作者:Jevgeni Kabanov日期:2009.12.10
链接:http://www.zeroturnaround.com/blog/rjc201/
For the full article series on Reloading Java Classes, see:
- RJC101: Objects, Classes and ClassLoaders
- RJC201: How do ClassLoader leaks happen?
- RJC301: Classloaders in Web Development --- Tomcat, GlassFish, OSGi, Tapestry 5 and so on
- RJC401: JRebel --- what does it do and how does it do that?
- RJC501: The social impact of Turnaround
From ClassLoaders to Classes
If you have programmed in Java for some time you know that memory leaks do happen. Usually it’s the case of a collection somewhere with references to objects (e.g. listeners) that should have been cleared, but never were. Classloaders are a very special case of this, and unfortunately, with the current state of the Java platform, these leaks are both inevitable and costly: routinely causing OutOfMemoryError
’s in production applications after just a few redeploys.
Let’s get started. Recalling RJC101: to reload a class we threw away the old classloader and created a new one, copying the object graph as best we could:
Every object had a reference to its class, which in turn had a reference to its classloader. However we didn’t mention that every classloader in turn has a reference to each of the classes it has loaded, each of which holds static fields defined in the class:
This means that
- If a classloader is leaked it will hold on to all its classes and all their static fields. Static fields commonly hold caches, singleton objects, and various configuration and application states. Even if your application doesn’t have any large static caches, it doesn’t mean that the framework you use doesn’t hold them for you (e.g. Log4J is a common culprit as it’s often put in the server classpath). This explains why leaking a classloader can be so expensive.
- To leak a classloader it’s enough to leave a reference to any object, created from a class, loaded by that classloader. Even if that object seems completely harmless (e.g. doesn’t have a single field), it will still hold on to its classloader and all the application state. A single place in the application that survives the redeploy and doesn’t do a proper cleanup is enough to sprout the leak. In a typical application there will be several such places, some of them almost impossible to fix due to the way third-party libraries are built. Leaking a classloader is therefore, quite common.
To examine this from a different perspective let’s return to the code example from our previous article. Breeze through it to quickly catch up.
Introducing the Leak
We will use the exact same Main class as before to show what a simple leak could look like:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 |
|
The ExampleFactory
class is also exactly the same, but here’s where things get leaky. Let’s introduce a new class called Leak
and a corresponding interface ILeak
:
1 2 3 4 5 6 7 8 9 10 |
|
As you can see it’s not a terribly complicated class: it just forms a chain of objects, with each doing nothing more than holding a reference to the previous one. We will modify the Example
class to include a reference to the Leak
object and throw in a large array to take up memory (it represents a large cache). Let’s omit some methods shown in the previous article for brevity:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 |
|
The important things to note about Example
class are:
Example
holds a reference toLeak
, butLeak
has no references toExample
.- When
Example
is copied (methodcopy()
is called) a newLeak
object is created holding a reference to the previous one.
If you try to run this code an OutOfMemoryError will be thrown after just a few iterations:
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space at example.Example.<clinit>(Example.java:8)
With the right tools, we can look deeper and see how this happens.
Post Mortem
Since Java 5.0, we’ve been able to use the jmap
command line tool included in the JDK distribution to dump the heap of a running application (or for that matter even extract the Java heap from a core dump). However, since our application is crashing we will need a feature that was introduced in Java 6.0: dumping the heap on OutOfMemoryError. To do that we only need to add -XX:+HeapDumpOnOutOfMemoryError
to the JVM command line:
java.lang.OutOfMemoryError: Java heap space Dumping heap to java_pid37266.hprof ... Heap dump file created /[57715044 bytes in 1.707 secs/] Exception in thread "main" java.lang.OutOfMemoryError: Java heap space at example.Example.<clinit>(Example.java:8)
After we have the heap dump we can analyze it. There are a number of tools (including jhat, a small web-based analyzer included with the JDK), but here we will use the more sophisticated Eclipse Memory Analyzer (EMA).
After loading the heap dump into the EMA we can look at the Dominator Tree analysis. It is a very useful analysis that will usually reliably identify the biggest memory consumers in the heap and what objects hold a reference to them. In our case it seems quite obvious that the Leak
class is the one that consumes most of the heap:
Now let’s run a search for all of the Leak objects and see what are they holding to. To do that we run a search List objects /-> with outgoing references for “example.Leak”:
The results include several Leak
objects. Expanding the outgoing references we can see that each of them holds on to a separate instance of Example
through a bunch of intermediate objects:
You may notice that one of the intermediate objects is ExampleFactory$1
, which refers to the anonymous subclass of URLClassLoader
we created in the ExampleFactory
class. In fact what is happening is exactly the situation we described in the beginning of the article:
- Each
Leak
object is leaking. They are holding on to their classloaders - The classloaders are holding onto the
Example
class they have loaded:
Conclusions
Though this example is slightly contrived, the main idea to take away is that it’s easy to leak a single object in Java. Each leak has the potential to leak the whole classloader if the application is redeployed or otherwise a new classloader is created. Since preventing such leaks is very challenging, it’s a better idea to use Eclipse Memory Analyzer and your understanding of classloaders to hunt them down after you get an OutOfMemoryError
on redeploy.
This article addressed the following questions:
- How does reloading a class cause the classloader to leak?
- What are some consequences of leaking classloaders?
- What tools can be used to troubleshoot these memory leaks?
Check out the next articles in the series:
- RJC101: Objects, Classes and ClassLoaders
- RJC201: How do ClassLoader leaks happen?
- RJC301: Classloaders in Web Development --- Tomcat, GlassFish, OSGi, Tapestry 5 and so on
- RJC401: JRebel --- what does it do and how does it do that?
- RJC501: The social impact of Turnaround