[Preliminary Review]: Proposal for back-porting JFR to OpenJDK8u

Andrey Petushkov andrey at azul.com
Thu Feb 28 10:10:03 UTC 2019


Sorry, no matter how hard I try I cannot repeat this problem :(
If I configure with --disable-jfr there are no JFR-related VM flags, diagnostic commands or beans (checked with command line and JMC). 
So I see no means to initiate JFR dump.
Could you please elaborate what target platform you're building for (I've checked on x86_64) and your exact configure parameters

Andrey

> On 27 Feb 2019, at 20:12, Mario Torre <neugens at redhat.com> wrote:
> 
> On Wed, Feb 27, 2019 at 5:45 PM Andrey Petushkov <andrey at azul.com> wrote:
>> 
>> Hi Mario,
>> 
>> First of all great thank you for the interest and such immediate involvement into this effort.
>> I'm sure it will help to move it faster towards having done and adopted into jdk8u mainline :)
>> 
>> To your particular question, what are the evidence that jfr files get compiled? It's a bit tricky thing hence the question.
>> There are in fact two processes which might or might not happen:
>> - jfr*.cpp files get compiled and linked into libjvm.so. Also JFR classes get compiled and put into jfr.jar.
> 
> Well, the thing is that I tried passing  --disable-jfr and I can dump
> a perfercly functioning JFR file.
> 
> I think the option to not build JFR is not working properly, I can see
> in the config.log that is being registered correctly:
> 
> configure:19778: checking whether to build jfr
> configure:19794: result: false
> 
> However then the same build can produce a JFR file.
> 
> Cheers,
> Mario
> 
> 
>> This should happen only when --enable-jfr is turned on (and yes, it's on by default)
>> - jfr metadata is processed and respective java and c++ classes are generated (into hotspot/linux_amd64_compiler2/generated/tools/jfr,
>> hotspot/linux_amd64_compiler2/generated/jfrfiles and jdk/lib/jfr under the build directory). This in fact happens regardless of whether
>> JFR is enabled or not
>> 
>> So if you only see the latter but don't see any jfr*.o or jfr.jar files that is normal (and that's what I observe with the code I've used to generate webrev from).
>> Otherwise something is wrong. Please let me know what is the case for you
>> 
>> Regarding why it works that way, it's not something I've invented. It's exactly the way JFR gets built in jdk11 codebase. If it's unacceptable I'll work to change it,
>> but please be aware that in order to avoid generation of jfr*.hpp files one would need to put a lot of extra #ifdefs into non-jfr specific files. Which I tried to avoid
>> to keep the code as close to the original as possible (and seeing no value in doing such change, in fact).
>> 
>> Re default value for --enable-jfr, thank you, I've noted that. Will fix
>> 
>> Thanks,
>> Andrey
>> 
>>> On 27 Feb 2019, at 17:00, Mario Torre <neugens at redhat.com> wrote:
>>> 
>>> On Fri, 2019-02-08 at 17:39 +0000, Andrey Petushkov wrote:
>>>> Dear team!
>>> 
>>> Hello Andrey (and everyone else),
>>> 
>>> Some notes first :)
>>> 
>>> I gave a look at both Azul and Alibaba patches, they are obviously very
>>> similar so I went with the latest consensus that is to integrate Azul
>>> patch first and then add some of the missing functionality from what
>>> Alibaba did, I also believe this makes it a tiny bit easier since it
>>> can be considered a progressive delta, however if the opposite is
>>> desired, please let me know.
>>> 
>>> I changed the subject line and cleaned the recipient list to make it a
>>> bit easier to follow the discussion, I expect people interested in this
>>> patch to be subscribed to jdk8u-dev mailing list anyway.
>>> 
>>> I will file a bug report to track the back porting effort, I think we
>>> will need to include the CSR at some point in the process, however I'll
>>> leave this to Andrew Haley as project maintainer, I think for the time
>>> being we can experiment on the incubator branch.
>>> 
