RFR(xs): 8225200: runtime/memory/RunUnitTestsConcurrently.java has a memory leak

[Thomas Stüfe]

Thanks Coleen!

On Mon, Jul 1, 2019, 21:14 <coleen.phillimore at oracle.com> wrote:

> +1
> Thank you for taking care of this!
> Coleen
>
> On 7/1/19 3:07 PM, Thomas Stüfe wrote:
> > Thanks Stefan!
> >
> > On Mon, Jul 1, 2019, 21:06 Stefan Karlsson <stefan.karlsson at oracle.com>
> > wrote:
> >
> >> On 2019-07-01 20:56, Thomas Stüfe wrote:
> >>> Hi all,
> >>>
> >>> may I please have reviews and opinions about the following patch:
> >>>
> >>> Issue: https://bugs.openjdk.java.net/browse/JDK-8227041
> >>> cr:
> >>>
> >>
> http://cr.openjdk.java.net/~stuefe/webrevs/8227041-rununittestsconcurrently-has-a-mem-leak/webrev.00/webrev/index.html
> >>> There is a memory leak in test_virtual_space_list_large_chunk(), called
> >> as
> >>> part of the whitebox tests WB_RunMemoryUnitTests(). In this test
> >> metaspace
> >>> allocation is tested by rapidly allocating and subsequently leaking a
> >>> metachunk of ~512K. This is done by a number of threads in a tight loop
> >> for
> >>> 15 seconds, which usually makes for 10-20GB rss. Test is usually OOM
> >> killed.
> >>> This test seems to be often excluded, which makes sense, since this
> leak
> >>> makes its memory usage difficult to predict.
> >>>
> >>> It is also earmarked by Oracle for gtest-ification, see 8213269.
> >>>
> >>> This leak is not easy to fix, among other things because it is not
> clear
> >>> what it is it wants to test. Meanwhile, time moved on and we have quite
> >>> nice gtests to test metaspace allocation (see e.g.
> >>> test_metaspace_allocation.cpp) and I rather would run those gtests
> >>> concurrently. Which could be a future RFE.
> >>>
> >>> So I just removed this metaspace related test from
> >> WB_RunMemoryUnitTests()
> >>> altogether, since to me it does nothing useful. Once you remove the
> >> leaking
> >>> allocation, not much is left.
> >>>
> >>> Without this part RunUnitTestsConcurrently test runs smoothly through
> its
> >>> other parts, and in that form it is still useful.
> >>>
> >>> What do you think?
> >> I think this makes sense and it looks good to me.
> >>
> >> Thanks,
> >> StefanK
> >>
> >>> Cheers, Thomas
> >>
>
>