hsx/hotspot-gc/hotspot: 8009536: G1: Apache Lucene hang during reference processing

[John Cuthbertson]

Hi Bengt,

I'm not sure. I don't think we want to make the GC test suite dependent 
on ant and ivy (currently you only need shell and a jdk to run them). At 
the moment I'm running the command Uwe gave by hand .

But it would be great to be able to run them consistently.

Uwe:  do you think there would be any licensing issues?

JohnC

On 4/3/2013 11:30 PM, Bengt Rutisson wrote:
>
> Uwe,
>
> Thanks for verifying the fix! You really provided an excellent bug 
> report.
>
> John,
>
> Great that you got this fixed so quickly! I like that you added the 
> Lucene test to your testing portfolio. Is there any way we can add it 
> to the common testing? I'm guessing that we don't have a lot of tests 
> that stress the marking overflow handling since we didn't catch this 
> in our internal testing. Are there any license issues with including 
> the Lucene test in our nightly testing?
>
> Thanks,
> Bengt
>
> On 4/2/13 10:48 PM, John Cuthbertson wrote:
>> Hi Uwe,
>>
>> Thanks for the update. That is good to know. The bulk of the credit 
>> goes to Bengt who did the initial diagnosis and the prototype fix.
>>
>> Thanks also for walking me through how to reproduce the issue. Once 
>> the magic formula for the proxy was found - your instructions were 
>> spot on. The tests were awesome as marking stress tests and uncovered 
>> several issues associated with concurrent marking overflows. I've 
>> added them to the set of tests I run for marking changes. :)
>>
>> Cheers,
>>
>> JohnC
>>
>> On 4/2/2013 12:16 PM, Uwe Schindler wrote:
>>> Hi John,
>>>
>>> today I installed JDK8 b83 on one of the Jenkins build servers used 
>>> by Apache Lucene: The problem with 32 bit JVMs and G1 collector is 
>>> solved. I compared the trunk checkout of Lucene with both JVMs: In 
>>> b82 it hangs, in b83 it passes. So I was able to re-enable 32 bit 
>>> builds without the extra workaround stacksize option again.
>>>
>>> Thanks for fixing this issue.
>>>
>>> Uwe
>>>
>>> -----
>>> Uwe Schindler
>>> uschindler at apache.org
>>> Apache Lucene PMC Member / Committer
>>> Bremen, Germany
>>> http://lucene.apache.org/
>>>
>>>> -----Original Message-----
>>>> From: hotspot-gc-dev-bounces at openjdk.java.net [mailto:hotspot-gc-dev-
>>>> bounces at openjdk.java.net] On Behalf Of john.cuthbertson at oracle.com
>>>> Sent: Tuesday, March 19, 2013 2:21 AM
>>>> To: hotspot-gc-dev at openjdk.java.net
>>>> Subject: hg: hsx/hotspot-gc/hotspot: 8009536: G1: Apache Lucene hang
>>>> during reference processing
>>>>
>>>> Changeset: fa08949fe0cb
>>>> Author:    johnc
>>>> Date:      2013-03-18 11:05 -0700
>>>> URL:       http://hg.openjdk.java.net/hsx/hotspot-
>>>> gc/hotspot/rev/fa08949fe0cb
>>>>
>>>> 8009536: G1: Apache Lucene hang during reference processing
>>>> Summary: In CMTask::do_marking_step(), Skip offering termination and
>>>> entering the first and second synchronization barriers if called 
>>>> from a serial
>>>> context, i.e. the VM thread.
>>>> Reviewed-by: brutisso, tschatzl
>>>>
>>>> ! src/share/vm/gc_implementation/g1/concurrentMark.cpp
>>>> ! src/share/vm/gc_implementation/g1/concurrentMark.hpp
>>
>