2 Replies Latest reply on Dec 17, 2015 2:18 AM by Daniel Nuss

    Messages on clustered topic occasionally are not received when sent from specific cluster node

    Daniel Nuss Newbie



      i'm using the following scenario:

      - 2 JBoss7 nodes, each configuring a colocated Hornetq Live/Backup-server pair

      - A clustered JEE application, deployed on both nodes

      - Multiple clients, sending some EJB calls to the clustered application, receiving JMS notifications on a clustered topic


      The scenario works fine so far, but every now and then a strange scenario occurs:

      The clients seems to stop receiving messages sent from one of the cluster nodes. Specifically, they are no more receiving messags from the server to which they are NOT connecting when creating the listener on the topic (e.g. when they initially make a connection to server1, they stop getting messages sent from server2). The same behaviour can be observed when trying to connect on the same topic with MDBs deployed in the application: In case i deploy a MDB within the application on server1 and server2, the messages which are not received by the clients are always received by the MDB on server2. If i disable this MDB on server2, so that it is only enabled on server1, the messages are not received anymore by this MDB. So it looks like whatever client is connected to the topic, it doesn't get the clustered one but gets only the connection to the node itself. In most cases, a server restart fixes this bug (but i did not yet figure out an automated way to discover sth. goes wrong without implementing some kind of lifecycle heartbeat mechanism, so manual monitoring and restarting is involved - very unwanted..)

      Do you have experienced such behavior before? Do you have any hints to isolate the error which causes this scenario? Or is this maybe even a known bug or some kind of misconfiguration? (i doubt that it is a misconfiguration because it is working in 99% of the time, additionally i can see in the log that the HornetQ servers are clustered and no failover of the backup server occurs or sth. like that).



      I really appreciate any help!