CacheFullException not recovered after rollback

Question

I have been encountering a problematic entity-bean-related CacheFullException situation: after a CacheFullException has been thrown (for good reasons), and the transaction is rolledback, I keep getting CacheFullException in later transactions, which shouldn't get it.
It seems as though the rolled-back transaction due to CacheFullException is not cleanning-up the cache. It's not marking the entity beans that did manage to enter the cache in that failed transaction as no longer active.
I'm using WLS 8.1 SP 3. The entity bean is CMP. It's PK class is Integer. Using Container Managed Transaction.
I originally encountered the problem when using the default max-beans-in-cache = 1000 during load.
In order to better understand the problem, I now reduced max-beans-in-cache to 5 and am performing manual testing, not load.
Now, if I try to get more than 5 entity beans into the cache in a single transaction (returned from a finder and accessed), I get CacheFullException (rightfully).
Then when I try, in a new transaction, to simply create one entity bean of that type, I get CacheFullException.
I'm stuck in repeating CacheFullException, and must restart the server in order to solve it.
The repeating CacheFullException has a growing 'size' value. For
example:
weblogic.ejb20.cache.CacheFullException: size=7, target=5, incr=1, then
weblogic.ejb20.cache.CacheFullException: size=8, target=5, incr=1, then,
weblogic.ejb20.cache.CacheFullException: size=9, target=5, incr=1, etc.
In the original load scenario, it grew from close to 1000 up to 6 digits!
Using the Monitoring of the Weblogic Server Console, I see that "Cached Beans Current Count" is 0. This leaves me puzzled.
Has anyone else encountered this problem? And possibly has an idea if this is a problem in my WLS configuration / descriptors / code, or a WLS problem?
Thanks,
Eynat

Answer 1

I have been encountering a problematic
entity-bean-related CacheFullException situation: after a
CacheFullException has been thrown (for good reasons),
and the transaction is rolledback, I keep getting
CacheFullException in later transactions, which shouldn't
get it.I appear to have the same problem. At some point (whether it's after a rollback or not, I don't yet know), I get a CacheFullException when trying to create a new Entity Bean.
The CacheFullException exception shows a size that increments by one each time this error occurs. The reported size in the exception continually increases way past the max-beans-in-cache amount of 1000. Yet, the Administrator console reports the "Cached beans current count" as zero. So, the cache appears empty, yet every attempt to create a new entity bean fails with a CacheFullException!
I am using Weblogic Server 8.1 SP3, BMP, and Oracle 9i. The Entity Bean's primary key class was generated by EJBGen (primary key is a java long).
Stopping and restarting the Weblogic server fixes the problem temporarily.
Any help would be greatly appreciated.
Jim