RFR (XS) 8061467: Add UseLargePages to TestNUMAPageSize

JC Beyler jcbeyler at google.com
Mon Aug 13 16:22:46 UTC 2018


Hi all,

I added Per's fix into the webrev:
http://cr.openjdk.java.net/~jcbeyler/8063110/webrev.02/

(I also put you Thomas as reviewer, I was not sure Per wanted to be put as
reviewer, let me know).

Not sure if you wanted to wait or not to push it but at least this version
would be the version that does not run with ZGC.

Thanks for your help,
Jc

On Mon, Aug 13, 2018 at 5:54 AM Per Liden <per.liden at oracle.com> wrote:

> On 08/13/2018 11:41 AM, Thomas Schatzl wrote:
> > Hi,
> >
> >    while testing this patch before push it seems like the problem is ZGC
> > which does not fall back to using small pages (which I did not know).
> >
> >> [5.048s][error][gc,init] Not enough space available on the backing
> > filesystem to hold the current max
> >> [5.048s][error][gc,init] Java heap size (128M). Forcefully lowering
> > max Java heap size to 0M (0%).
> >> Error occurred during initialization of VM
> >> java.lang.OutOfMemoryError: Java heap too small
> >
> > I will hold off pushing this before this issue has been resolved in one
> > way or the other.
>
> UseNUMA is enabled by default in ZGC, so it's tested frequently anyway.
> I guess the easiest solution is to just not run this test when ZGC is
> enabled, something like this:
>
> diff --git a/test/hotspot/jtreg/gc/TestNUMAPageSize.java
> b/test/hotspot/jtreg/gc/TestNUMAPageSize.java
> --- a/test/hotspot/jtreg/gc/TestNUMAPageSize.java
> +++ b/test/hotspot/jtreg/gc/TestNUMAPageSize.java
> @@ -27,7 +27,8 @@
>    * @summary Make sure that start up with NUMA support does not cause
> problems.
>    * @bug 8061467
>    * @requires vm.opt.AggressiveOpts != true
> - * @run main/othervm -Xmx128m -XX:+UseNUMA TestNUMAPageSize
> + * @requires vm.gc != "Z"
> + * @run main/othervm -Xmx128m -XX:+UseNUMA -XX:+UseLargePages
> TestNUMAPageSize
>    */
>
>   public class TestNUMAPageSize {
>
>
> cheers,
> Per
>
>
> >
> > Thanks,
> >    Thomas
> >
> > On Mon, 2018-08-13 at 10:23 +0200, Thomas Schatzl wrote:
> >> Hi,
> >>
> >> On Mon, 2018-08-13 at 09:58 +0200, Per Liden wrote:
> >>> Hmm, I suspect this might fail on many Linux systems, since it
> >>> might require configuring the huge page pool with enough pages.
> >>
> >>    from my understanding -XX:+UseLargePages will just give some
> >> warnings
> >>   at startup if it can not allocate pages. It will not fail.
> >>
> >> Assuming that the intent is to check whether when trying to use large
> >> pages (and the machine is properly configured), does not in any way
> >> make the VM crash, the warnings can be ignored.
> >>
> >> I am not aware there is a way to run tests on machines only that have
> >> the property "properly configured large pages" at this time.
> >>
> >> Thanks,
> >>    Thomas
> >>
> >
>


-- 

Thanks,
Jc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20180813/25ce2697/attachment.htm>


More information about the hotspot-gc-dev mailing list