Objects not being finalized and Finalizer thread n

问题:

On our server, we started to have problems with OutOfMemoryError. We analyzed the heap dumps using Eclipse Memory Analysis, and found, that many objects were held to do finalization (about 2/3 of the heap):

We found, that it could be some finalize() method blocking. I found several bug reports of this problem (here or here), and it always manifested itself in the Finalizer thread stack, that it was blocked somewhere. But in our case, this thread was WAITING:

"Finalizer" daemon prio=10 tid=0x43e1e000 nid=0x3ff in Object.wait() [0x43dfe000]
   java.lang.Thread.State: WAITING (on object monitor)
        at java.lang.Object.wait(Native Method)
        - waiting on <0x4fe053e8> (a java.lang.ref.ReferenceQueue$Lock)
        at java.lang.ref.ReferenceQueue.remove(ReferenceQueue.java:133)
        - locked <0x4fe053e8> (a java.lang.ref.ReferenceQueue$Lock)
        at java.lang.ref.ReferenceQueue.remove(ReferenceQueue.java:149)
        at java.lang.ref.Finalizer$FinalizerThread.run(Finalizer.java:189)

EDIT:

We then tried to add -XX:+UseConcMarkSweepGC, but with no success, only the frequency of OutOfMemoryErrors diminished, so we first thought it helped.

Finally, we suspected the JVM bug and upgraded from OpenJDK 1.6.0_30 to Oracle JDK 1.7.0_51, and the problem disappeared (at least it seems so, during last 4 hours the used heap does not grow). We do not remember any change in finalize method, nor did we upgrade any library, there were only minor developments during that time. The problem does not reproduce on our test server, with same configuration except that it is 64bit JVM while the production server is 32bit.

The question is: what could be the cause for Objects not being finalized and Finalizer thread waiting for next object? Did we analyze the heap dump correctly?

Thanks for all answers.

回答1:

We think it was related to the OpenJDK version 1.6.0_30. After upgrading to Oracle JDK 1.7.0_51, the problem disappeared. And it probably appeared after automatic update of openJDK, but we cannot confirm this either. We could not find a relevant bug report.

回答2:

The Finalizer thread has low-priority, and thus will spend a good amount of time WAITING rather than finalizing. I wouldn't conclude from that stack trace that the thread is blocked; it's simply relinquished control to other threads. It's instead likely you are introducing a pathological number of objects into the finalizer queue and the JVM simply can't keep up.

There are unfortunately far too many possible explanations for why this behavior changed between versions to identify an exact cause, but here's a possible explanation. Oracle's Java 7 has a a new, more efficient garbage collector. It's reasonable to imagine that lightened load on the GC means the finalizer queue gets more time on the processor, and is therefore able to keep up with the number of objects being added to it.

However regardless of the underlying cause the correct solution is to reduce your usage of finalizers. Except in very limited circumstances they introduce more problems than they solve, not the least of which is GC and memory overhead. If you ever find yourself investigating objects that are pending finalization cleanup you are constructing too many finalizable objects.