Pending message redelivery wait period after server startup

Question

Hi, I have seen an strange behaviour in the way pending JMS messages are
          redelivered to MDBs at server startup.
          In my unit test I have seen that, killing Weblogic during an MDB code
          execution as expected makes the container to redelivery unconfirmed message
          at server startup. I have also seen that it's redelivered about 4 minutes
          after server is started (more info of this in thread: "Problem with
          persistent messages and MDBs" 25 October 2002)
          After a recent crash we have had on the system in wich there were a lot of
          messages pending to deliver to an MDB we saw that they were being
          redelivered just after the MDBs were deployed while system was starting. Of
          course due to the fact that some components needed by the MDB were not
          deployed by that time, a lot of exceptions were raised making messages to be
          queued again and promptly delivered wich caused more exceptions. Just a big
          mess!
          I'm tring to figure a good work-around about this, what I wanted to know is
          how is implemented the "delay" to send messages after server startup. Is it
          done by comparing timestamp of message (when did it enter the queue and
          thereby persisted on the store) with current system time?. That approach of
          course will fail if any of the messages were sent before that "delay" time.
          If this is the case and there isn't a fix for it this is what I have in
          mind:
          Each MDB will have a private attribute wich will be used to detect when
          system has completely started. By default it will be false and after a
          message is sent to the MDB the first line of the onMessage method will test
          wether system has started or not (by checking weblogic connection port i.e.
          7001. Btw. is there any better way of checking server has finished start-up)
          if not, a runtime exception will be launched that will force message to be
          redelivered (without trying to execute any MDB code), I think setting a
          proper redelivery delay i.e. 30-60 secs. will allow server to startup with
          less problems (in the situation I described above JVM crashed with an 11
          system signal)
          Any other ideas to deal with this?
          Thanks in advance.
          Regards.
          Ignacio.
          

Answer 1

Hi Ignacio,
          I think you can usually control the order in which ejb's are booted,
          through a combination of console settings and
          meta file entries. I suggest posting to the ejb newsgroup
          to find out how.
          The four minute delay you see is normal for resolving
          interrupted transactions, but I believe that an enhancement
          was put into a 7.0 SP to reduce or eliminate this interval,
          and that 8.1 has no interval. Post to the transaction
          newsgroup for more info, or contact customer support.
          Given that you referenced earlier posts on this issue, you
          probably already know how to tune the interval down to one
          minute even without the enhancements.
          There is no other delay for sending messages at system startup,
          as soon as the MDB is booted it attaches to JMS, and JMS
          starts delivering messages whose transactional state is known.
          Tom
          P.S. I suppose one way to detect end-of-boot is to
          place a startup class last in the boot order.
          P.P.S. For MDBs one alternative is to code a "sleep" in the onMessage if
          an unavailable resource is detected - this
          is hokey, but works as long as you have made sure that
          max-beans-in-free-pool totals do not come near or exceed
          available thread pool size.
          Ignacio G. Dupont wrote:
          > Hi, I have seen an strange behaviour in the way pending JMS messages are
          > redelivered to MDBs at server startup.
          >
          > In my unit test I have seen that, killing Weblogic during an MDB code
          > execution as expected makes the container to redelivery unconfirmed message
          > at server startup. I have also seen that it's redelivered about 4 minutes
          > after server is started (more info of this in thread: "Problem with
          > persistent messages and MDBs" 25 October 2002)
          >
          > After a recent crash we have had on the system in wich there were a lot of
          > messages pending to deliver to an MDB we saw that they were being
          > redelivered just after the MDBs were deployed while system was starting. Of
          > course due to the fact that some components needed by the MDB were not
          > deployed by that time, a lot of exceptions were raised making messages to be
          > queued again and promptly delivered wich caused more exceptions. Just a big
          > mess!
          >
          > I'm tring to figure a good work-around about this, what I wanted to know is
          > how is implemented the "delay" to send messages after server startup. Is it
          > done by comparing timestamp of message (when did it enter the queue and
          > thereby persisted on the store) with current system time?. That approach of
          > course will fail if any of the messages were sent before that "delay" time.
          >
          > If this is the case and there isn't a fix for it this is what I have in
          > mind:
          >
          > Each MDB will have a private attribute wich will be used to detect when
          > system has completely started. By default it will be false and after a
          > message is sent to the MDB the first line of the onMessage method will test
          > wether system has started or not (by checking weblogic connection port i.e.
          > 7001. Btw. is there any better way of checking server has finished start-up)
          > if not, a runtime exception will be launched that will force message to be
          > redelivered (without trying to execute any MDB code), I think setting a
          > proper redelivery delay i.e. 30-60 secs. will allow server to startup with
          > less problems (in the situation I described above JVM crashed with an 11
          > system signal)
          >
          > Any other ideas to deal with this?
          >
          > Thanks in advance.
          >
          > Regards.
          >
          > Ignacio.
          >
          >