RFR 6987597: ManagementFactory.getGarbageCollectorMXBeans() returns empty list with CMS

[Jaroslav Bachorik]

On 21.11.2013 17:51, Mandy Chung wrote:
> Hi Jaroslav,
>
> On 11/19/2013 6:23 AM, Jaroslav Bachorik wrote:
>> Please, review this test fix.
>>
>> Issue : https://bugs.openjdk.java.net/browse/JDK-6987597
>> Webrev: http://cr.openjdk.java.net/~jbachorik/6987597/webrev.03
>>
>> The fix moves the execution of the test routine to the
>> GarbageCollectorMXBean notification handler - at that moment it's
>> certain that GC has already happened and it is safe to assert the data
>> provided by the MBean.
>
> This patch may have a potential issue when GC happens during the process
> of adding the notification handlers such as the number of increments is
> greater than the number of registered handlers.   By the time the main
> thread reaches p.arriveAndAwaitAdvance, the phaser has advanced to the
> next phase with few arrived parties and will wait forever (until the
> next GC happens).   This should rarely happen though.  Note that
> p.arriveAndAwaitAdvance continue to run even if the thread is
> interrupted and what happens if jtreg attempts to time out the test?

Thanks for catching this. It seems that using a Semaphore with adding 
permissions when a GC notification arrives would be more stable.

>
> Does this issue only happen to CMS with background GC thread?   The
> proposed patch seems a little overkill.   I wonder if the test should
> skip if running in CMS background mode.  Does
> -XX:+ExplicitGCInvokesConcurrent flag will get System.gc() to run in
> foreground mode in CMS (I think that may get the GC to complete before
> System.gc() returns?)

Yes, currently only CMS. It seems that the semantics of System.gc() 
("When control returns from the method call, the Java Virtual Machine 
has made a best effort to reclaim space from all discarded objects.") 
doesn't apply to the CMS collector.

Adding "-XX:-ExplicitGCInvokesConcurrent" should force System.gc() to 
wait till the GC has been performed. This could help resolve this 
particular problem.

-JB-

>
> Mandy
>