>>> Also, we will need to keep this branch updated with latest changes from
>>> the primary repository, it is appreciated if other committers can
>>> participate in this task.
>>> 
>>> Finally, I take this opportunity to thank you and Guangyu and everyone
>>> else involved for those patches and for your patience as well as
>>> understanding regarding the process.
>>> 
>>> Ok, now on with the interesting bits ;)
>>> 
>>> I applied the patch from http://cr.openjdk.java.net/~apetushkov/jfr8/
>>> as described in you email below and compiled everything, --enable-jfr is enabled by default I think this should be disabled by default instead, however, when compiling even if I pass --disable-jfr I still see that the jfr files gets compiled and everything, so the option doesn't appear to do anything. What is the proper way to disable it?
>>> 
>>> Cheers,
>>> Mario
>>> 
>>> 
>>>> It happened that in the meanwhile Azul was also working on preparing
>>>> the backport of JFR to jdk8 codebase. We as well have finished the
>>>> implementation recently and would like to contribute it to the
>>>> community. Our approach and platform coverage is a bit different so
>>>> patches are definitely not the same. One might want to compare and
>>>> merge somehow to get most of both. We are performing difference
>>>> analysis but not yet finished. Will update once we have full
>>>> information.
>>>> The following is different to your approach:
>>>> - We did not add EnableJFR flag, JFR could be used in the same way as
>>>> in jdk11
>>>> - we have added —enable-jfr parameter to configure script to control
>>>> inclusion of JFR into binary
>>>> - -XX:+/-LogJFR VM parameter to control JFR logging (analogous to
>>>> -Xlog:jfr*=info), with -XX:+Verbose acting like -Xlog:jfr*=trace
>>>> - we have removed the “trace*” build files while adding “jfr*” files.
>>>> We believe such approach makes it easier to backport further jfr
>>>> fixes from jdk11+
>>>> - we have the following platforms supported: Linux x86, ppc, arm, all
>>>> in both 32-bit and 64-bit variants. Also Mac and Windows x86 32/64
>>>> are supported (Solaris on x86 and sparc have some bugs which are not
>>>> resolved yet)
>>>> We as well have JFR test suite passed with some limitations coming
>>>> from missing VM features (we did not add some missing WhiteBox APIs
>>>> so test coverage is a bit smaller than yours)
>>>> There might be as well difference in data being supported, e.g. Azul
>>>> implementation does not support G1 memory types, while Alibaba seem
>>>> to have added missing G1 functionality to get this data
>>>> 
>>>> The webrev is at
>>>> http://cr.openjdk.java.net/~apetushkov/jfr8/
>>>> Baseline is 8u202.
>>>> The changes to aarch64- and aarch32-port projects are delta on top of
>>>> generic hotspot patch.
>>>> Shenandoah GC is not integrated with JFR. BTW, related question to
>>>> RedHat, I see that aarch64-port/jdk8u repos are stale for ~5 months,
>>>> and latest update level is u181. Does that mean that OpenJDK aarch64-
>>>> jdk8u mainline is considered to be hosted in
>>>> http://hg.openjdk.java.net/aarch64-port/jdk8u-shenandoah/ repos
>>>> (where 8u192 is propagated for some time already)? Just in case I’ve
>>>> posted integration patches for both repos
>>>> 
>>>> Regards,
>>>> Andrey
>>>> 
>>>> 
>>>> Begin forwarded message:
>>>> 
>>>> From: "guangyu.zhu" <guangyu.zhu at aliyun.com<mailto:
>>>> guangyu.zhu at aliyun.com>>
>>>> Subject: Re: Proposal for back-porting JFR to OpenJDK8u
>>>> Date: January 29, 2019 at 3:41:35 AM PST
>>>> To: "Andrew Haley" <aph at redhat.com<mailto:aph at redhat.com>>, "Mario
>>>> Torre" <neugens.limasoftware at gmail.com<mailto:
>>>> neugens.limasoftware at gmail.com>>, "jdk8u-dev" <
>>>> jdk8u-dev-bounces at openjdk.java.net<mailto:
>>>> jdk8u-dev-bounces at openjdk.java.net>>
>>>> Cc: yumin qi <yumin.qi at gmail.com<mailto:yumin.qi at gmail.com>>,
>>>> kingsum.chow <kingsum.chow at gmail.com<mailto:kingsum.chow at gmail.com>>,
