FW: is this a bug ?

[Aleksey Shipilev]

On 10/16/19 3:23 PM, Yangfei (Felix) wrote:
> Based on that, I think "org.openjdk.jcstress.tests.atomicity.buffers.ByteBufferViewsAtomicityTests.LongViewTest" has the same problem.  
> 
> I can also reproduce this problem on X86-64 platform with 8u jdk by running: 
> $ java -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI -XX:-RestrictContended -XX:ReservedCodeCacheSize=128M -jar /home/yangfei/tools/jcstress.jar -t "org.openjdk.jcstress.tests.atomicity.buffers.ByteBufferViewsAtomicityTests.LongViewTest" -time 50000

"-XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI -XX:-RestrictContended
-XX:ReservedCodeCacheSize=128M" does not make sense in this context: jcstress would fork the JVMs
and set up their own command line options, enabling WhiteBoxAPI and @Contended along the way.

> Looks like both tests always fail with the following exception: 
> Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
>         at org.openjdk.jcstress.util.OpenAddressHashCounter.resize(OpenAddressHashCounter.java:130)
>         at org.openjdk.jcstress.util.OpenAddressHashCounter.recordWithTries(OpenAddressHashCounter.java:108)
>         at org.openjdk.jcstress.util.OpenAddressHashCounter.record(OpenAddressHashCounter.java:77)
>         at org.openjdk.jcstress.util.OpenAddressHashCounter.merge(OpenAddressHashCounter.java:83)
>         at org.openjdk.jcstress.tests.atomicity.buffers.ByteBufferViewsAtomicityTests_LongViewTest_jcstress.internalRun(ByteBufferViewsAtomicityTests_LongViewTest_jcstress.java:78)
>         at org.openjdk.jcstress.infra.runners.Runner.run(Runner.java:101)
>         at org.openjdk.jcstress.JCStress.runEmbedded(JCStress.java:246)
>         at org.openjdk.jcstress.ForkedMain.main(ForkedMain.java:55)

It looks like heap overflow, which is not surprising, given the ByteBuffer tests and the very old
jcstress, which misses at least this fix:
  https://hg.openjdk.java.net/code-tools/jcstress/rev/0eecf3f89210

> Note that I am using an old jcstress version in order to run it with a 8u jdk.  I execute " hg update -r 0de8079ec104 " to get the old version.  

Why? Current jcstress requires JDK 11 to build, but it can run with JDK 8 (it would yield some "soft
errors" on tests that are not executable with JDK 8). 0de8079ec104 is very, very, very old and
misses many important harness and test fixes.

I took the today's build of jcstress:
  https://builds.shipilev.net/jcstress/jcstress-tests-all-20191016.jar

...and today's 8u232:
  https://builds.shipilev.net/openjdk-jdk8/openjdk-jdk8-latest-linux-x86_64-release.tar.xz

...and it runs to completion:

$ jdk8u232/bin/java -jar jcstress.jar -t ByteBufferViewsAtomicityTests.LongViewTest -time 50000

-- 
Thanks,
-Aleksey