RFR (M) JDK-6479360 - detailed dumping of class size statistics

Staffan Larsen staffan.larsen at oracle.com
Thu Jan 24 23:43:43 PST 2013 

Hi Ioi,

I was looking at the change in diagnosticCommand.cpp. In your change the new command is only available when running with -XX:+ UnlockDiagnosticVMOptions. I was wondering if it would make sense to instead always expose the diagnostic command, and check UnlockDiagnosticVMOptions in the execute() method. The execute() method can then print a helpful message saying that you need to enable UnlockDiagnosticVMOptions if it isn't enabled. This would make it easier to discover the availability of the diagnostic command and also give users a hint on how to enable it.

Thanks,
/Staffan

On 24 jan 2013, at 20:52, Ioi Lam <ioi.lam at oracle.com> wrote:

> Hi,
> 
> I have updated the patch:
> 
>     http://cr.openjdk.java.net/~iklam/6479360/class_stats_012/
> 
> [1] Removed FOREACH_COLUMN and SELECTED per Harold
> [2] Modified HeapInspection constructor per Coleen.
> [3] Renamed a few columns in the output to match the capitalizations
>     in source code:
>     Constmethod -> ConstMethod
>     Methoddata -> MethodData,  etc
> 
> Thanks!
> - Ioi
> 
> 
> On 01/24/2013 08:34 AM, harold seigel wrote:
>> Hi Ioi,
>> 
>> I think we want to get away from using lots of macros, such as these that were added to heapinspection.cpp:
>> +#define FOREACH_COLUMN(c) \
>>  + for (int c=0; c<KlassSizeStats::_num_columns; c++)
>> +
>> +#define SELECTED(c) selected_columns_table[c]
>> Would you consider removing them?
>> 
>> Thanks, Harold
>> 
>> On 1/23/2013 8:24 PM, Coleen Phillimore wrote:
>>> 
>>> 
>>> In vmGCOperations.cpp the type HeapInspection inspect looks like the only initialization of these fields is in the set functions.
>>> 
>>> It would be cleaner to declare the constructor to take the initial values in the declaration, ie:
>>> 
>>>     HeapInspection inspect(_csv_format, _print_help, _print_class_stats, _colmns);
>>> 
>>> and then you can delete the setter functions.
>>> 
>>> The rest of the code looks good.
>>> Coleen
>>> 
>>> On 1/23/2013 6:59 PM, Karen Kinnear wrote:
>>>> 
>>>> Ioi,
>>>> 
>>>> Thanks for finding that - I had missed that. That makes a lot of sense. Thank you.
>>>> So - you have my approval for checking this in. You need one more code reviewer.
>>>> 
>>>> thanks,
>>>> Karen
>>>> 
>>>> On Jan 23, 2013, at 6:57 PM, Ioi Lam wrote:
>>>> 
>>>>> Karen,
>>>>> 
>>>>> Thanks for the review.
>>>>> 
>>>>> hotspot/make/excludeSrc.make already excludes the entire file if INCLUDE_SERVICES=0. I verified that heapInspection.o does not exist for the "linux_i486_minimal1" build, which has INCLUDE_SERVICES=0.
>>>>> 
>>>>> ifeq ($(INCLUDE_SERVICES), false)
>>>>>       CXXFLAGS += -DINCLUDE_SERVICES=0
>>>>>       CFLAGS += -DINCLUDE_SERVICES=0
>>>>> 
>>>>>       Src_Files_EXCLUDE += heapDumper.cpp heapInspection.cpp \
>>>>>     attachListener_linux.cpp attachListener.cpp
>>>>> endif
>>>>> 
>>>>> - Ioi
>>>>> 
>>>>> On 01/23/2013 03:31 PM, Karen Kinnear wrote:
>>>>>> Ioi,
>>>>>> 
>>>>>> Thank you for the cleanups. Looks good. Ok to check in.
>>>>>> 
>>>>>> Or - if you are up to adding INCLUDE_SERVICES around a bit more - 
>>>>>> 
