WebappClassLoader memory leak even with no gc roots

问题

HERE IS THE HEAD DUMP (UPDATED ON 10/29/2013)

I'm working in a webapp with:

Tomcat 7.0.24
Java 6
Spring 3 (with aop - cglib)
SLF4J over Log4j
Oracle Coherence

After a lot of work, I managed to remove all the strong reference to the class loader and now it is a candidate for the garbage collector. So, memory leak solved? Of course not! Because after several hot deployments a OOME appears due to PermGen space.

Thanks to Yourkit, I was able to check that the WebappClassLoader was Pending Finalization which means that is waiting in the finalizer queue (actually, is not the WebappClassLoader itself but one of his referents). Checking a memory snapshot I found several Finalizer references to Oracle Coherence classes...

This seems "okey": Coherence objects are waiting to be garbage collected thanks to all the hard work done removing all the strong references (killing all coherence threads, removing java security providers, etc). I think that there is nothing to do here.

So, I was thinking about some finalize execution that breaks something and then not allowing empty the finalizer queue. But the weird thing is that using JMX or jmap -finalizerinfo the finalizer queue seems to be empty! All this is very confusing so I kept searching in other places...

Do you think that is something to do here? I've read something about CGLIB enhancing the finalize method. If I have access to Enhancer I can create a callback filter as explained here but I dunno how to manage this with Spring AOP.

Well, searching in other places, I found several weak references from java.lang.reflect.Proxy. These are jdk dynamic proxies right? or they are related to Introspection memory leaks? with weak references?

INFO: I'm using a Spring's context listener that flushes instrospector's caches (java.beans.Introspector.flushCaches()). What else I can do with this?

Lets continue.

Then, we have several other weak references from java.io.ObjectStreamClass$Caches. A lot of my business objects have these kind of weak references.

Maybe I need flushing these caches. But how??

Then we have these weak reference related to com.sun.internal.ResourceManager, java.util.logging.Logging and java.lang.reflect.Proxy

What I can do with this weak references? Do I need to worry about this or the problem is on the finalizer queue? Any clue will be of help ... really :-D

Ah, another thing, I found a weak reference from a tomcat "main" thread that will not be renewed ever by tomcat. I know that my application can leave some thread local var in some tomcat threads, but tomcat 7 renew these threads to avoid class loader memory leaks.

I think this is the oddest thing on my memory snapshot, but is a weak referece right? What I can do with this?

EDIT: Reading the java.lang.ref javadoc i found this:

An object is weakly reachable if it is neither strongly nor softly reachable but can be reached by traversing a weak reference. When the weak references to a weakly-reachable object are cleared, the object becomes eligible for finalization.

So, can weak references retain objects in the heap when they implement a finalize method?

Meanwhile i found an answer to this, I manage to remove all the weak references to my classloader but two: ClassLoaderLogManager.classLoaderLoggers and the one related to the tomcat thread.

NOTE: Actually, I managed to remove the first one but this reference is set again by tomcat after/during undeployment.

EDIT: PLUMBR RESULTS

I've tried plumbr and no reports on the web console. Only this message on the standard output

Dumping heap to /opt/tomcat7/headdumps/java_pid9478.hprof ...
Heap dump file created [348373628 bytes in 3.984 secs]
#
# An unexpected error has been detected by Java Runtime Environment:
#
#  Internal Error (javaCalls.cpp:40), pid=9478, tid=1117813056
#  Error: guarantee(!thread->is_Compiler_thread(),"cannot make java calls from the compiler")
#
# Java VM: Java HotSpot(TM) 64-Bit Server VM (11.2-b01 mixed mode linux-amd64) [thread 1110444352 also had an error]
# An error report file with more information is saved as:
# [thread 1110444352 also had an error]
#
# If you would like to submit a bug report, please visit:
#   http://java.sun.com/webapps/bugreport/crash.jsp
#
******************************************************************************
*                                                                            *
* Plumbr has noticed that JVM has thrown an OutOfMemoryError: PermGen space. *
*                                                                            *
* You can increase PermGen size with -XX:MaxPermSize parameter.              *
* If you encountered this error after a redeploy, please read next article:  *
* http://plumbr.eu/blog/what-is-a-permgen-leak                               *
*                                                                            *
******************************************************************************

回答1:

I think Yourkit has led you to the wrong path.

I have looked into your heap dump using Eclipse Memory Analyzer. It showed, that WebappClassLoader is referenced by class com.inovasoftware.iap.data.access.platform.datarepository.CoherenceDataRepository$$EnhancerByCGLIB$$180c0a4e, which instance is alive in some thread local variable. Some googling showed this: https://hibernate.atlassian.net/browse/HHH-2481

So may be upgrading Hibernate version will be of help.

来源：https://stackoverflow.com/questions/19470328/webappclassloader-memory-leak-even-with-no-gc-roots

标签

java

Tomcat

memory-leaks

classloader

yourkit