Request for review (s) 8025635: SoftReferences are not cleared before metaspace OOME are thrown
Stefan Karlsson
stefan.karlsson at oracle.com
Thu Oct 17 12:55:17 UTC 2013
On 2013-10-16 22:56, Jon Masamitsu wrote:
> I've fixed the CMS case and updated the webrev.
>
> http://cr.openjdk.java.net/~jmasa/8025635/webrev.01/
Looks good.
StefanK
>
> With the fix an unneeded call to expand_and_allocate() may be
> made for CMS. I chose to keep the code simpler rather
> than to treat that CMS case.
>
> While working on this change I decided that a CMS concurrent
> cycle could be started inappropriately (after a STW full GC or
> after a foreground CMS collection) and have filed
>
> 8026752: Cancel MetaspaceGC request for a concurrent collection after GC
>
> See that CR for comments on the CMS behavior regarding collections for
> Metaspace allocations.
>
> Jon
>
>
> On 10/3/13 3:44 PM, Jon Masamitsu wrote:
>> 8025635: SoftReferences are not cleared before metaspace OOME are thrown
>>
>> If native memory has been exhausted or MaxMetaspaceSize has been set and
>> has been reached, a GC will be done to try and unload classes and
>> free metadata
>> space. If not enough metadata space has been freed, another GC which
>> clears
>> SoftReferences will be done.
>>
>> Note that GC's are normally done to unload classes and free metadata
>> space.
>> If not enough metadata space is freed by the GC, additional
>> space is allocated and used for the metadata allocation. It is only in
>> the situation where no additional native memory can be allocated for
>> metadata that SoftReferences are cleared.
>>
>> http://cr.openjdk.java.net/~jmasa/8025635/webrev.00/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20131017/11076a14/attachment.htm>
More information about the hotspot-gc-dev
mailing list