We're seeing some odd behavior that our research hasn't been able to resolve. We have a tool that uses a QueueBrowser to return all entries in a queue for display in a browser. This is primarily a testing tool. Under most situations this works fine.
However, when we use clustering and point our JNDI context to one instance in the cluster the QueueBrowser object sometimes returns 100 entries, and sometimes 300. Even if there are several thousand entries in each queue (of the same name) in the cluster the browser still returns 100 or 300. It is consistently one of these numbers and implies some type of caching limit.
As I indicated, this is not a problem on a non-clustered queue. Do you have any information on this behavior that can help us out?