Using Optimistic Concurrency in a cluster environment

Question

Hi,
In weblogic 8.1 sp3 cluster environment, I deployed CMP Entity Beans with the following settings:
In weblogic-ejb-jar.xml:
<entity-cache>
<concurrency-strategy>OptimistÂic</concurrency-strategy>
<cache-between-transactions>trÂue</cache-between-transactionsÂ>
</entity-cache>
In weblogic-cmp-rdbms-jar.xml:
<verify-columns>Version</verifÂy-columns>
<optimistic-column>VERSION</opÂtimistic-column>
And I deployed the CMP Entity beans into a cluster which has two managed servers.
When I only do findByPrimaryKey, on both managed servers, the cache-between-transaction functions well and only call ejbLoad() when it first loads that Entity Bean Instance.
However, If I do any updates to this bean, the performance change. After the updates, I issued a lot findByPrimaryKey calls for this bean to test. If the call reach the managed server where the update for the
bean happens, it is fine and still perform like cache-between-transaction. But if the call reach other managed servers, the ejbLoad() get called for that bean in every transactions; and cache-between-transaction seems to be disabled on the other managed servers since the updates.
I test this senario a lot of times and the problem is very consistant. According to my understanding, the other managed servers should only do ejbLoad() at the first time after the updates happened, and the transactions after that shouldn't call ejbLoad everytime.
Does anyone encounter the same problem like this? And is there anyway to optimize it?
Thanks!!

Answer 1

Did you figure out how to do this? We ended up having to track the number of sessions using the service and close it only when there were none. However, this did not solve the problem completely. There seems to be a conflict when running our servlet app (which uses PAPI) on different machines talking to the same BPM. A thread on one machine opens, uses, closes a session and service while a thread on another machine opens a session and in the middle of using it, it dies.