Probable JDK6 regression: SSLSocketImpl leak

[Nathan Bryant]

Hi,

Looking for some advice on how to debug an issue that's hitting our
production stack.

We have a grails app that's using apache httpclient 4.3.3 for transport to
our backend tier over SSL. This worked fine for a long time but started to
leak after our latest push.

Symptoms: heap dump shows a large number of SSLSocketImpl instances (and
associated byte[] and SocksSocketImpl) that can not be reclaimed. Clicking
through everything in jmap, I can not find a path to the rootset (the paths
to rootset query times out). There are references from finalizers, but I am
consistently seeing "Number of objects pending for finalization: 0" via
jhat and via jmap -finalizerinfo and other heap analysis tools. Thus, these
are objects that "should" be reclaimed by GC.

The finalizer thread itself appears to be idle:

"Finalizer" daemon prio=10 tid=0x0000000001dcc000 nid=0x2437 in
> Object.wait() [0x00007f3b8f5f4000]
>    java.lang.Thread.State: WAITING (on object monitor)
> at java.lang.Object.wait(Native Method)
> - waiting on <0x0000000180014918> (a java.lang.ref.ReferenceQueue$Lock)
> at java.lang.ref.ReferenceQueue.remove(ReferenceQueue.java:133)
> - locked <0x0000000180014918> (a java.lang.ref.ReferenceQueue$Lock)
> at java.lang.ref.ReferenceQueue.remove(ReferenceQueue.java:149)
> at java.lang.ref.Finalizer$FinalizerThread.run(Finalizer.java:189)


The app works on:

$ java -version
> java version "1.6.0_27"
> OpenJDK Runtime Environment (IcedTea6 1.12.6)
> (6b27-1.12.6-1ubuntu0.10.04.4)
> OpenJDK Client VM (build 20.0-b12, mixed mode, sharing)
>


But fails on:


$ java -version
> java version "1.6.0_30"
> OpenJDK Runtime Environment (IcedTea6 1.13.1) (6b30-1.13.1-1ubuntu2~0.12.04.1)
> OpenJDK 64-Bit Server VM (build 23.25-b01, mixed mode)


It also fails on the client/32-bit version of 1.6.0_30.

If these objects are pinned by references from native code, I am
assuming this doesn't show up in the heap dump?