Re: Collision SPI Not Adhering to Specification

2021-07-01 Thread Atri Sharma
I have opened a JIRA for the same:

https://issues.apache.org/jira/browse/IGNITE-15043

Unless objections, I plan on sketching an implementation plan.

On Thu, Jul 1, 2021 at 1:07 AM Atri Sharma  wrote:
>
> Hi All,
>
> I have been playing around with Collision SPI and specifically used
> FifoQueueCollisionSPI and noticed that it is not really adhering to
> the specified task of restricting the number of concurrent tasks that
> can be run.
>
> Specifically, if there is only one slot available and N tasks
> concurrently land onto the node, all nodes will take the current
> active count, compare it with maximum jobs count and proceed.
>
> Essentially, we have no concurrency safety there.
>
> I propose refactoring FifoQueueCollisionSPI to use semaphores and/or
> atomic variables to ensure that all jobs get a realistic view of the
> active count.
>
> Please share your thoughts,
>
> Regards,
>
> Atri
>
> --
> Regards,
>
> Atri
> Apache Concerted



-- 
Regards,

Atri
Apache Concerted


Collision SPI Not Adhering to Specification

2021-06-30 Thread Atri Sharma
Hi All,

I have been playing around with Collision SPI and specifically used
FifoQueueCollisionSPI and noticed that it is not really adhering to
the specified task of restricting the number of concurrent tasks that
can be run.

Specifically, if there is only one slot available and N tasks
concurrently land onto the node, all nodes will take the current
active count, compare it with maximum jobs count and proceed.

Essentially, we have no concurrency safety there.

I propose refactoring FifoQueueCollisionSPI to use semaphores and/or
atomic variables to ensure that all jobs get a realistic view of the
active count.

Please share your thoughts,

Regards,

Atri

-- 
Regards,

Atri
Apache Concerted