Test client pinned to single node in production

Question

WL 6.1 sp2, Solaris 2.8
          Currently we have a bunch of SLSBs deployed in cluster out in production and
          a web tier that usually gets and invokes a single SLSB, and they're running
          happily. But everyone once in a while, we get an asymmetrical exception,
          where one node in the cluster is giving us bad results. What I like to do
          is write some simple test clients that can pin to a particular node and
          diagnose just that node while the regular client (web tier) still
          round-robins in production.
          Our SLSBs do not have <stateless-clustering> elemements at all in
          weblogic-ejb-jar.xml, so <stateless-bean-is-clusterable> defaults to true.
          My understanding is this means WL will round robin at 3 different levels:
          jndi Context, EJBHome and EJBObject, unless server-client is co-located in
          which WL will always pick the local object.
          What I have tried to do is to write the test client with a single url in
          PROVIDER_URL
          and PIN_TO_PRIMARY_SERVER set to true in InitialContext construction. This
          does not seem to work; by the time I get the EJBHome, create the EJBObject
          and invoke a test method, I see round-robin occuring. I can understand a
          reason FOR this behavior, and a counter-argument AGAINST this behavior. The
          reason why WL is still round-robining is because only the Context is pinned
          to the primary server; subsequent EJBHome and EJBObject are cluster-aware,
          and hence will round-robin, which in fact it is doing. But then the reason
          against this observation is that once I retrieve InitialContext, the
          subsequent EJBHome and EJBObject are all available locally. So shouldn't WL
          do co-location optimization and hence never round-robin???
          Here are some alternative framework I've thought up so I can write a test
          client that pins to a specific ejb server:
          1) Create a second set of DD's in weblogic-ejb-jar.xml, this time setting
          stateless-bean-is-clusterable to false, and have the test client us this for
          pinning.
          2) Expose a co-located servlet that will accept ejb invocation (via SOAP or
          customized RPC). Servlet invocation will always be ip-specific, and
          hopefully co-location of web and ejb tier will
          Problem with #1 is 2 sets of DD's, hence 2 sets of EJBHomes/Objects that
          behave slightly differently.
          Problem with #2 is the complexity of a new web tier just for pinning, which
          then also means the test client doesn't exactly replicate my actual web
          client calls.
          Is there a simple solution to isolate and diagnose a single node in a
          production cluster? Am I missing something? Much appreciated!
          Gene
          

Answer 1

WL 6.1 sp2, Solaris 2.8
          Currently we have a bunch of SLSBs deployed in cluster out in production and
          a web tier that usually gets and invokes a single SLSB, and they're running
          happily. But everyone once in a while, we get an asymmetrical exception,
          where one node in the cluster is giving us bad results. What I like to do
          is write some simple test clients that can pin to a particular node and
          diagnose just that node while the regular client (web tier) still
          round-robins in production.
          Our SLSBs do not have <stateless-clustering> elemements at all in
          weblogic-ejb-jar.xml, so <stateless-bean-is-clusterable> defaults to true.
          My understanding is this means WL will round robin at 3 different levels:
          jndi Context, EJBHome and EJBObject, unless server-client is co-located in
          which WL will always pick the local object.
          What I have tried to do is to write the test client with a single url in
          PROVIDER_URL
          and PIN_TO_PRIMARY_SERVER set to true in InitialContext construction. This
          does not seem to work; by the time I get the EJBHome, create the EJBObject
          and invoke a test method, I see round-robin occuring. I can understand a
          reason FOR this behavior, and a counter-argument AGAINST this behavior. The
          reason why WL is still round-robining is because only the Context is pinned
          to the primary server; subsequent EJBHome and EJBObject are cluster-aware,
          and hence will round-robin, which in fact it is doing. But then the reason
          against this observation is that once I retrieve InitialContext, the
          subsequent EJBHome and EJBObject are all available locally. So shouldn't WL
          do co-location optimization and hence never round-robin???
          Here are some alternative framework I've thought up so I can write a test
          client that pins to a specific ejb server:
          1) Create a second set of DD's in weblogic-ejb-jar.xml, this time setting
          stateless-bean-is-clusterable to false, and have the test client us this for
          pinning.
          2) Expose a co-located servlet that will accept ejb invocation (via SOAP or
          customized RPC). Servlet invocation will always be ip-specific, and
          hopefully co-location of web and ejb tier will
          Problem with #1 is 2 sets of DD's, hence 2 sets of EJBHomes/Objects that
          behave slightly differently.
          Problem with #2 is the complexity of a new web tier just for pinning, which
          then also means the test client doesn't exactly replicate my actual web
          client calls.
          Is there a simple solution to isolate and diagnose a single node in a
          production cluster? Am I missing something? Much appreciated!
          Gene