RFR: 8314139: TEST_BUG: runtime/os/THPsInThreadStackPreventionTest.java could fail on machine with large number of cores

[Aleksey Shipilev]

On Fri, 11 Aug 2023 05:41:52 GMT, Man Cao <manc at openjdk.org> wrote:

> Hi all, @tstuefe 
> 
> Could anyone review this fix for a test failure that could happen on machines with many cores? This fix caps `-XX:ActiveProcessorCount` to 64, which limits the number of internal threads the JVM creates.
> 
> An alternative is to bump up the `acceptableRSSLimitMB` such as increasing `basicRSSOverheadMB`. However, this approach could be fragile for machines with even more number of cores (e.g. 256 or more).
> 
> -Man

This makes sense to me. 

Since we know the internal VM threads participate in calculation, I think we would be even better off with unconditionally setting `-XX:ActiveProcessorCount=2` to cover all machine shapes at once. Setting it to `1` would treat the machine as uni-processor, which we don't want, hence `2`.

-------------

PR Review: https://git.openjdk.org/jdk/pull/15241#pullrequestreview-1573296501