit. The processing will have a shorter elapsed time as we have not had to perform 99 commit processes. So the BatchPayment session bean will want to explicitly specify its transaction demarcation and will therefore be defined with bean-managed transaction demarcation.
How would this be implemented? A possible solution is:
Payment session bean implemented with container-managed transaction demarcation with transaction scope set to Required.
BatchPayment session bean implemented with bean-managed transaction demarcation with transaction scope set to Required.
When the client application is run it calls the Payment bean and the container-managed transaction demarcation ensures the transactional integrity of that transaction.
When a BatchPayment process is run it explicitly determines the transaction demarcation. Lets say that after every 100 Payment transactions (through 100 calls to the Payment session bean) have been processed the BatchPayment bean issues a commit. In this scenario however we have mixed container-managed and bean-managed transaction demarcation. Hence my original question. Can container-managed and bean-managed transaction demarcation be mixed? If not how is it possible to implement the requirements as described above?
Thanks for any thoughts.
Paul

Answer 1

BatchPayment session bean implemented with bean-managed transaction demarcation with transaction scope set to Required.Didn't quite understand this sentence.... if it's BMT it has no declarative transaction attributes such as "Required"....
Anyway, first of all I'll have to ask, Why at all would you want to commit in the middle of the business method? to get as much through as possible before a potential crash? :-)
Can container-managed and bean-managed transaction demarcation be mixed?Yes, of course. Just remember that the "direction" you are refering to ->
a BMT SB that propagates it's transaction to a method in a CMT SB that is demarcated with "Required" is the simplest case. If it were "reversed", or for that matter any BMT that might be called within an active transaction context must perform logic to manipulate the transaction state. For instance(and most common case), checking to see if a transaction is active and if so not to do anything(just use the one that is already active).
If not how is it possible to implement the requirements as described above?You could also implement this scenario with CMTs all the way through. your BatchPayment SB could consist of two methods, one (say, execute(Collection paymentsToExecute) ) with "Supports", and another(say executeBatchUnit(Collection paymentsToExecute, int beginIndex, int endIndex) ) with "RequiresNew".
then have the first just call the other with indexes denoting each time a group of payments.
Still, it does seem more suitable using BMT for these kind of things.....
Hope this helped....