On Dec 8, 2007, at 6:48 AM, Matthew Toseland wrote:
> On Friday 07 December 2007 02:13, you wrote:
>
> With regards to Timer: Why not just use the Ticker ? Admittedly they
> will
> sometimes get scheduled in parallel, but iirc it doesn't do that
> much work
> on-thread so it's not really a problem...
I was not aware of Ticker. It would seem to be more consistent with
the code base to use Ticker.
>>
>> Modified: trunk/freenet/src/freenet/client/async/USKFetcher.java
>> ===================================================================
>> --- trunk/freenet/src/freenet/client/async/USKFetcher.java 2007-12-07
> 01:41:35 UTC (rev 16387)
>> +++ trunk/freenet/src/freenet/client/async/USKFetcher.java 2007-12-07
> 02:13:36 UTC (rev 16388)
>> @@ -252,43 +255,28 @@
>> long now = System.currentTimeMillis();
>> synchronized(this) {
>> started = false; // don't finish before have
>> rescheduled
>> +
>> + //Find out when we should check next ('end'), in an
> increasing delay (unless we make progress).
>> + long newSleepTime = sleepTime * 2;
>> + if(newSleepTime > maxSleepTime) newSleepTime =
>> maxSleepTime;
>> + sleepTime = newSleepTime;
>> + end = now + sleepTime;
>> +
>> if(valAtEnd > valueAtSchedule) {
>> - // Have advanced.
>> + // We have advanced; keep trying as if
>> we just started.
>> minFailures = origMinFailures;
>> sleepTime = origSleepTime;
>> + end = now;
>
> end = now + origSleepTime. What you have there makes it retry
> *immediately*.
Correct. If a USKFetch makes progress it must retry with a distinct
window-of-five, and the original code appeared to do this immediately,
so I maintained that. This code should only be executed when the very
last fetch attempt is complete.
--
Robert Hailey
