Making the application cluster-aware

Question

Hi,
          I have just been assigned a multi-tier J2EE application with the task of
          making it "work in a cluster". The earlier architects (if i may use the
          term), just "assumed" that weblogic would provide support for
          high-availability and they would not need to do any code changes.
          So I have this application that has about 100 EJBs - about 40 Entity Beans,
          only two SFSB and the rest are SLSB.
          What would be the best strategy to go about making this application
          cluster-aware?
          I could think of the following points so far:
          1. Identify idempotent methods in all EJBs.
          2. Make all session-stored objects serializable in the web-tier. Make sure
          objects are not too huge to be serialized (don't know if I would be actually
          able to modify - too late for that).
          3. Modify bean descriptors to make them clusterable.
          4. Make web app clusterable.
          5. Look for code that uses non-replicable services (file system access,
          sockets, etc.).
          I'm sure there must be more that I've missed. Anything more that needs to be
          added to the list above? Would there be a scenario where the EJB clients
          (other beans themselves, the web tier, RMI clients, etc) would need to be
          changed?
          I understand identification of idempotent methods would be manual (or is
          not?). But what would be the best way to modify the deployment descriptors
          of all these beans?
          TIA
          With Warm Regards,
          Manav.
          

Answer 1

Hi Manavendra,
          By default all EJBs are clusterable because the default value of
          <home-is-clusterable/> is true and <stateless-bean-is-clusterable/> default
          is true. For stateful session EJBs, you will have to configure the
          replication type to InMemory as it defaults to none. Unless you have
          specifically added these elements to turn off clustering, they should
          already be clusterable.
          Typically, you will find performance is better if the web app and EJBs are
          deployed in the same JVM, because there are no remote calls. By default
          <enable-call-by-reference/> is set to true which allows arguments to be
          passed by reference rather than copying parameters. If calls are remote,
          then <enable-call-by-reference/> must be false as pass by value will always
          be necessary in this case.
          Jane
          "Manavendra Gupta" <[email protected]> wrote in message
          news:[email protected]...
          > Jane,
          >
          > I appreciate you taking out time to reply.
          >
          > <headers trimmed>
          > > > 1. Identify idempotent methods in all EJBs.
          > >
          > > You only need to do this if your EJB clients are remote. If the client
          is
          > a
          > > web app deployed in the same JVM as your EJBs, you don't need to worry
          > about
          > > it since WebLogic will create the EJBs on the same JVM as the web app.
          > Like I said, its a multi-tier application, so there is the web-tier making
          > calls to the EJB tier (though thankfully, there is a EJB "API" layer -
          that
          > consists of multiple Session Facades, ensuring network calls are
          minimized).
          > I'm not sure if the web-tier and the EJB-tier in the same application are
          in
          > the same VM in weblogic?
          >
          > Finally, I might need to separate the two tiers physically, though I'm
          still
          > trying to find out the parameters on which I'd need to do that (when to
          farm
          > and when to cluster?).
          >
          > So, the short reply, yes, the clients would be remote.
          >
          > > > 3. Modify bean descriptors to make them clusterable.
          > >
          > > They are clusterable by default. For stateful session EJBs you will
          want
          > to
          > > use in-memory replication. You didn't mention the version of WLS you
          are
          > > using pre-6.1SP3 has a few problems with this, so you might want to
          check
          > > with Customer Support if you are using SP1 or SP2.
          >
          > Sorry about that. The OS used is RHL6.2 and we are using weblogic 6.1 with
          > SP1 (thats the only one certified on 6.2). Its intriguing when you say
          that
          > all EJBs in WL 6.1 SP3 would be "clusterable by default". How does it do
          > that? Does it generate clustering descriptors for all beans deployed?
          Could
          > you shed more light on this?
          >
          > > > 4. Make web app clusterable.
          > > If using in-memory replication for HttpSession state, make sure you set
          > the
          > > session descriptor persistence type to "replicated" not "memory". This
          is
          > a
          > > point of confusion for a lot of folks.
          >
          > Thanks for the heads-up on the that. I'll keep this in mind.
          >
          > > > 5. Look for code that uses non-replicable services (file system
          access,
          > > > sockets, etc.).
          > > EJBs should not do things like this even if not running in a cluster.
          >
          > I agree EJBs should not be doing this, and chances are they will not be
          > doing this in the code either. But I'm not sure. Also, they might be
          using
          > objects that are doing this, and I'm not sure if everything will work
          > seamlessly in that case.
          >
          > > > I'm sure there must be more that I've missed. Anything more that needs
          > to
          > > be
          > > > added to the list above? Would there be a scenario where the EJB
          clients
          > > > (other beans themselves, the web tier, RMI clients, etc) would need to
          > be
          > > > changed?
          > >
          > > If your EJB clients are remote clients, you might need to modify code to
          > > handle failover (depending on the type of EJB you are using). The
          > > examples.ejb.cluster sample that ships with WebLogic is a good example
          to
          > > look at.
          >
          > Thanks again. I'll check this one out.
          >
          > Finally, the prime reason for posting this was to gain insights into such
          a
          > situation that I was hoping someone on this list had handled before.
          >
          > --
          > With Warm Regards,
          > Manav.
          >
          >
          >