>>>>>> including heapInspection.cpp/hpp (at least that entire table of
>>>>>> strings) and print_class_stats and the call to it. The goal is to reduce memory
>>>>>> usage when not needed.
>>>>>> 
>>>>>> And thank you for testing with that flag off.
>>>>>> 
>>>>>> thanks,
>>>>>> Karen
>>>>>> 
>>>>>> (p.s. you might ask Jiangli for a code review)
>>>>>> 
>>>>>> On Jan 23, 2013, at 5:18 PM, Ioi Lam wrote:
>>>>>> 
>>>>>>> Thanks to Karen for the feedback. I have updated the webrev:
>>>>>>> 
>>>>>>>     http://cr.openjdk.java.net/~iklam/6479360/class_stats_011/
>>>>>>> 
>>>>>>> Changes since last time
>>>>>>> 
>>>>>>>     + Cleaned up typos found by Karen
>>>>>>> 
>>>>>>>     + Use JULONG_FORMAT. Remove all uses of PRIu64 and 
>>>>>>>       UINT64_FORMAT which may be unportable on MacOS (see 7102489).
>>>>>>> 
>>>>>>>     + Updated copyright to 2013
>>>>>>> 
>>>>>>>     + Added more #if INCLUDE_SERVICES. GC.class_histogram and 
>>>>>>>       GC.class_stats are no longer accessible by jcmd if
>>>>>>>       INCLUDE_SERVICES=0.
>>>>>>> 
>>>>>>> Thanks
>>>>>>> - Ioi
>>>>>>> 
>>>>>>> 
>>>>>>> On 01/22/2013 02:07 PM, Karen Kinnear wrote:
>>>>>>>> Ioi,
>>>>>>>> 
>>>>>>>> Thank you for doing this. I like the way you have the size collection in the file
>>>>>>>> with the metadata definitions.
>>>>>>>> 
>>>>>>>> 1. diagnosticCommand.hpp line 200 - inspeactheap -> inspectheap
>>>>>>>> 
>>>>>>>> 2. please update copyrights on all files you changed to 2013
>>>>>>>> 
>>>>>>>> 3. I sent an email to the embedded folks to find out if they want this
>>>>>>>>    under INCLUDE_SERVICES.
>>>>>>>>    If they do ...
>>>>>>>> 
>>>>>>>>    - did you build with and without INCLUDE_SERVICES?
>>>>>>>>    What happens when you build with it off and run the jcmd ?
>>>>>>>>    I am confused that in diagnosticCommand.cpp ClassStatsDcmd is available without
>>>>>>>>    INCLUDE_SERVICES, but the actual details below the collect_statistics are not?
>>>>>>>>    Did you talk to the embedded team about whether they want this or not?
>>>>>>>> 
>>>>>>>>    - I would recommend that include files also add the new methods conditional
>>>>>>>>   on INCLUDE_SERVICES - you did that in some of them, but I think you missed:
>>>>>>>>   method.hpp, methodData.hpp, constantPool.hpp, constMethod.hpp, annotations.hpp
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 4. heapInspection.hpp line 56: is InstBytesi supposed to be InstBytes?
>>>>>>>> 
>>>>>>>> 5. heapInspection.hpp line 122 - you have some funny "\" - is that intentional?
>>>>>>>> 
>>>>>>>> 6. INT64_FORMAT: I think Harold just put back a change for julong handling
>>>>>>>> that switched to using JULONG_FORMAT- see KlassInfoHisto::print_class_stats - 2 places
>>>>>>>> (to ensure this also works on Mac OS/X)
>>>>>>>> 
>>>>>>>> thanks,
>>>>>>>> Karen
>>>>>>>> 
>>>>>>>> On Jan 16, 2013, at 7:39 PM, Ioi Lam wrote:
>>>>>>>> 
>>>>>>>>> Please review:
>>>>>>>>>     http://cr.openjdk.java.net/~acorn/class_stats_010/
>>>>>>>>> 
>>>>>>>>> Bug: RFE: PrintClassHistogram improvements
>>>>>>>>>     https://jbs.oracle.com/bugs/browse/JDK-6479360
>>>>>>>>> 
>>>>>>>>> Sponsor: Karen Kinnear
>>>>>>>>> 
>>>>>>>>> Summary:
>>>>>>>>> 
>>>>>>>>> A new diagnostic command "jcmd GC.class_stats" is added. The user
