OutOfMemoryError and Unloading class sun.reflect.GeneratedMethodAccessor

Question

I have application with settings: -Xmx2048M, -Xms2048M, -XX:MaxPermSize=256M.
Sometimes I get a lot of messages in log:

[Unloading class sun.reflect.GeneratedMethodAccessor9]  
[Unloading class sun.reflect.GeneratedMethodAccessor129]  
[Unloading class sun.reflect.GeneratedMethodAccessor12]  
[Unloading class sun.reflect.GeneratedMethodAccessor11]  
[Unloading class sun.reflect.GeneratedMethodAccessor12]  
[Unloading class sun.reflect.GeneratedMethodAccessor11]  
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor29]  
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor25]  

and get error:

OutOfMemoryError: Java heap space

After reading this article:
http://anshuiitk.blogspot.com/2010/11/excessive-full-garbage-collection.html
I know, that classes are load in Perm Gen and should occur error:

OutOfMemoryError: PermGen space.

My question, why I have error OutOfMemoryError: Java heap space instead of
OutOfMemoryError: PermGen space?

Solution 1

This means you are running out of heap space (or you have a space which cannot grow for some reason)

You only get PermGen space. as a reason when the maximum for this space runs out.

In short, you are getting this error because your heap space cannot grow to hold the objects in it.

Solution 2

Like @Peter tells it's a heap issue. Unloading is a side effect of the full garbage collection, not an indication of anything going amiss.

Here is how to resolve and prevent OOMs:

• Allow heap dumps on OOM -XX:+HeapDumpOnOutOfMemoryError on hotspot, the option has no performance penalties
• Study the heap dump and understand where you have a leak or overcommitted use of heap, e.g. huge selects from a database.
• Once you get proficient at heap analysis, take heap dumps via jmap during development and analyze for leaks.
• If you believe you may have leaks in production: jmap -histo is a good option as it enjoys little performance hit only.

Comments

[luk4443]

Thanks for response. But this messages [Unloading class sun.reflect.GeneratedMethodAccessor9] indicates problem in Perm Gen and why I don't get OutOfMemoryError: PermGen space, or maybe first OutOfMemoryError: PermGen and then next OutOfMemoryError: Java heap space?

[Vishy]

The message indicates the PermGen was cleaned up. No more than that. You will always get a Full GC before an OOME Java heap space and that will clean up all areas including the perm gen (even if there is plenty of space there)

[luk4443]

So far I thought, that this OutOfMemoryError is trigger by soft references when GC cannot properly clean it. But it seems, there are another reason of OOME?

[Vishy]

Soft references can change your GC behaviour, but won't trigger an OMEE unless you have many millions of them.

[luk4443]

Thanks. I have several hundred messages like Unloading class sun.reflect.GeneratedMethodAccessor. Heap spase rises to about 1,8 G and trigger an OOEM. So there are very intensive object creating or just memory leaks?

[Vishy]

If you are using lots of reflection in your libraries you can get lots of these messages. I have been working on 64-bit JVMs for a few years and I start with a heap of 8 GB and increase or decrease it as required. 1.8 GB doesn't sound like too much unless you have a 32-bit machine.

[luk4443]

A lot of time I spend to find cause of trigger an OOEM Java heap space. It is triggered when heap space grow to about 1,8 G (-Xmx2048M). But twice there are another situation - heap grow to maximum (several MB free), but there aren't OOEM Java heap space. But application server restarted. Maybe could you aim me to reason of this problems?

[Vishy]

What do you mean application server restarted ?

[luk4443]

After about hour of application running with 95% heap space use, application server itself restarted.But when occurs an OOEM, situation last about 15 min. before an OOEM: Java heap space (but server doesn't restart itself).

[Cameron]

It's no secret he has run out of heap space, that's in the error message. But the question is how are these messages related to garbage collection and memory issues? I see these messages often in GC logs before long GC pauses, not out of memory errors. But it's not clear to me what these messages are indicative of and how I can reduce their negative consequences.

[Vishy]

The best thing to do is to use a memory profiler to minimize object creation and usage. Once you have done this you can look at how the GC logs compare with the graphs you see in VisualVM.