Review Request for 7142919 TEST_BUG: java/nio/channels/AsyncCloseAndInterrupt.java failing intermittently [sol11]

Dan Xu dan.xu at oracle.com
Mon Sep 17 16:22:57 PDT 2012


Alan,

Thanks for your good comments. I changed my fix basing on your 
suggestions and generated a new webrev at 
http://cr.openjdk.java.net/~dxu/7142919/webrev.01/

In the new version, I also made a small change in pumpRefuser(). Now I 
assume the connection backlog is saturated if a client cannot connect 
to the refuser within 50 milliseconds.

Thanks,

-Dan

On Sun 16 Sep 2012 12:54:28 PM PDT, Alan Bateman wrote:
> On 11/09/2012 23:54, Dan Xu wrote:
>> Please review the fix of CR 7142919 at
>> http://cr.openjdk.java.net/~dxu/7142919/webrev/.
>>
>> The change is to create a pump thread dedicated to saturate refuser's
>> connection backlog. And it also removes this test from problem list.
>> Thanks!
>>
>> -Dan
> Thanks for working on this one, this is a difficult scenario to test.
>
> If I understand the changes correctly then the pump thread loops
> continuously to ensure that the listener backlog is saturated. If it's
> the only test running then this should be fine but I worry that it
> might not be as effective as it initially looks when running
> concurrently with other tests. I know you've tried various things on
> this so maybe we go with this approach for now and see how reliable it
> is over the next few weeks.
>
> It would be great if the pump could close the connections it has
> accumulated in refuserClient so that it doesn't cause problems for
> tests that run subsequently in the same VM.  Hmm, this reminds me that
> will also be connections that are eagerly accepted and they should be
> closed too. It might be simpler to have the test run in othervm mode.
> One other thing about refuserClients is that it could be local to the
> pump task, it doesn't need to be a static.
>
> A couple of minor comments:
>
> At L714 then pumperExecutor cannot be null. Same thing at L723 with
> pumpFuture.
>
> For consistently I would suggest renaming donePump to pumpDone.
>
> Otherwise thanks again for spending time on this awkward test.
>
> -Alan.


More information about the nio-dev mailing list