I fetch an instance of class A with id of 1 from a hibernate session then close the session. I store the fetched instance of class A to a variable named aFromFirstSession.

I open a new session and retrieve an instance of class B whose a1 references the same record (id of 1) that was fetched in my first hibernate session and a2 is null and close the session.

I then call setA2 passing aFromFirstSession, open a hibernate session and call saveOrUpdate on b.

This give a NonUniqueObjectException. This makes sense, because b is referencing two seperate instances that represent the same record (id of 1) since they were retrieved in seperate hibernate sessions.

I thought I could solve this by turning on the second-level cache for class A. With this approach, the instance will be the same across the multiple hibernate sessions because it is being retrieved from the second-level cache.

I still get the NonUniqueObjectException on the saveOrUpdate even though the log says that it was retrieved from the second-level cache.

Am I correct to think that if the instance is retrieved from the second-level cache I shouldn't get the NonUniqueObjectException in the above scenario?

You are correct in that the instance is recovered from the second level cache because it uses the id returned from the query. But your second assumption that it is the same object is incorrect. To stop you from corrupting the cached version or from having multiple threads update tha same object, it gives you different copies of the same object. If you have not overridden the equals and hashCode methods, it will think these are different objects and it will try to save the second version. The point of the second-level cache is not to minimise object creation, but unnecessary(duplicated) queries. Hope this answers your question.