RFR: 8040935: -XX:+AggressiveOpts broken: GC triggered before VM initialization completed on several tests

Kim Barrett kim.barrett at oracle.com
Wed Jan 14 18:13:39 UTC 2015


On Jan 12, 2015, at 5:01 AM, Stefan Johansson <stefan.johansson at oracle.com> wrote:
> 
> Please review this testfix to avoid the following issue:
> https://bugs.openjdk.java.net/browse/JDK-8040935
> 
> Webrev:
> http://cr.openjdk.java.net/~sjohanss/8040935/hotspot.00
> 
> Summary:
> Making the test a driver test to avoid dealing with externally specified flags. One problematic flag is AggressiveOpts, it causes the VM to allocate more stuff during startup and that together with UseNUMA the eden space gets to little to handle all startup allocations. This will in turn lead to a GC being triggered before initialization is finished, and because of that failing the test.
> 
> Testing:
> * Tested via JPRT

test/gc/TestNUMAPageSize.java
Summary: Changed test to not be affected by external options.

I'm not sure I like this approach to resolving the bug.  I think it
results in many potentially interesting configurations never being
tested, e.g. this tests with one and only one GC, the default GC.  It
ignores the nightly build cycling through various configurations to
increase test coverage.



More information about the hotspot-gc-dev mailing list