EAP 602: JVMTI error 98 on Solaris

mschulz · Post by **mschulz** » Tue May 17, 2005 12:50 am

I am getting this error on Solaris 9, Java 1.5.0, Tomcat 4.1.30:

Picked up JAVA_TOOL_OPTIONS: -agentlib:yjpagent=sessionname=Tomcat
[YourKit Java Profiler 4.5 EAP build 602] Using JVMTI
ERROR: JVMTI error err=98(JVMTI_ERROR_NOT_AVAILABLE) in JVMTI_Agent.cpp:447
Assertion failed: false, file Utils.cpp, line 26

Any ideas?

Thx.

mschulz · Post by **mschulz** » Tue May 17, 2005 3:03 am

Same story with 1.5.0_03:

Picked up JAVA_TOOL_OPTIONS: -agentlib:yjpagent=sessionname=Tomcat
[YourKit Java Profiler 4.5 EAP build 602] Using JVMTI
ERROR: JVMTI error err=98(JVMTI_ERROR_NOT_AVAILABLE) in JVMTI_Agent.cpp:447
Assertion failed: false, file Utils.cpp, line 26
java -version
^C$ java -version
java version "1.5.0_03"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_03-b07)
Java HotSpot(TM) Server VM (build 1.5.0_03-b07, mixed mode)

Post by **Anton Katilin** » Tue May 17, 2005 8:31 am

Hmm, very strange indeed.

This error happens when JVM is asked to turn on some profiling capabilities, and looks like it has some internal problems.

What is the exact output of java -version?

Are there any other VM parameters in the command line of profiled application (e.g. explicit choice of garbage collector)?

What does uname -a prints?

Post by **Anton Katilin** » Tue May 17, 2005 8:33 am

I'm sorry, I didn't notice that you have already provided the 'java -version' output...

mschulz · Post by **mschulz** » Tue May 17, 2005 3:58 pm

hmmh, yes, the GC and related options are quite elaborate:

-server
-XX:+UseConcMarkSweepGC -XX:+UseParNewGC
-XX:+CMSPermGenSweepingEnabled -XX:+CMSClassUnloadingEnabled
-XX:SurvivorRatio=3 -XX:MaxTenuringThreshold=4
-XX:MaxPermSize=64m -XX:PermSize=64m
-Xms384m -Xmx512m -XX:MaxNewSize=128m

Any suggestions what could be interfering here?

Martin

Post by **Anton Katilin** » Tue May 17, 2005 4:39 pm

Please try removing -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+CMSPermGenSweepingEnabled

JVM cannot do memory profiling with concurrent mark sweep GC, at least Java 5.0.

Post by **Anton Katilin** » Tue May 17, 2005 4:40 pm

At least everything should be OK if you leave only "-Xms384m -Xmx512m"

mschulz · Post by **mschulz** » Wed May 18, 2005 4:28 am

Thanks, CMS GC seems to conflict with JVMTI.
After removing the CMS related options, everythings wroks as expected.

I could use -XX:+UseParallelGC, though.

brettb · Post by **brettb** » Wed Nov 30, 2005 7:24 pm

This is a big problem for us - we have production systems which require CMS for performance reasons, and we'd like to have YJP installed.

Is this a limitation with JDK 1.5? Is there any workaround, and / or is Sun working on a fix?

Thu Dec 01, 2005 10:21 am

By the current JVM design UseConcMarkSweepGC blocks all profiling capabilities. Unfortunately there is no workaroung for this situation. We heard that Mustang will bring some changes, but this is not official information.

brettb · Post by **brettb** » Thu Dec 01, 2005 2:16 pm

Thanks for the info.

By "blocks all profiling capabilities" I assume you mean both JVMTI and JVMPI? I haven't got around to trying JVMPI with CMS yet in our labs.

If you have any reference to Mustang's changes, please let me know. We have some contacts at Sun to which we can request this capability.

Post by **Vladimir Kondratyev** » Thu Dec 01, 2005 3:43 pm

This bug is fixed in Mustang
http://bugs.sun.com/bugdatabase/view_bu ... id=6278106

I've just tried latest Mustang buid #61 (https://mustang.dev.java.net) and JVMTI agent works fine.