RFR: 8306825: Monitor deflation might be accidentally disabled by zero intervals [v2]

Thu Apr 27 14:00:53 UTC 2023

On Thu, 27 Apr 2023 09:14:55 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

>> Found this mistake while backporting [JDK-8305994](https://bugs.openjdk.org/browse/JDK-8305994): if we put GuaranteedAsyncDeflationInterval=0 (defined as "disable guaranteed deflation"), then the MonitorDeflationThread would call `wait(0)`, that is, wait indefinitely for the notify. This breaks triggering the threshold heuristics!
>> 
>> But what is even more concerning, the same thing would happen even prior to [JDK-8305994](https://bugs.openjdk.org/browse/JDK-8305994), if we put GuaranteedSafepointInterval=0 too. Which means users who kept GSI=0 as the way to dodge excess safepoints, now effectively disable monitor deflation!
>> 
>> New test cases capture most of the interesting paths/combinations. Since I needed to add more test cases, I ended up refactoring the test for more clarity. The test is also renamed, because now it tests not only GADI, but other intervals too.
>
> Aleksey Shipilev has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Drop unneccessary options

Marked as reviewed by dcubed (Reviewer).

Yup. I just kicked off a Mach5 Tier1 and I'll do the Repeat-50 runs...

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

PR Review: https://git.openjdk.org/jdk/pull/13635#pullrequestreview-1404090641
PR Comment: https://git.openjdk.org/jdk/pull/13635#issuecomment-1525733166