RFR of JDK-8168975: java/rmi/activation/Activatable tests fail due to "Port already in use" in RMID.restart()

[joe darcy]

In terms of timeout, please use much shorter timeout, seconds not 
minutes. Most JDK regression tests complete in well under 10 seconds and 
few tests should run for as long as a minute. Also, please scale any 
timeout if a timeout factor retrieved from the jtreg environment. This 
allows the test to behave better on heavily loaded or slow machines 
where the test invoker has requested more time.

Thanks,

-Joe


On 11/16/2016 9:01 AM, Chris Hegarty wrote:
>> On 16 Nov 2016, at 07:36, Hamlin Li <huaming.li at oracle.com> wrote:
>>
>> Would you please review below fix?
>>
>> bug: https://bugs.openjdk.java.net/browse/JDK-8168975
>> webrev: http://cr.openjdk.java.net/~mli/8168975/webrev.00/
> The approach builds on the mechanism put in for 8168975, and seems
> reasonable.
>
> The system property being used to “pass” the port number is specific to
> the test, and not JDK related so should not be named
> java.nio.channels.spi.SelectorProviderPort.  Maybe something like
> openjdk.test.RMIDSelectorProvider.port ?
>
> -Chris.
>
>> Root Cause: There is a time window between RMID.start() and RMID.restart(), interloper can step in and bind to the port used by RMID in RMID.start().
>>
>> Solution:   Modify RMID.java to use inherited channel when calling RMID.restart. And extend RMIDSelectorProvider.java to try to bind to specific port, no matter 0 or none-zero(for RMID.restart) in a while loop, if timeout just call System.exit to exit rmid and let test fail.
>>
>>
>> Thank you
>> -Hamlin