RFR for JDK-6963118 Intermittent test failure: test/java/nio/channels/Selector/Wakeup.java fail intermittently (win)

[srikalyan]

Hi Alan, please find the new webrev at 
http://cr.openjdk.java.net/~srikchan/Regression/6963118-Wakeup-webrev-V2/  
, this includes the refactoring you did + timing adjustments. Ideally 
with the fix the test should never fail which holds true on all 
platforms except windows(though failures reduced to 2/1000 as opposed to 
10/1000(per our experiments)).

Possible root cause:
There are atleast 2 places in the windows specific Selector 
implementation(WindowsSelectorImpl) which has wakeup() calls, this is 
not the case in Solaris/Linux implementations(DevPollSelectorImpl and 
EPollSeclectorImpl). It would be helpful if some from NIO team can 
explain when the wakeup() in these 2 places are invoked which i suspect 
could cause spurious wakeup from select(during double wakeup phase of 
test/java/nio/channels/Selector/Wakeup.java on windows). Please see the 
attachment in JDK-6963118 
<https://bugs.openjdk.java.net/browse/JDK-6963118> .

---
Thanks
kalyan

On 12/30/2013 10:47 AM, srikalyan wrote:
> Hi Alan, sorry for the delay. I have been trying to incorporate sleep 
> and make the code succeed but getting intermittent failures. The 
> reason is it is extremely difficult to synchronize the Sleeper and 
> Checker(main) threads as there is not way the Sleeper could 
> communicate to Checker that it is successfully blocked on the select() 
> call. At most we can mitigate by bring both threads synchronize at a 
> certain point and then make the Checker wait for some long time to 
> allow the Sleeper to march into and wait at select() call.I will do 
> some repeated runs of your version as well(meanwhile trying other 
> ways) and let you know before we flag ok on this.
>
> -- 
> Thanks
> kalyan
> Ph: (408)-585-8040
>
>
> On 12/30/13, 6:24 AM, Alan Bateman wrote:
>>
>> I didn't see any more on this one but I think it would be good to get 
>> it fixed and removed from the exclude list. Here's a slightly 
>> modified fix that might be a bit clearer:
>>
>> http://cr.openjdk.java.net/~alanb/6963118/webrev/
>>
>> If we can agree this then I'll get it into jdk9/dev.
>>
>> -Alan.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/nio-dev/attachments/20140110/bf8e5e2e/attachment.html