Clustering read-only bean-managed entity ejbs

Question

          I'm designing a data caching approach that relies on using read-only entity ejbs with bean-managed persistence. My design is based on the fact that WebLogic blocks on entity bean access by concurrent users for a given bean instance (unique primary key). I would like to keep only one entity bean instance active(timeoutSetting=0) for eacy primary key for all users to share. That way I only have to hit the database one time to initially populate data in the entity bean. I'm worried about this approach in a WebLogic clustered environment. From reading notes in this newsgroup and other doc, it appears that WebLogic might not use one instance of the entity bean (based upon unique primary key) in a clustered environment. Is that true (that being multiple users could get their own instance of the entity bean with the same primary key)?
          Thanks,
          Bryan
          

Answer 1

Typically, the read-write EJBs are on each WL server instance, so there is
          no remote invocation -- it is all done by reference.
          Cameron Purdy
          Tangosol, Inc.
          http://www.tangosol.com
          +1.617.623.5782
          WebLogic Consulting Available
          "Bryan Dixon" <[email protected]> wrote in message
          news:[email protected]...
          >
          > I guess I'm confused about read-write (not read-only) entity beans being
          pinned or not. This is from WebLogic 5.1 EJB doc:
          > "read-write entity EJBs do not use a clustered EJBObject stub; a client's
          method calls to a particular EJB always go to a single WebLogic Server
          instance. If the server that a client is using fails, the client must
          re-find the entity EJB using the cluster-aware home stub."
          >
          > Doesn't that mean the entity bean instance for a primary key is pinned to
          a single WebLogic Server instance? Maybe I'm just misunderstanding
          terminology about what a "particular EJB" is - I was thinking it is an
          entity bean instance for a unique primary key.
          >
          > Thanks,
          > Bryan
          >
          >
          > "Cameron Purdy" <[email protected]> wrote:
          > >>I was thinking that read-write entity beans were pinned.
          > >
          > >Not unless you pin them. Basically, that means that the JAR/XML that
          > >contains/specifies the EJB only is on one server.
          > >
          > >> So if two separate weblogic instances in a cluster did a find on an
          entity
          > >ejb with the same primary key and then performed some business method on
          > >that entity ejb, would there really be two separate bean instances in
          each
          > >weblogic instance for the same primary key?
          > >
          > >If it is not pinned, yes.
          > >
          > >--
          > >Cameron Purdy
          > >Tangosol, Inc.
          > >http://www.tangosol.com
          > >+1.617.623.5782
          > >WebLogic Consulting Available
          > >
          > >
          > >"Bryan Dixon" <[email protected]> wrote in message
          > >news:[email protected]...
          > >>
          > >> Toa, thanks again.
          > >>
          > >> There are a couple of things I'm not clear about though. One is that I
          > >want one instance of an entity bean per primary key, not a singleton of
          the
          > >entity bean itself. How would JNDI in a clustered environment help me
          > >there?
          > >>
          > >> The other question I have is about having multiple instances of an
          entity
          > >bean for the same primary key in the cluster. I was thinking that
          > >read-write entity beans were pinned. So if two separate weblogic
          instances
          > >in a cluster did a find on an entity ejb with the same primary key and
          then
          > >performed some business method on that entity ejb, would there really be
          two
          > >separate bean instances in each weblogic instance for the same primary
          key?
          > >>
          > >> Thanks again,
          > >> Bryan
          > >>
          > >> "Tao Zhang" <[email protected]> wrote:
          > >> >
          > >> >"Bryan Dixon" <[email protected]> wrote:
          > >> >>
          > >> >>The reason I was wanting one instance per primary key is that I want
          to
          > >use this entity bean to cache some data from database tables. This data
          > >doesn't change frequently, so we were wanting to get it from this entity
          > >bean's memory instead of constantly hitting the database. This data is
          > >global to all users, so we don't want to store it in stateful session
          beans.
          > >> >>
          > >> >>After reading more about the read-only cache-strategy it doesn't
          appear
          > >that any sycnhronization will occur if the entity bean for a given
          primary
          > >key is updated (state data is updated) in one weblogic instance, that
          change
          > >will not get synced up with other weblogic instances for that same
          primary
          > >key. Is that correct?
          > >> >>
          > >> >It's correct. If you do want to have exact one copy in the cluster.
          You
          > >can read Using JNDI in cluster environment.
          > >> >
          > >> >>If I deploy this entity bean with read-write cache-strategy the
          WebLogic
          > >doc reads as if I will get one instance per primary key that is pinned to
          > >one WebLogic instance and I won't get any fail-over or load-balancing on
          the
          > >ejbObject (the entity bean instance). Did I read this correctly? If
          that
          > >is the case, what is the advantage of setting up read-write entity beans
          to
