RFR (XS) 8251949: ZGC: Set explicit heap size for compiler/gcbarriers tests

Erik Österlund erik.osterlund at oracle.com
Tue Aug 18 15:45:25 UTC 2020


Hi Aleksey,

This seems fine. Although one of the reasons we implemented support for 
smaller heaps with ZGC
was as a response to various test failures, where tests would run on 
small heaps. And it seemed to
make more sense to implement support for small heaps to make all tests 
happy, than to try to find
all the tests that would have small heap sizes. With that being said, 
there might be more tests
popping up after a while, as we found ourselves. But until that day, 
this seems fine for now.

Thanks,
/Erik

On 2020-08-18 14:41, Aleksey Shipilev wrote:
> Testbug:
>   https://bugs.openjdk.java.net/browse/JDK-8251949
>
> On large TR 3970X, jtreg asks for MaxRAMFraction=0, which makes ZGC 
> use 128M heap for some tests. These tests reliably fail with OOM in 
> jdk-updates/jdk11u-dev:
>   compiler/gcbarriers/UnsafeIntrinsicsTest.java
>   compiler/loopopts/TestRangeCheckPredicatesControl.java
>
> While they do not fail in jdk/jdk, I think it makes more sense to add 
> explicit heap sizes in jdk/jdk, and then backport to 11u. The 
> alternative is to backport ZGC improvements that make it more 
> resilient at small heap sizes (like JDK-8234338?); but I am not 
> feeling very comfortable doing and testing that.
>
> Webrev:
>   https://cr.openjdk.java.net/~shade/8251949/webrev.01/
>
> Testing: affected tests in jdk/jdk and jdk-updates/jdk11u-dev
>




More information about the hotspot-gc-dev mailing list