[concurrency-interest] Fork and and timed get

Romain Colle rco at quartetfs.com
Wed Nov 27 11:09:20 EST 2013


Hi all,

I'd like to second Jonathan's post on this subject.
The basic idea here is that, as far as I can tell, the get(long timeout,
TimeUnit unit) method of a ForkJoinTask does not respect the Future
contract.

Before blocking and waiting for the task to complete, the timed get()
method tries to either steal and execute the task, or at least help it
complete.
If it succeeds doing so, it will for instance execute the task that it is
waiting for. And in this case, there is no telling how long it will take to
execute it ...

Although this is great for the overall throughput, a thread calling the
timed get() method with a timeout of 1 second might end up returning a
minute later (if the task does take a minute to execute).
This is what Jonathan reproduced in its unit test (with a parallelism of 1
in its pool to consistently reproduce this behavior).

This behavior does causes some problems in our code, and to my knowledge
there are no ways of invoking a reliable timed get() on a ForkJoinPool.
Any feedback on this topic would be greatly appreciated!

Thanks
Romain


On Thu, Nov 21, 2013 at 8:35 AM, Martin Buchholz <martinrb at google.com>wrote:

> Your message got ignored...
>
> You created a pool with parallelism of 1, so it's not surprising that no
> new thread was created to run action1.  The action of the pool seems
> reasonable to me - how would you do better?
>
> Also, posting your code with a friendlier legal notice might help.
>
>
> On Mon, Nov 18, 2013 at 8:35 AM, Jonathan Soto <jso at quartetfs.com> wrote:
>
>> Hello everyone,
>> It has been a few days I am thinking about this problem. Could you help
>> me ?
>>
>> Suppose that we have two tasks
>> - task1 waits for 5s
>> - task2 forks task1 and does a timed get of 2.5 seconds
>>
>> The main thread submits task2 to a FJP.
>>
>> I observe the following behavior.
>> - Thread Main   :
>> FPJ.submit(task2)
>>
>> - Thread FJPO-1 : task1.fork
>> As we are in the FJP it adds task1 into the workQueue of the current
>> thread
>> Task1 does not start in another thread
>>
>> - Thread FJPO-1 : task1.get(2.5, SECONDS)
>> Starts by checking if the current thread has some local tasks to be done.
>> As there is task1 that has been previously stored, we execute it.
>>
>> Task1 takes 5 seconds to be executed. The 2.5 seconds timeout is totally
>> ignored.
>>
>> Is that the expected behavior ?
>>
>> Thanks
>>
>> Jonathan
>>
>> P.s in attachment a unit test that prints FAIL when the test fails
>>
>> _______________________________________________
>> Concurrency-interest mailing list
>> Concurrency-interest at cs.oswego.edu
>> http://cs.oswego.edu/mailman/listinfo/concurrency-interest
>>
>>
>
> _______________________________________________
> Concurrency-interest mailing list
> Concurrency-interest at cs.oswego.edu
> http://cs.oswego.edu/mailman/listinfo/concurrency-interest
>
>


-- 
Romain Colle
R&D Project Manager
QuartetFS
2 rue Jean Lantier, 75001 Paris, France
http://www.quartetfs.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.oswego.edu/pipermail/concurrency-interest/attachments/20131127/6f3bed89/attachment.html>


More information about the Concurrency-interest mailing list