[ https://issues.jboss.org/browse/JBTM-2356?page=com.atlassian.jira.plugin.... ] Michael Musgrove commented on JBTM-2356: ---------------------------------------- Gytis tracked it down. The bug is in the afterCompletion call where the coordinator does cleanup - the (rest-at) coordinator maintains an in memory map of active transactions (and a one time read of the recovering transactions from disk at boot time). In this case it should be checking the transaction status before deciding to remove its in-memory copy of the transaction uid. So when a participant asks for the status the coordinator has no knowledge of it. We didn't spot this in our tests because we only test crash recovery on a single server and in this case on restart the coordinator repopulates its view of recovering transactions. Gytis has written a 2 server arquillian test to catch these cases in the future. -- This message was sent by Atlassian JIRA (v6.3.11#6341)