It was found that if you use a cache which is transactional and has cluster listeners that if you try to perform an operation with the SKIP_LISTENER_NOTIFICATION flag the listeners are still notified to some extent.

It isn't clear how listener notifications should work in a TX in the first place. It is possible that a subset of operations may have the flag and others may not. And also it is completely possible the FLAG is set on a commit itself but not on the begin, which seems very problematic.

My take as to how to solve this would be to save the flag with each command individually and completely ignore the flag for transaction begin and transacation end notifications (always raising them).

We could have special operating for an implicit TX as well where we could safely ignore the TX begin/end notifications if the single operation has the SKIP_LISTENER_NOTIFICATION flag set.