Strange behaviour on half-occupied heap

[Aleksey Shipilev]

On 08/30/2018 07:38 PM, Aleksey Shipilev wrote:
> On 08/30/2018 07:28 PM, Simon Ogorodnik wrote:
>> Find out that issue related to my local build, and not reproduces with build from CI with same sources
> 
> I can see how the zero space for CSet can happen, even if it is intermittent. I think evac-reserve
> would help to mitigate bad behaviors on densely populated heaps:
>   http://mail.openjdk.java.net/pipermail/shenandoah-dev/2018-August/007270.html
> 
> If everything goes well, it should land in sh/jdk8 early next week.

Current sh/jdk8 repository and builds contain evac-reserve work, along with other touchups. This
should work much better on small + densely populated heaps. I had a few runs with CLion and -Xmx2g
with LDS of ~1G, and it seems to work fine. Please try again?

The other question, why Full GC reclaimed a lot of memory, but Degen did not -- these are softrefs,
and only Full GC used to clean them up. Should be better with the patch currently in testing:
  http://mail.openjdk.java.net/pipermail/shenandoah-dev/2018-September/007337.html

The sh/jdk8u variant of that patch is:
  http://cr.openjdk.java.net/~shade/shenandoah/8u-softrefs.patch

You may want to apply it separately, if interested. Otherwise, it should arrive in the next batch of
backports next week.

-Aleksey