>>>>>>>>> can invoke this command to connect to a live JVM and dump a detailed
>>>>>>>>> report of size statistics of all loaded classes (including array
>>>>>>>>> classes and anonymous classes).
>>>>>>>>> 
>>>>>>>>>     ==========SYNOPSIS===================================
>>>>>>>>>     $ jcmd $PID help GC.class_stats
>>>>>>>>>     Provide statistics about Java class meta data.
>>>>>>>>>     Impact: High: Depends on Java heap size and content.
>>>>>>>>> 
>>>>>>>>>     Syntax : GC.class_stats [options] [<columns>]
>>>>>>>>> 
>>>>>>>>>     Arguments:
>>>>>>>>> 	columns : [optional] Comma-separated list of all the columns to
>>>>>>>>>                   show. If not specified, the following columns are shown:
>>>>>>>>>                   InstBytes,KlassBytes,CpAll,annotations,MethodCount,Bytecodes,
>>>>>>>>>                   MethodAll,ROAll,RWAll,Total (STRING, no default value)
>>>>>>>>> 
>>>>>>>>>     Options: (options must be specified using the <key> or <key>=<value> syntax)
>>>>>>>>> 	-all : [optional] Show all columns (BOOLEAN, false)
>>>>>>>>> 	-csv : [optional] Print in CSV (comma-separated values) format for 
>>>>>>>>>                 spreadsheets (BOOLEAN, false)
>>>>>>>>> 	-help : [optional] Show meaning of all the columns (BOOLEAN, false)
>>>>>>>>>     ======================================================
>>>>>>>>> 
>>>>>>>>> By default, the output is human-readable tabulated text format. The
>>>>>>>>> user can also select CSV format (comma-separated values) to be
>>>>>>>>> used with spreadsheets.
>>>>>>>>> 
>>>>>>>>> A large number of columns are available. By default, a few "summary 
>>>>>>>>> columns" (e.g., size of Klass objects, total read-only data, etc) 
>>>>>>>>> are printed. The user can also manually select the columns.
>>>>>>>>> 
>>>>>>>>> Please see the attachments in the bug for sample output, as well as
>>>>>>>>> a listing of all the columns and their meanings:
>>>>>>>>> 
>>>>>>>>> https://jbs.oracle.com/bugs/browse/JDK-6479360?focusedCommentId=13290360&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13290360
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> Impact:
>>>>>>>>> 
>>>>>>>>> If this diagnostic command is not used, there should be no
>>>>>>>>> impact to the JVM's execution, except for
>>>>>>>>> 
>>>>>>>>>    + libjvm.so footprint increase (about 10KB larger on Linux/x64)
>>>>>>>>>    + one additional virtual method in Klass.
>>>>>>>>> 
>>>>>>>>> This feature is excluded when INCLUDE_SERVICES=0 (e.g., 
>>>>>>>>> embedded builds).
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> Tests run:
>>>>>>>>> 
>>>>>>>>> + JPRT -- (hotspot only) to verify build-ability
>>>>>>>>> 
>>>>>>>>> + Manually ran "jcmd $PID help GC.class_stats" on Linux/x64
>>>>>>>>> 
>>>>>>>>> + I intend to add a new testcase once this is committed:
>>>>>>>>>   https://jbs.oracle.com/bugs/browse/JDK-8006413
>>>>>>>>>   Add utility classes for writing better multiprocess tests in jtreg
>>>>>>>>> 
>>>>>>>>> Thanks,
>>>>>>>>> Ioi
>>>>>>>> 
>>>>>>> 
>>>>>> 
>>>>> 
>>>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/attachments/20130125/7553dd7e/attachment-0001.html

More information about the hotspot-runtime-dev mailing list