RFR: 8059551: JEP-JDK-8043304: Test task: stress tests

[Vladimir Kozlov]

On 12/29/14 9:19 AM, Pavel Chistyakov wrote:
> Hi all,
>
> webrev: http://cr.openjdk.java.net/~pchistyakov/8059551/webrev.04/
> <http://cr.openjdk.java.net/%7Epchistyakov/8059551/webrev.04/>
>
> We need small update in compileBroker.cpp:255 for testing overload
> compile queue.
>
> 255     assert (_num_allocated_tasks < 10000, "Leaking compilation
> tasks?");
>
> This assert fails during test if we lock compilation and trying to
> enqueue a lot of methods for compilation.

The change seems reasonable.

Could you explain the logic which put methods on queue?
JVM don't add a method compilation task if such task already exist. How 
you bypass that check?

>
> The suggestion is to change condition like this:
>
> 255     assert (WhiteBoxAPI || WhiteBox::compilation_locked ||
> _num_allocated_tasks < 10000, "Leaking compilation tasks?");
>
> Checking only WhiteBox::compilation_locked flag is not enough, because
> even after compilation is unlocked we can fall into situation when
> _num_allocated_tasks will be greater than 10000 and test fails.

Why check WhiteBox::compilation_locked then? Should WhiteBoxAPI check be 
enough?

Thanks,
Vladimir

>
> -------------
> Thanks,
> Pavel
>
> On 19.12.2014 13:53, Pavel Chistyakov wrote:
>> Hi Christian,
>>
>> Yes, here is updated version:
>> http://cr.openjdk.java.net/~iignatyev/pchistyakov/8059551/webrev.03
>> Could you please take a look into it?
>>
>> --------
>> Thanks,
>> Pavel
>>
>> ----- Original Message -----
>> From: christian.thalinger at oracle.com
>> To: igor.ignatyev at oracle.com
>> Cc: pavel.chistyakov at oracle.com, hotspot-compiler-dev at openjdk.java.net
>> Sent: Friday, December 19, 2014 1:17:48 AM GMT +04:00 Abu Dhabi / Muscat
>> Subject: Re: RFR: 8059551: JEP-JDK-8043304: Test task: stress tests
>>
>>
>>> On Dec 15, 2014, at 5:34 AM, Igor Ignatyev <igor.ignatyev at oracle.com>
>>> wrote:
>>>
>>> Pavel,
>>>
>>> http://cr.openjdk.java.net/~iignatyev/pchistyakov/8059551/webrev.02/test/compiler/codecache/stress/Helper.java.html
>>>
>>>
>>>> 45             CLASS_DATA =
>>>> loadClassData(TestCaseImpl.class.getName());
>>>> 95             String className = TestCaseImpl.class.getName();
>>> these lines will load 'TestCaseImpl' class via default class loader,
>>> could you try to avoid it?
>> Any update on that?
>>
>>> Igor
>>>
>>> On 12/15/2014 04:15 PM, Pavel Chistyakov wrote:
>>>> Hi,
>>>>
>>>> please take a look into slightly updated webrev:
>>>> http://cr.openjdk.java.net/~iignatyev/pchistyakov/8059551/webrev.02/
>>>> <http://cr.openjdk.java.net/%7Eiignatyev/pchistyakov/8059551/webrev.02/>
>>>>
>>>>
>>>> --------------
>>>> Thanks,
>>>> Pavel
>>>>
>>>> On 12.12.2014 23:05, Vladimir Kozlov wrote:
>>>>> Looks fine.
>>>>>
>>>>> Thanks,
>>>>> Vladimir
>>>>>
>>>>> On 12/12/14 6:07 AM, David Chase wrote:
>>>>>> On 2014-12-12, at 5:14 AM, Pavel Chistyakov
>>>>>> <pavel.chistyakov at oracle.com> wrote:
>>>>>>
>>>>>>> David, thank you for review.
>>>>>>> Here is updated webrev:
>>>>>>> http://cr.openjdk.java.net/~iignatyev/pchistyakov/8059551/webrev.01
>>>>>>> <http://cr.openjdk.java.net/%7Eiignatyev/pchistyakov/8059551/webrev.01>
>>>>>>>
>>>>>> That is excellent -- now we just need a real reviewer.
>>>>>>
>>>>>> David
>>>>>>
>