RFR 8164656: krb5 does not retry if TCP connection timeouts

Weijun Wang weijun.wang at oracle.com
Wed Aug 24 02:00:50 UTC 2016


I've updated the test at

   http://cr.openjdk.java.net/~weijun/8164656/webrev.01/

No more ratio things.

When the ratio was invented for BadKdc, it was because it uses real KDCs 
and we have to find a balance between unexpected timeout and test speed. 
Changing a single ratio is much easier to modify hundreds of timeout and 
pattern across 6 test files.

In KdcPolicy.java, most requests go to fake KDCs and the timeout is 
already super short. For the few cases where a real KDC is needed, even 
if we need to tweak the timeout later (if intermittent failures still 
happen), it will be much easier to just tweak on the actual timeout 
value directly.

Thanks
Max

On 8/24/2016 8:06, Xuelei Fan wrote:
> Looks fine to me.
>
> Xuelei
>
> On 8/24/2016 12:14 AM, Weijun Wang wrote:
>> Please review the change at
>>
>>    http://cr.openjdk.java.net/~weijun/8164656/webrev.00/
>>
>> The old try-catch is only on send/receive, but NetClient.getInstance()
>> could also timeout for TCP when it calls Socket::connect(dest,timeout).
>> Therefore move the catch to a higher level.
>>
>> The test is brand new and it's meant to replace existing BadKdc?.java,
>> TcpTimeout.java and MaxRetries.java. In fact, I discover this bug when I
>> am writing it. The new test is much faster because in most cases there
>> is no real KDC and timeout can be very short. It contains 2 @run and
>> each spends a little more than half a minute to run on my laptop.
>>
>> Thanks
>> Max



More information about the security-dev mailing list