8202466: Test serviceability/tmtools/jstat/GcTest01.java fails: Number of concurrent GC events is 1, but CGCT is 0

[Chris Plummer]

Hi Yasumasa,

On 5/2/18 5:46 AM, Yasumasa Suenaga wrote:
> Hi all,
>
> I've tried to fix 8202466, but I do not yet certain which solution is 
> the best for it.
> So I want your opinion for it.
>
> https://bugs.openjdk.java.net/browse/JDK-8202466
>
> David reports serviceability/tmtools/jstat/GcTest01.java failed in 
> Mach5. It seems to appear on OS X only.
> He has shared jtreg report, and I found as below:
>
> ```
> stdout: S0C S1C S0U S1U EC EU OC OU MC MU CCSC CCSU YGC YGCT FGC FGCT 
> CGC CGCT GCT
>  0.0 0.0 0.0 0.0 3072.0 0.0 11264.0 2825.6 9472.0 9003.3 1024.0 845.3 
> 26 0.204 3 0.292 1 0.000 0.497
> ```
>
> YGCT: 0.204
> FGCT: 0.292
> CGCT: 0.000
>
>  GCT: 0.497
>
> I guess this failure was caused by rounding error because (YGCT + FGCT 
> + CGCT) < GCT.
> CGC is 1, so CGC operation might finish in very short time.
>
>
> GcTest01.java:
> http://hg.openjdk.java.net/jdk/jdk/file/4da7dce7e2bf/test/hotspot/jtreg/serviceability/tmtools/jstat/GcTest01.java#l57
>
> GcCauseTest01.java:
> http://hg.openjdk.java.net/jdk/jdk/file/4da7dce7e2bf/test/hotspot/jtreg/serviceability/tmtools/jstat/GcCauseTest01.java#l53
>
> GcTest01 and GcCauseTest01 are very similar, but GcCauseTest01 passed.
> Both tests use GcProvoker::provokeGc() to inflate memory usage. So I 
> wonder why GcTest01 just only failed.
> I guess we might encounter similar issue(s) in the future if we get 
> more fast machines.
>
> Hence I think we can take two approaches as below:
>
>   1. Add all tests in serviceability/tmtools/jstat to ProblemList.
I don't think we want to do that at this time since this is a newly 
introduced problem, and is intermittent. So the goal would be to fix it 
in the near term without problem listing it.
>   2. Change all JstatGc*Tool to use custom jstat_options - to show raw 
> values in PerfCounters
I'm not sure what is meant by this.

thanks,

Chris
>
>
> What do you think?
> I can start to work for it if they are OK.
>
>
> Thanks,
>
> Yasumasa
>