Lost (and found) JMS - messages

Question

          Using WLS 6.1 and have experienced that JMS messages has not been delivered to
          its consumer correctly. We are interested in any suggestions to what has happend,
          what can be done to avoid the problem in the future, or perhaps how we can find
          out more about the problem.
          We have a message driven bean A which sends messages to the queue Q, of which
          the message driven bean B is the only consumer.
          What we recently have experienced is that some of the messages delivered by A
          to Q never are delivered to B. But the messages seems to be persisted correctly,
          because after a restart they reappear!
          Following is a more detailed description of the scenario:
          1. A message is sendt with DeliveryMode=PERSISTENT and TimeToLive=0. QueueSender.send
          is invoked and returns normally.
          2. The consumers never gets the message (B.onMessage i never invoked with this
          message).
          3. Point 1-2 is repeated for 13 different messages over a period when ~100 similar
          messages are delivered normally.
          4. The server is then restarted.
          5. In the starup log it is indicated that 44 records are found in the JMSStore.
          6. When the server restarts, the "lost" messages gets delivered to B.
          After the server restart we have not experienced the problem again.
          Logs does not show anything unusual about the previous restart.
          Other details:
          There is always exatly 1 instance of A running, and 5-20 instances of B. A implements
          and uses a pool of javax.jms.XAQueueSession instances, which shares a single instance
          of javax.jms.XAQueueConnection. The aknowledge-mode for B is "Auto-aknowledge".
          We're using Weblogic Server 6.1.3.0 / Red Hat Linux 7.3.
          Any help on this problem is appreciated!
          

Answer 1

          I believe we've seen just the behaviour you describe on 6.1 SP2, but only on a
          multi-cpu box (Solaris), and only under load. We can't reproduce the problem reliably,
          and get no error messages/warnings anywhere, so very little to send to customer
          support; by the time we notice this, the problem's long passed, so a thread dump's
          of no use.
          We're going to try and see how SP5 (which has loads of JMS fixes) pans out.
          But if you find anything out, could you post here?
          simon.
          "Harald Stendal" <[email protected]> wrote:
          >
          >Using WLS 6.1 and have experienced that JMS messages has not been delivered
          >to
          >its consumer correctly. We are interested in any suggestions to what
          >has happend,
          >what can be done to avoid the problem in the future, or perhaps how we
          >can find
          >out more about the problem.
          >
          >We have a message driven bean A which sends messages to the queue Q,
          >of which
          >the message driven bean B is the only consumer.
          >What we recently have experienced is that some of the messages delivered
          >by A
          >to Q never are delivered to B. But the messages seems to be persisted
          >correctly,
          >because after a restart they reappear!
          >
          >Following is a more detailed description of the scenario:
          >
          >1. A message is sendt with DeliveryMode=PERSISTENT and TimeToLive=0.
          >QueueSender.send
          >is invoked and returns normally.
          >2. The consumers never gets the message (B.onMessage i never invoked
          >with this
          >message).
          >3. Point 1-2 is repeated for 13 different messages over a period when
          >~100 similar
          >messages are delivered normally.
          >4. The server is then restarted.
          >5. In the starup log it is indicated that 44 records are found in the
          >JMSStore.
          >6. When the server restarts, the "lost" messages gets delivered to B.
          >
          >After the server restart we have not experienced the problem again.
          >
          >Logs does not show anything unusual about the previous restart.
          >
          >Other details:
          >There is always exatly 1 instance of A running, and 5-20 instances of
          >B. A implements
          >and uses a pool of javax.jms.XAQueueSession instances, which shares a
          >single instance
          >of javax.jms.XAQueueConnection. The aknowledge-mode for B is "Auto-aknowledge".
          >
          >We're using Weblogic Server 6.1.3.0 / Red Hat Linux 7.3.
          >
          >Any help on this problem is appreciated!
          >