On 2021-05-17 10:19, Aleksey Shipilev wrote:
On Mon, 17 May 2021 13:24:16 GMT, Jie Fu <ji...@openjdk.org> wrote:
Hi all,
vmTestbase/vm/mlvm/anonloader/stress/oome/metaspace/Test.java fails on our
many-core machines due to `-XX:MaxRAMPercentage=0`.
This is because `MaxRAMPercentage` will be always 0 if JTREG_JOBS > 25 [1].
It can be reproduced by: `make test
TEST="vmTestbase/vm/mlvm/anonloader/stress/oome/metaspace/Test.java"
JTREG="JOBS=26"` on almost all machines.
Setting `-XX:MaxRAMPercentage=0` on many-core machines seems unreasonable.
It would be better to fix it.
Thanks.
Best regards,
Jie
[1] https://github.com/openjdk/jdk/blob/master/make/RunTests.gmk#L741
`-XX:MaxRAMPercentage` is actually `double`, so it can accept values below 1.
The last time I looked into trouble like this, it was a problem with doing
floating-point division in the make files -- not sure if something can be done
about that.
Oh, if it's double, we can just switch to using awk to make the
calculation, just like we do for TEST_JOBS. I just did a quick check and
it produces float values.
But my point is that the failing test -- is that the only test that fails? --
expects some heap size to accommodate Java allocations until the expected
Metaspace OOM happens. In that case, the fix should be in the test itself.
Because even if we do +1 to `MaxRAMPercentage`, it might still be not enough.
If this test has special needs, those should of course be handled by the
test itself.
/Erik