          > >be clusterable - just the Home objects? I definitely could be
          > >misunderstanding something in the doc since I'm very new to clustering.
          > >> "Tao
          > >> >Zhang"
          > >> ><[email protected]> wrote:
          > >> >>>
          > >> >
          > >> >It's not only instance in the cluster. Probably many instances.
          > >> >
          > >> >If you use 2 tier clustering, failover will not happen because of
          > >co-location. But if you use 3 tier cluster, you can write the special
          code
          > >in the client side to do failover and load-balance.
          > >> >
          > >> >In a 2 tier cluster, actually the ejb load balancing and failover is
          > >almost useless.
          > >> >
          > >> >But in 3 tier, you can use it.
          > >> >
          > >> >Hope this help.
          > >> >
          > >> >
          > >> >
          > >> >>>"Bryan Dixon" <[email protected]> wrote:
          > >> >>>>
          > >> >>>>Thanks Tao.
          > >> >>>>
          > >> >>>>A couple more questions...
          > >> >>>>I was planning deploying this entity bean with the read-only
          > >cache-strategy which means our transaction attribute would be
          > >TXN_NOT_SUPPORTED. Also, our db isolation is TRANSACTION_READ_COMMITTED.
          > >> >>>>
          > >> >>>>Based upon how I was planning on deploying this entity bean, would
          > >WebLogic create an instance of the bean for each primary key in each
          > >cluster? I'm just trying to figure out how many duplicate bean instances
          > >for the primary key I could have across all clusters. I was really just
          > >wanting one instance that is shared among all clients and was hoping that
          > >the clustering would provide me with fail-over if that one cluster went
          > >down.
          > >> >>>>
          > >> >>>If you use 3 tier cluster structure, it's impossible to know how
          many
          > >instances of ejb with the same primary key. Probably one instance for
          each
          > >wls instance.
          > >> >>>In wls5.1, it's impossible to host only read only entity bean
          instance
          > >in the 3-tier cluster. Because read only entity bean are clusterable in
          both
          > >home and remote interface.
          > >> >>>
          > >> >>>>Regarding making the bean a pinned service, which it sounds like I
          > >might have to do to get the results I want, how do I do that? Is that a
          > >deployment descriptor setting? Also, if I make it a pinned service, do I
          > >get any fail-over suport by clustering the bean?
          > >> >>>>
          > >> >>>For the pinned service, you can just deployed on one or several
          server
          > >instances. The per-server properties file is a good place to put the
          > >weblogic.ejb.deploy property. If only one pinned service in the cluster,
          you
          > >can't get fail over. If the that server instance fails, the home stub
          will
          > >be removed from the jndi in other server instances.
          > >> >>>
          > >> >>>Why do you must need only one instance in the cluster? Do you want
          > >exact-only-copy? You can read Using JNDI doc about its in cluster
          > >environment.
          > >> >>>
          > >> >>>
          > >> >>>
          > >> >>>
          > >> >>>>Thanks again,
          > >> >>>>Bryan
          > >> >>>>
          > >> >>>>
          > >> >>>>
          > >> >>>>"Tao Zhang" <[email protected]> wrote:
          > >> >>>>>
          > >> >>>>>
          > >> >>>>>Bryan Dixon <[email protected]> wrote in message
          > >> >>>>>news:[email protected]...
          > >> >>>>>>
          > >> >>>>>> I'm designing a data caching approach that relies on using
          > >read-only
          > >> >>>>>entity ejbs with bean-managed persistence. My design is based on
          the
          > >fact
          > >> >>>>>that WebLogic blocks on entity bean access by concurrent users for
          a
          > >given
          > >> >>>>>bean instance (unique primary key). I would like to keep only one
          > >entity
          > >> >>>>>bean instance active(timeoutSetting=0) for eacy primary key for
          all
          > >users to
          > >> >>>>>share. That way I only have to hit the database one time to
          > >initially
          > >> >>>>>populate data in the entity bean. I'm worried about this approach
          in
          > >a
          > >> >>>>>WebLogic clustered environment. From reading notes in this
          newsgroup
          > >and
          > >> >>>>>other doc, it appears that WebLogic might not use one instance of
          the
          > >entity
          > >> >>>>>bean (based upon unique primary key) in a clustered environment.
          Is
          > >that
          > >> >>>>>true (that being multiple users could get their own instance of
          the
          > >entity
          > >> >>>>>bean with the same primary key)?
          > >> >>>>>>
          > >> >>>>>
          > >> >>>>>
          > >> >>>>>It's true. In a cluster environment, each wls instance can have
          their
          > >ejb
          > >> >>>>>instance. The block of concurrent access to the ejb data is up to
          > >your
          > >> >>>>>transaction attribute and isolation level and your database.
          > >> >>>>>
          > >> >>>>>If you only want to keep one instance active, you can make the
          > >read-only
          > >> >>>>>entity bean a pinned service, to be deployed in one instance. But
          the
          > >> >>>>>network overhead is worse.
          > >> >>>>>
          > >> >>>>>
          > >> >>>>>> Thanks,
          > >> >>>>>> Bryan
          > >> >>>>>
          > >> >>>>>
          > >> >>>>
          > >> >>>
          > >> >>
          > >> >
          > >>
          > >
          > >
          >