RFR: JDK-8302385: Remove MetaspaceReclaimPolicy=none

Thomas Stuefe stuefe at openjdk.org
Thu Feb 16 07:04:38 UTC 2023


On Wed, 15 Feb 2023 21:14:35 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> This patch removes the "MetaspaceReclaimPolicy=none" mode. It had been a safety measure when JEP-387 came out, but it has never really been needed. Since it adds complexity (especially in tests) I'd like to get rid of it.
>> 
>> For more details, please see the reasoning in CSR [JDK-8302130](https://bugs.openjdk.org/browse/JDK-8302130).
>> 
>> -------
>> 
>> Tests:
>> - I manually ran metaspace tests (see test group "hotspot_metaspace") on linux x64 (release, fastdebug) and x86 (fastdebug) 
>> - GHAs are scheduled.
>> 
>> -------
>> 
>> Reviewer hint:
>> 
>> Removing the "none" Policy basically hardcodes the behavioral settings to:
>> 
>> Settings::**uncommit_free_chunks** => **true**; (as in, we always do reclaim upon classloader death)
>> Settings::**new_chunks_are_fully_committed** => **false**; (as in, newly allocated chunks are only committed as far as needed)
>
> I looked at the diff with whitespace turned off. It seems to be a clean removal of the `none` policy.

Thanks @iklam and @dholmes-ora !

> 
> Not sure how that missing license header in the test was not previously detected!

Yes, that's strange, I thought someone was running automated checks for that. Maybe everyone thinks that :-)

-------------

PR: https://git.openjdk.org/jdk/pull/12554


More information about the hotspot-runtime-dev mailing list