RFR: 8254027: gc/g1/TestHumongousConcurrentStartUndo.java failed with "'Concurrent Mark Cycle' missing from stdout/stderr"

[Thomas Schatzl]

On Thu, 8 Oct 2020 02:35:40 GMT, Kim Barrett <kbarrett at openjdk.org> wrote:

>> Hi all,
>> 
>>   can I get reviews for this change to the test to verify concurrent mark/undo cycles to make it more stable?
>> 
>> In the failing scenario, the second concurrent cycle that would typically result in a complete concurrent mark would
>> start with too few strongly reachable humongous objects, so instead g1 performs an Undo cycle, resulting in that
>> message.  The changes make sure that if the second concurrent start pause starts there are always enough strongly
>> reachable humongous objects there to disallow an undo cycle (because the reason we started the cycle has been
>> allocation of that amount of humongous objects that are all guaranteed to be kept reachable).  Testing: could not
>> reproduce the failure, but 2k runs on linux/aarch64 did not show an issue.
>
> Looks good.

Thanks @kimbarrett , @kstefanj for your reviews.

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

PR: https://git.openjdk.java.net/jdk/pull/523