Withdrawn: 8332596: Shenandoah: Make TestAllocIntArrays.java#[iu-]aggressive and TestAllocObjectArrays.java#[iu-]aggressive more robustness

SendaoYan syan at openjdk.org
Fri Jul 26 01:00:50 UTC 2024


On Tue, 21 May 2024 09:12:09 GMT, SendaoYan <syan at openjdk.org> wrote:

> Hi all,
> The OOM failure seems a JVM bug, not a test bug. Testcase run OOM on Neoverse-N1 and Neoverse-N2, run pass on Kunpeng-920. It seems that, if the cpu work quickly, there testcases failed; if the cpu work slowly, there testcases passed.
> There 4 testcase run OOM:
> gc/shenandoah/TestAllocObjectArrays.java#aggressive
> gc/shenandoah/TestAllocObjectArrays.java#iu-aggressive
> gc/shenandoah/TestAllocIntArrays.java#aggressive
> gc/shenandoah/TestAllocIntArrays.java#iu-aggressive
> 
> 
> Before the OOM issue was investigated,  There two methods to avoid the OOM failures, maybe the 2rd method more reasonable.
> 
> 1. Problemlist these OOM testcases scenarios.
> 2. Use larger ShenandoahNoProgressThreshold and/or a smaller ShenandoahCriticalFreeThreshold to avoid the OOM failures.
> 
> In this PR, I choose the 2rd method. And I split `#aggressive` `#iu-aggressive` two parts as `os.arch!="aarch64"` and `os.arch=="aarch64"` separately, as show below:
> 
> `#aggressive` split to `#aggressive` and `#aggressive-aarch64`
> `#iu-aggressive` split to `#iu-aggressive` and `#iu-aggressive-aarch64`
> 
> The change has been verified locally, no risk.

This pull request has been closed without being integrated.

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

PR: https://git.openjdk.org/jdk/pull/19322


More information about the hotspot-gc-dev mailing list