(XS) RFR: 8148766: Test AvailableProcessors.java got wrong number of processors
Dmitry Samersoff
dmitry.samersoff at oracle.com
Tue Feb 2 09:50:44 UTC 2016
David,
Should we check that available > 0 ?
Looks good for me.
-Dmitry
On 2016-02-02 05:15, David Holmes wrote:
> Bug: https://bugs.openjdk.java.net/browse/JDK-8148766
>
> webrev: http://cr.openjdk.java.net/~dholmes/8148766/webrev/
>
> It seems some (arguably ill-behaved) systems can use aggressive power
> management to disable cores on the fly and at high frequency. As a
> result even using taskset the number of processors reported as available
> to the JVM can be less than what we asked for. This small tweak to the
> test simply changes the success criteria to only see more than the
> expected number of processors as an error.
>
> Thanks,
> David
>
> --- old/test/runtime/os/AvailableProcessors.java 2016-02-01
> 21:14:07.721558368 -0500
> +++ new/test/runtime/os/AvailableProcessors.java 2016-02-01
> 21:14:05.681443769 -0500
> @@ -27,7 +27,7 @@
>
> /*
> * @test
> - * @bug 6515172
> + * @bug 6515172 8148766
> * @summary Check that availableProcessors reports the correct value
> when running in a cpuset on linux
> * @requires os.family == "linux"
> * @library /testlibrary
> @@ -93,7 +93,9 @@
>
> static void checkProcessors(int expected) {
> int available = Runtime.getRuntime().availableProcessors();
> - if (available != expected)
> + // available can dynamically drop below expected due to
> aggressive power management
> + // but we should never have more than expected, else taskset is
> broken
> + if (available > expected)
> throw new Error("Expected " + expected + " processors, but
> found "
> + available);
> else
--
Dmitry Samersoff
Oracle Java development team, Saint Petersburg, Russia
* I would love to change the world, but they won't give me the sources.
More information about the hotspot-runtime-dev
mailing list