To address a leak with some resource adapters, WildFly is de-listing resources during the (IronJacamar) Synchronization.beforeCompletion time, however this can happen before all of the persistence providers have saved their changes to the database. This is likely to be the case for applications that have more than one persistence unit used per transaction. The impact is that only the first referenced persistence provider, is part of the transaction. The other database changes will run as local database changes. This was reported recently via [1] + [2] + [3]. Discussion about how to best address is ongoing (would be nice to have both the resource leak fix and a way to run the IronJacamar Synchronization.beforeCompletion last). This is a blocking issue for any jiras that we create to handle this. This problem is not in earlier application server versions (only in WildFly 8.0 and higher). Workaround: Using a single persistence unit per transaction should workaround this but a code fix will be better. Scott [1] WF user forum description of problem https://community.jboss.org/message/901613 [2] Narayana dev mailing list post about sync ordering https://community.jboss.org/message/901880 [3] original jira that reported this issue that I rejected https://issues.jboss.org/browse/WFLY-3619