RFR: Keep root bits until the end of partial
Aleksey Shipilev
shade at redhat.com
Wed Jun 14 13:44:26 UTC 2017
Not really, because the first step clear *all* regions just in case, and
root_regions get populated much later. I'd prefer to keep it this way, until
further refactoring.
-Aleksey
On 06/14/2017 03:43 PM, Zhengyu Gu wrote:
> Probably should encapsulate the logic inside root_regions->clear()?
>
> -Zhengyu
>
> On 06/14/2017 09:36 AM, Aleksey Shipilev wrote:
>> Actually that was wrong. We still need to clean root bits at the first step,
>> because partial may get cancelled on OOME, leaving roots set indefinitely:
>> http://cr.openjdk.java.net/~shade/shenandoah/partial-keeproots/webrev.02/
>>
>> -Aleksey
>>
>> On 06/14/2017 03:34 PM, Zhengyu Gu wrote:
>>> Okay.
>>>
>>> -Zhengyu
>>>
>>> On 06/14/2017 09:31 AM, Aleksey Shipilev wrote:
>>>> We need to keep region->is_root bit until partial is over, because otherwise
>>>> hs_err from the mid-partial crash looks misleading: it could say that some
>>>> region is not root, while it actually is!
>>>>
>>>> See:
>>>> http://cr.openjdk.java.net/~shade/shenandoah/partial-keeproots/webrev.01/
>>>>
>>>> Thanks,
>>>> -Aleksey
>>>>
>>
>>
More information about the shenandoah-dev
mailing list