Hello!

Using hibernate 2.1 beta 6.

I profiled my app because I got OutOfMemoryError after some redeployments within JBoss app server (3.2.2).

Looking with the profiler I see some strange things happening.

For example: My app contains a persistent class called com.select.adsystem.model.Element. After the app is deployed, Hibernate configuration is processed and some persistence operations are performed, memory profiler reports several classes that only have one instance allocated and all look like that:

com.select.adsystem.model.Element$$ProxiedByCGLIB$$82
com.select.adsystem.model.Element$$ProxiedByCGLIB$$81
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8006
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8005
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8004
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8003
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8002
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8001
com.select.adsystem.model.Element$$ProxiedByCGLIB$$8000
com.select.adsystem.model.Element$$ProxiedByCGLIB$$80
com.select.adsystem.model.Element$$ProxiedByCGLIB$$7999
...
...

... the strange thing being that there are more that 4000 of them. Is this realy what it is supposed to be happening? Are they all different? If they are all the same, there should be some cache preventing from generating a new class just to instantiate only one object of it. If they are different it seems that this is a strange design. Or maybe this is just a bug.


Regards, Peter

There is a lazy proxy for each field of the class.

Ok, but I don't have 4000+ fields in my Element class.

There is something broken.
Cglib fails to find generated class in cache. It can be caused by some bug in LazyInitializer implementation (All parameters to "enhance" must implement "equals/hashCode" for cache).

Sorry, Juozas - could you explain a bit further?

enhance() takes some classes, a class loader and a MethodInterceptor.

Class + ClassLoader already implement equals()/hasCode(), presumably (well, unless the ClassLoader doesn't, I suppose).

I don't have to implement equals()/hashCode() on the MethodInterceptor do I? That makes no sense.....

No MethodInterceptor doe's not need to implemet it, Filter implementation must implement it (if not singleton).
It can be something broken in cglib too, have you changed cglib version in hibernate 2.1 beta 6 ?

Whats a Filter?



No, not for five months.

Filter can be used to strip not interesting methods (default is to accept all methods).

It will be something more interesting , if this code was not changed (possible classloader implementation), cache is tested in cglib.

"Hot fix" can be cache in LazyInitilizer implemention, it must be very trivial to implement for hibernate use case (sole "class" key for synchronized WeakHashMap must work).

AFAIK Hibernate does not use a ClassFilter so I'm not sure what's going on here. Hopefully the switch CGLIB2 will make it go away (only uses 2 classes per type, instead of 1+N) but it does seem like there is an underlying ClassLoader bug or something.