RFR 8204525: [TESTBUG] runtime/NMT/MallocStressTest.java ran out of java heap

[David Holmes]

Hi Harold,

On 21/11/2018 11:08 pm, Harold David Seigel wrote:
> Hi David,
> 
> Thanks for looking at this change.
> 
> When growing the mallocd_memory array list, the JDK creates a completely 
> new copy.  This can a large allocation and is more likely to cause an 
> OutOfMemoryException than the small println() call.  

Sure, but you have many of these threads running at once so the heap 
could be being exhausted in one thread at the same time as another hits 
the malloc limit and goes to do the println. And println can trigger a 
lot of memory use on its first use.

> But, since the 
> println() call could be the proverbial straw that breaks the camel's 
> back, how about if I just remove the println() call?

Works for me. :)

Thanks,
David

> Thanks, Harold
> 
> 
> On 11/20/2018 5:44 PM, David Holmes wrote:
>> Hi Harold,
>>
>> Fix seems reasonable but the problem suggests that the test could 
>> still fail spuriously if it runs out of Java heap around the same time 
>> as malloc'd memory - the println will throw OOME.
>>
>> Cheers,
>> David
>>
>> On 21/11/2018 12:46 am, Harold David Seigel wrote:
>>> Hi,
>>>
>>> Please review this small change to fix test bug JDK-8204525. Test 
>>> MallocStressTest quickly allocates lots of malloc memory and slowly 
>>> releases it.  It checks that NMT properly measures the amount of 
>>> allocated memory.  The test occasionally fails when it runs out of 
>>> Java heap before the malloc memory allocations completed.
>>>
>>> The test has several threads that loop, allocating lots of memory.  A 
>>> thread's allocations stop when its malloc allocation fails, or when a 
>>> certain amount of time expires.  This fix also stops a thread's 
>>> allocation when an OutOfMemoryError exception gets thrown.
>>>
>>> Open Webrev: http://cr.openjdk.java.net/~hseigel/bug_8204525/webrev/
>>>
>>> JBS Bug:  https://bugs.openjdk.java.net/browse/JDK-8204525
>>>
>>> The fix was tested on Linux X64, Windows, Solaris, and Mac.  It was 
>>> also tested on Linux X64 with -Xmx20M, which, before this fix, 
>>> reliably reproduced the failure.
>>>
>>> Thanks, Harold
>>>
>