Mustang b71/b73 + YP 5.1: 1000s of JNI Local GC roots

plethora · Post by **plethora** » Sat Feb 25, 2006 3:52 pm

Environment:
-Win32
-Mustang b73
-YP 5.1
-Tomcat 5.5.15

A typical snapshot contains 1000s or even 10000s of GC roots of type JNI Local.

Is it a JVM regression, or is YP to blame?

Post by **Anton Katilin** » Sun Feb 26, 2006 2:38 pm

Roots are shown exactly as they are reported by JVM.

So this must be a JDK issue.

Do you have similar number of such roots profiling with other JDK?

plethora · Post by **plethora** » Sun Feb 26, 2006 4:03 pm

Anton Katilin wrote: Do you have similar number of such roots profiling with other JDK?

No. I'll do an experiment with older Mustang builds.

plethora · Post by **plethora** » Tue Feb 28, 2006 7:02 pm

I tried the following Mustang build:
69, 70, 71, 73

I tried different versions of Tomcat: 5.5.9 and 5.5.15

Version of YourKit used was always 5.1

All combination exhibit the same symptoms: excessive amounts of JNI Local GC roots. In our application, the number of reported GC roots was around 15000.

I'll try to verify this again on a completely clean system.

p.s. What is the "Held by JVM" GC root type? I often see this for classloader instances.

plethora · Post by **plethora** » Tue Feb 28, 2006 7:08 pm

I'd like to tese with earlier releases, but the build history on java.net only goes back to build 69.

Using 1.5.0 releases before update 7 has other JVMTI issues. (Some reported on this forum)

plethora · Post by **plethora** » Sat Mar 04, 2006 4:48 pm

I've used 1.6 b74 and b64 for testing. (Using JVMTI)

Target application was a simple 4K Java game:
http://javaunlimited.net/games/view.php?id=102

YourKit 5.1 shows +/- 1300 JNI Local GC roots.

I'm pretty sure that 5.0.x releases did not exhibit this problem. Are these releases archived somewhere?

plethora · Post by **plethora** » Sat Mar 04, 2006 4:59 pm

Actually, this problem also occurs when using the last GA JRE: 1.5.0 update 6.

Taking a snapshot of the minimal application mentioned above gives such results.

Unless something is broken in my local JVM or YourKit installation it seems like a YK bug.

Post by **Vladimir Kondratyev** » Mon Mar 06, 2006 1:35 pm

Can you provide the small application wich demonstrate the problem?

plethora · Post by **plethora** » Mon Mar 06, 2006 8:13 pm

I'll try to prepare a testcase in the next few days.

Post by **Anton Katilin** » Tue Mar 07, 2006 9:51 am

It looks like I have managed to reproduce this.

We're investigating the problem.

Post by **Anton Katilin** » Tue Mar 07, 2006 1:48 pm

This problem is solved in 5.1.1

It is available at http://www.yourkit.com/download

plethora · Post by **plethora** » Tue Mar 07, 2006 6:51 pm

5.1.1 does indeed fix the issue I saw, thanks for the rapid response.