RFR: 8339538: Wrong timeout computations in DnsClient [v2]

[Aleksei Efimov]

On Mon, 9 Sep 2024 15:03:46 GMT, Daniel Fuchs <dfuchs at openjdk.org> wrote:

>> Sounds like a right thing to do:  measuring time in the loop should give us better estimation on time DNS client spends waiting on the response after submiting a query (that's how environment property value is defined in [javadoc here](https://docs.oracle.com/en/java/javase/22/docs/api/jdk.naming.dns/module-summary.html)).
>> I've tried to move `start` and `end` like:
>> 
>>                  do {
>> +                    long start = System.nanoTime();
>> <....>
>> -                    long start = System.nanoTime();
>>                      gotData = blockingReceive(udpChannel, ipkt, timeoutLeft);
>> -                    long end = System.nanoTime();
>> <...>
>> +                    long end = System.nanoTime();
>>                      long elapsedMillis = TimeUnit.NANOSECONDS.toMillis(end - start);
>> 
>> As a result the tests showed that the timeout happened with a bit better precision (10th of milliseconds).
>> I will run more tests and incorporate the suggestions. Thank you.
>
> I have been wondering why the timeout was measured in this way. My explanation was that the original author of the API (duke? ;-) ) wanted to measure "actual timeout" (time actually spent waiting for a packet) as opposed to perceived timeout (time the caller spent waiting). Rationale might have been to exclude potential time spent waiting for a lock before entering receive() - for instance. That said - I'm not opposed to extend the scope of the timeout to make it closer to the perceived timeout, which is what the test expect. 
> If we do this maybe we could move `start` out of the `do` block  - which would make the computation of `timeoutLeft` simpler (we wouldn't need the `-=`).

I think that "the perceived timeout" described by you above is closer to the following description of the timeout environment property:
```The provider submits a query to a DNS server and waits for a response to arrive within a timeout period (1 second by default)```
And the query is submitted to a DNS server right before entrering the while loop.
I will apply the suggested simplification of the timeout left now, thanks.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/20892#discussion_r1750628602