hotspot-gc-dev Digest, Vol 198, Issue 56

[Thomas Schatzl]

Hi,

On 28.12.23 12:59, 劉俠麟 wrote:
> Hi Thomas
> Thank you very much for your suggestions. Increasing G1HeapRegionSize 
> really helped me.
> I successfully reproduce the To-space Exausted with the attached file. I 
> also attached the gclog. Here's my java version and cmds:
> 
> openjdk version "11.0.18" 2023-01-17
> 
> OpenJDK Runtime Environment Temurin-11.0.18+10 (build 11.0.18+10)
> 
> OpenJDK 64-Bit Server VM Temurin-11.0.18+10 (build 11.0.18+10, mixed mode)
> 
> 
> java -Xmn1g -Xmx4g -Xms4g -Xlog:gc,gc+heap,gc+region*:gc.log:tags 
> -XX:+UseG1GC -XX:G1HeapRegionSize=1M 
> -XX:MaxTenuringThreshold=1TospaceExaustedReproduce
> 
> 

Thanks for the reproducer. I played a bit with it, and could actually 
reproduce evacuation failures with all versions I tried. However this is 
no surprise: there is some strong element of timing involved in such cases.

Great that increasing heap region size worked for you.

> I also tried java-21.0.1 but  can not reproduce it again. I'm still 
> wondering whether it's a bug in JDK11 or not.
> 

I could not find any particular oddities with jdk 11; later versions 
optimized generation sizes quite a bit, which will change timing.

In particular there is also some changes in later versions that tries to 
address such humongous allocation spikes (or iow, taking them into 
account much better).

Thanks for your reproducer (and apologies for the somewhat late reply...),
   Thomas