Running the JTS recovery quickstart (ArjunaJTS/recovery) sometimes, but not always, throws an org.omg.CORBA.OBJECT_NOT_EXIST exception during recovery. When it does happen the relevant XA resoruces still complete but the corresponding log record is left in the ObjectStore and then moved to AssumedComplete on the next pass. When it does happen I notice that there are two threads replaying phase 2, the one that completes removes the servants (for the XA resources) from the ORBs active object map so that when the other thread replays phase 2 the CORBA object no longer exists.