RFR: 7109087: gc/7072527/TestFullGCCount.java fails when GC is set in command-line

[Mikael Gerdin]

Kevin,

On 04/17/2013 05:53 PM, Kevin Walls wrote:
> On 17/04/13 13:05, Mikael Gerdin wrote:
>> Kevin,
>>
>> On 2013-04-10 16:05, Kevin Walls wrote:
>>> Hi,
>>>
>>> I'd like a review of this testcase change.  It fails as it has a GC
>>> option set in the jtreg tag which may conflict with what jtreg sends,
>>> and fail to run.
>>>
>>> Although the test was created for a CMS problem, double-counting of full
>>> collections, here it is adapted to check all collectors:
>>>
>>> http://cr.openjdk.java.net/~kevinw/7109087/webrev/
>>
>> Since you know the number of iterations I think it would be better to
>> use an ArrayList and use list.ensureCapacity(20) to avoid allocations
>> in addCollectionCount.
>>
>> You could also make the GarbageCollectorMXBean list available from a
>> static variable since I'm not sure what happens in that call path.
>>
>> Here you access theseCounts as a List instead of a List<Long>, maybe
>> the "counts" map should be a Map<String, List<Long>>?
>>
>> +            for (int i = 0; i < iterations - 1; i++) {
>> +                List theseCounts = counts.get(collector);
>> +                long a = (Long) theseCounts.get(i);
>>
>> /Mikael
>>
>>
>>>
>>> Thanks
>>> Kevin
>
>
> Hi Mikael, thanks -
>
> ensureCapacity: not sure it stops the addCollectionCount() method
> changing/adding to the list?  We could verify the list is of the right
> length, although I don't think it's that important?...

I meant that ensureCapacity() on ArrayList will make sure that we won't 
need to reallocate the backing array when adding elements.
LinkedList on the other hand will allocate a new list node every time 
you add an element to the list.

We can't be 100% sure that we won't trigger a spurious GC when running 
the test but if we avoid allocating we can reduce the risk somewhat.

/Mikael

> addCollectionCount adds one thing to the List for each collector each
> time it's called...  I don't think we can guarantee that collectors
> aren't getting invoked at any time, but we can make it unlikely, like
> the next point you make:
>
> Yes, don't really need to call  into the ManagementFactory each
> iteration, and can reduce it to a single call to
> ManagementFactory.getGarbageCollectorMXBeans();  We don't need the
> List<MemoryPoolMXBean> at all.
>
> Yes, those casts to Long can be eradicated.
>
> Also there was an unused "count" variable in main(), removed it.. 8-)
>
> And I should clone from hotspot-gc for this change.
>
> New webrev: http://cr.openjdk.java.net/~kevinw/7109087/webrev.02/
>
> Thanks
> Kevin
>