RFR (S): 8012086: The object count event should only send events for instances occupying more than 0.5% of the heap

Erik Helin erik.helin at oracle.com
Wed May 8 13:35:24 UTC 2013


Thanks!

Erik

On 05/08/2013 01:29 PM, Jesper Wilhelmsson wrote:
> Erik Helin skrev 8/5/13 1:00 PM:
>> Jesper,
>>
>> On 05/08/2013 10:20 AM, Jesper Wilhelmsson wrote:
>>> Erik Helin skrev 8/5/13 10:00 AM:
>>>> Jesper,
>>>>
>>>> On 05/07/2013 09:52 PM, Jesper Wilhelmsson wrote:
>>>>> Erik Helin skrev 7/5/13 9:28 PM:
>>>>>> Jesper,
>>>>>>
>>>>>> I've changed the flag to be an experimental flag instead of a product
>>>>>> flag.
>>>>>
>>>>> Thanks!
>>>>>
>>>>>> Please see new webrev at:
>>>>>> http://cr.openjdk.java.net/~ehelin/8012086/webrev.02/
>>>>>
>>>>> If you want to change the description of the flag to say "... to
>>>>> generate a trace event.", I think it would read better. But either way
>>>>> I'm fine with this change.
>>>>
>>>> I think that an even better description would be "... to generate a
>>>> possible
>>>> trace event", since even if the instances of the class takes up more
>>>> than
>>>> ObjectCountCutOffPercent of the used heap, there is no guarantee that
>>>> an event
>>>> will be sent (the event might be turned off).
>>>>
>>>> What do you think?
>>>
>>> I think that the fact that the event might be turned off is unrelated to
>>> this flag and that the description would look better without the
>>> "possible".
>>
>> Ok sure, I will remove "possible" before pushing the change :)
>>
>> All good?
>
> Yes, ship it! :-)
>
> Thanks for fixing this!
> /Jesper
>
>
>>
>> Thanks,
>> Erik
>>
>>> /Jesper
>>>
>>>>
>>>> Thanks,
>>>> Erik
>>>>
>>>>> /Jesper
>>>>>
>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>> Erik
>>>>>>
>>>>>> On 05/07/2013 03:54 PM, Jesper Wilhelmsson wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> Please check with the serviceability team (CC:ed) before adding a
>>>>>>> flag
>>>>>>> to control the threshold. I suspect that they would prefer some
>>>>>>> tracing
>>>>>>> specific way to configure these thresholds.
>>>>>>> /Jesper
>>>>>>>
>>>>>>>
>>>>>>> Erik Helin skrev 7/5/13 2:58 PM:
>>>>>>>> Hi Bengt,
>>>>>>>>
>>>>>>>> thanks for the review!
>>>>>>>>
>>>>>>>> On 05/07/2013 11:42 AM, Bengt Rutisson wrote:
>>>>>>>>>
>>>>>>>>> Hi Erik,
>>>>>>>>>
>>>>>>>>> Looks good.
>>>>>>>>>
>>>>>>>>> Two minor nits:
>>>>>>>>>
>>>>>>>>> Could we make 0.05 into a flag or at least a declared constant?
>>>>>>>>
>>>>>>>> Good suggestion, I've turned into a flag: ObjectCountCutOffPercent.
>>>>>>>>
>>>>>>>> On 05/07/2013 11:42 AM, Bengt Rutisson wrote:
>>>>>>>>> Could we rename _used_heap_in_words to something like
>>>>>>>>> _total_size_in_words or _base_size_in_words since it is not
>>>>>>>>> really the
>>>>>>>>> used count for the heap?
>>>>>>>>
>>>>>>>> Agree, I renamed it to _total_size_in_words.
>>>>>>>>
>>>>>>>> Please see new webrev at:
>>>>>>>> http://cr.openjdk.java.net/~ehelin/8012086/webrev.01/
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> Erik
>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> Bengt
>>>>>>>>>
>>>>>>>>> On 5/7/13 10:14 AM, Erik Helin wrote:
>>>>>>>>>> Hi all,
>>>>>>>>>>
>>>>>>>>>> this change makes sure that we only send the trace event
>>>>>>>>>> vm/gc/detailed/object_count_after_gc if the total size of all the
>>>>>>>>>> instances of a class occupies more than 0.5% of the used heap.
>>>>>>>>>>
>>>>>>>>>> Webrev:
>>>>>>>>>> http://cr.openjdk.java.net/~ehelin/8012086/webrev.00/
>>>>>>>>>>
>>>>>>>>>> Testing:
>>>>>>>>>> JPRT
>>>>>>>>>>
>>>>>>>>>> Thanks,
>>>>>>>>>> Erik
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>
>>>>
>>




More information about the hotspot-gc-dev mailing list