Plz help!! idempotent methods stateful session beans? (failover)

Question

          plz help!! idempotent methods stateful session beans? (failover)
          Hi there,
          Im trying to failover my shopping cart implemented using stateful ejb, the replication
          is working but the failover is not (Object not found exception....).
          By design is hard to think on an stateful ejb using idempotent methods, but after
          checking the documentation in detail im completely confused, so im starting to
          wonder if i should use idempotent methods or manual retry in case one server is
          down.
          Plz take a loo at these explanations about faiolver in the WLS edocs.
          "....With clustered objects, automatic failover generally occurs only in cases
          where the object is idempotent.....Because of this, replica-aware stubs will not,
          by default, attempt to retry a method that fails after the request is sent but
          before it returns. This behavior can be overridden by marking a service idempotent..."
          BUT
          "By default, a stateful session bean's Home stub provides load balancing and failover
          for its method invocations to any clustered server where the bean is deployed...."
          So is possible to achieve automatic failover (not only load balancing) for stateful
          ejb without having to worry about state-changes or manual retry. (i know the retry
          is requiered if the server crashed just before finishing a transaction but this
          is not what im talking about).
          Any help will be highly appreciated
          Alan
          

Answer 1

          Ryan,
          Yes, im deploying using In-memory replication and clusterable elements, when try
          to invoke the method (after one of the nodes is down) i recieve an "java.rmi.NoSuchObjectException:
          Bean has been deleted".
          I believe all this could be related to some sort of issue with JNDI Lookup in
          the cluster, but haavent been able to get a clue to make it work....(almost desesperate
          by now)
          Alan
          "ryan upton" <[email protected]> wrote:
          >Alan,
          >
          >If the state isn't maintained during failover I would suspect a
          >configuration error. Have you set the <replication-type> element within
          ><stateful-session-clustering> tag in the bean's deployment descriptor?
          >
          >"Alan" <[email protected]> wrote in message
          >news:40e17cf7$1@mktnews1...
          >>
          >> Ryan,
          >>
          >>
          >> Maybe i didnt explain myself, in the scenario im dealing with I KNOW
          >the
          >method
          >> finished succesfully (example, the user succesfully added an item to
          >a
          >cart).
          >>
          >>
          >> Just after that one of the nodes goes down, and the user decides to
          >add a
          >new
          >> item (new invocation , not a retry of failed one) on the same session.
          >>
          >> This is the case in which failover should work automatically as you
          >say,
          >meaning
          >> the new method invokation should be able to retrieve the session state
          >info from
          >> the replica and switch the primary.....Well, this is the part wich
          >isnt
          >working
          >> :( as i understand it should transparently
          >>
          >> I believe could be a config issue , but have no way to make it work.
          >>
          >> Regards,
          >>
          >> Alan
          >> "ryan upton" <[email protected]> wrote:
          >> >
          >> >"Alan" <[email protected]> wrote in message
          >news:40e06296@mktnews1...
          >> >>
          >> >> Ryan,
          >> >>
          >> >> Thanks for clearing the both conditions a bit more, in this case
          >my
          >> >issue
          >> >has
          >> >> to do with the scenario #1 in which subsequent calls (invocations)
          >> >go to a
          >> >node
          >> >> which isnt avaliable(down), do you have any idea of in what conditions
          >> >the
          >> >"automnatic"
          >> >> failover would fail??? (meaning why the stub would try to dispatch
          >> >a call
          >> >to the
          >> >> same previous node)
          >> >>
          >> >> Thanks for the help again
          >> >>
          >> >
          >> >Subsequent calls won't go to a node that's down. That's the whole
          >> >point
          >> >;-). Failover is always automatic, what you are getting confused
          >on
          >> >is
          >> >automatic method re-execution upon failover which happens if the method
          >> >has
          >> >been marked as idempotent. The logic is this: if I failover to an
          >EJB
          >> >that
          >> >I know is still alive, I can't safely call the same method again.
          > Why?
          >> >Because I don't know how much of the method completed the last time
          >before
          >> >the previous bean's failure. So if the bean can yield multiple results
          >> >from
          >> >multiple invocations I had better play it safe and not re-execute
          >the
          >> >method
          >> >on the bean I failed over to (unless it's idempotent because in that
          >> >case it
          >> >doesn't matter how many times I call it, it always does the same thing).
          >> > In
          >> >scenario #1 and #2 failover is automatic and all new calls are executed
          >> >against the bean the client failed over to, but only idempotent methods
          >> >are
          >> >re-executed.
          >> >
          >> >
          >>
          >
          >