Currently, if a callback (jpa-style, Lifecycle, Listener) called during during a flush results in an action (as in ActionQueue) being added, that action is ignored until the next flush cycle. I thought we discussed (in st louis) adding a feature to allow those actions to be caught up in the original flush cycle. But I cannot remember the outcome of that discussion as to whether that was something we wanted to look at for 4.0. The solution is essentially a recursive flush. There is obviously a danger there and we would certainly want to depth-limit this. What do y'all think about getting this into 4.0? --- Steve Ebersole <steve(a)hibernate.org&gt; http://hibernate.org