RFR: JDK-8261034: improve jcmd GC.class_histogram to support parallel

[Hamlin Li]

On Wed, 17 Feb 2021 04:30:45 GMT, Chris Plummer <cjplummer at openjdk.org> wrote:

>> Dear All, 
>>     May I say that we all agreed that "noparallel" is not necessary at present? I think the PR #2519 and related CSR and issue could be all closed.
>
>>  May I say that we all agreed that "noparallel" is not necessary at present? I think the PR #2519 and related CSR and issue could be all closed.
> 
> Yes, assuming we get approval for `dumpheapext`. Unfortunately the email that went out did not get any feedback. Maybe we should just create a CSR for that and see what the CSR committee says.

Hi @plummercj @sspitsyn @linzang 
Thanks for discussion. I understood that currently the final decision is to still use "parallel=<N>" and do not introduce new option "noparallel".
Based on this decision, for jcmd GC.class_historgram, would you mind to double confirm whether currently implementation is fine? I think we only need to refine the words in help doc or CSR, am I right? please kindly correct me if I misunderstand it. :-)
If I'm understanding correctly, would you mind to suggest how should we describe <N>? IMHO, I don't think it's a good idea to expose to much details about the implementation, and it's hard to make its meaning precise. Following is my version:

"Degree of parallelism for heap iteration. "
"0 means let the VM determine the parallelism. "
"1 means use one thread, i.e. disable parallelism. "
"n means ask the VM try best to use n threads, but it's not guaranteed and dependant on the specific implementation and runtime environment. n must be positive."

How do you think about it?

-------------

PR: https://git.openjdk.java.net/jdk/pull/2379