Incorrect Memory Usage Reported When Using ZGC + sun.misc.Unsafe
Alex Elent
alex at scalyr.com
Fri Jan 4 17:25:57 UTC 2019
Hi Erik,
For verbosity I have hosted both files in S3:
https://s3.amazonaws.com/aelent-public/TestUnsafe.java
https://s3.amazonaws.com/aelent-public/top_res_mem.png
I will look at NMT, thanks for the pointer!
On Fri, Jan 4, 2019 at 1:13 AM Erik Österlund <erik.osterlund at oracle.com>
wrote:
> Hi Alex,
>
> Unfortunately, attachments to the mailing list get eaten by the evil
> mail server monster. So I can't see any of the attachments you mentioned.
>
> Having said that, the kernel is known to be pretty bad at accounting
> resident memory when using multi-mapping. ZGC multi-maps the Java heap,
> so that there are multiple virtual addresses for the same physical
> memory. Unfortunately, the kernel doesn't get that it is indeed the same
> physical memory, and messes up in the accounting. That is most likely
> why you are seeing weird numbers for resident memory used.
>
> If you are curious about how much memory is actually being used, I think
> you are better off checking that using NMT. For more information, please
> consult the following documentation:
> https://docs.oracle.com/en/java/javase/11/vm/native-memory-tracking.html
>
> Hope this helps.
>
> /Erik
>
> On 2019-01-04 02:22, Alex Elent wrote:
> > Hello,
> >
> > Looking for some help on an issue we are experiencing exclusively with
> ZGC.
> > I'm unsure if the problem lies with ZGC or the kernel or where. Looking
> for
> > your experienced guidance if possible.
> >
> > We experienced this in our actual application but I managed to create a
> > simpler reproducible case.
> >
> > I've attached the example Java file (TestUnsafe.java).
> > I've attached a screenshot of top reporting 108% resident memory used.
> >
> > This is running on an EC2 t2.medium instance (2 vCPU / 4GB RAM).
> >
> > I compile with:
> > javac -cp . TestUnsafe.java
> >
> > I run with:
> > java -XX:+UnlockExperimentalVMOptions -XX:+UseZGC -Xms3500M -Xmx3500M
> -cp .
> > TestUnsafe
> >
> > Thank you so much for your help!
>
>
More information about the zgc-dev
mailing list