>>>> jdk8u-dev <jdk8u-dev at openjdk.java.net<mailto:
>>>> jdk8u-dev at openjdk.java.net>>, denghui.ddh <denghui.ddh at antfin.com<mai
>>>> lto:denghui.ddh at antfin.com>>
>>>> Reply-To: guangyu.zhu <guangyu.zhu at aliyun.com<mailto:
>>>> guangyu.zhu at aliyun.com>>
>>>> 
>>>> Hi there,
>>>> 
>>>> JFR backport patch has been uploaded to cr.openjdk. Please have a
>>>> review for the patch.
>>>> 
>>>> Webrev:
>>>> http://cr.openjdk.java.net/~luchsh/hs_jfr_cr/
>>>> http://cr.openjdk.java.net/~luchsh/jdk_jfr_cr/
>>>> 
>>>> The original patch comes from webrev:
>>>> http://cr.openjdk.java.net/~mgronlun/JEP328_FlightRecorder/Preview/webrev/index.html
>>>> . We ported it to jdk8u192-b26 and passed most of the jfr jtreg
>>>> tests on Linux/x86-64 platform. Some features related to Module, AOT
>>>> and log level are removed because jdk8 does not support them.
>>>> 
>>>> We have added a new option ‘EnableJFR’ to enable or disable the JFR
>>>> feature. It’s disabled by default. To enable it, you can start java
>>>> with ‘-XX:+EnableJFR’.  For example:
>>>> java -XX:+EnableJFR
>>>> -XX:StartFlightRecording=duration=1m,filename=rec.jfr MyApp
>>>> 
>>>> When running the jtreg test, please add the extra option '-vmoption:-
>>>> XX:+EnableJFR'. Otherwise the test case will not execute correctly.
>>>> Here is an example of running jfr jtreg test:
>>>> make test JTREG_TEST_EXTRA_OPTIONS=-vmoption:-XX:+EnableJFR
>>>> TEST=jdk_jfr
>>>> 
>>>> There is one more thing worth noting, please use jmc version 7.0 or
>>>> later to open the jfr record file.
>>>> 
>>>> Your suggestions and comments are welcome.
>>>> 
>>>> Thanks,
>>>> Guangyu Zhu
>>>> ------------------------------------------------------------------
>>>> Sender:李三红(三红) <sanhong.lsh at alibaba-inc.com<mailto:
>>>> sanhong.lsh at alibaba-inc.com>>
>>>> Sent At:2018 Dec. 17 (Mon.) 21:21
>>>> Recipient:Andrew Haley <aph at redhat.com<mailto:aph at redhat.com>>; Mario
>>>> Torre <neugens.limasoftware at gmail.com<mailto:
>>>> neugens.limasoftware at gmail.com>>
>>>> Cc:jdk8u-dev <jdk8u-dev at openjdk.java.net<mailto:
>>>> jdk8u-dev at openjdk.java.net>>
>>>> Subject:Re: Proposal for back-porting JFR to OpenJDK8u
>>>> 
>>>> Thanks for comments, we are preparing our internal patches for
>>>> webrev.
>>>> In the meantime, Martijn/ Andrew mentioned we can use AdoptOpenJDK to
>>>> produce technical preview build.
>>>> Would like to know this... can Martijn point us some guide somewhere?
>>>> 
>>>> Thanks!
>>>> Sanhong
>>>> -----邮件原件-----
>>>> 发件人: Andrew Haley [mailto:aph at redhat.com]
>>>> 发送时间: 2018年12月13日 3:03
>>>> 收件人: Mario Torre <neugens.limasoftware at gmail.com<mailto:
>>>> neugens.limasoftware at gmail.com>>; 李三红(三红) <
>>>> sanhong.lsh at alibaba-inc.com<mailto:sanhong.lsh at alibaba-inc.com>>
>>>> 抄送: Volker Simonis <volker.simonis at gmail.com<mailto:
>>>> volker.simonis at gmail.com>>; jdk8u-dev at openjdk.java.net<mailto:
>>>> jdk8u-dev at openjdk.java.net>
>>>> 主题: Re: Proposal for back-porting JFR to OpenJDK8u
>>>> 
>>>> On 12/12/18 1:44 PM, Mario Torre wrote:
>>>> I think the first thing to do would be to post the patch for review,
>>>> a
>>>> shared repository would make the review process a bit more complex I
>>>> think, and only makes sense if there is a need to work further on the
>>>> patch collectively.
>>>> 
>>>> Yes, exactly. That's the normal process, and it's the best place to
>>>> start.
>>>> 
>>>> --
>>>> Andrew Haley
>>>> Java Platform Lead Engineer
>>>> Red Hat UK Ltd. <https://www.redhat.com>
>>>> EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>> --
>>> Mario Torre
>>> Associate Manager, Software Engineering
>>> Red Hat GmbH <https://www.redhat.com>
>>> 9704 A60C B4BE A8B8 0F30  9205 5D7E 4952 3F65 7898
>> 
> 
> 
> --
> Mario Torre
> Associate Manager, Software Engineering
> Red Hat GmbH <https://www.redhat.com>
> 9704 A60C B4BE A8B8 0F30  9205 5D7E 4952 3F65 7898



More information about the jdk8u-dev mailing list