-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/51962/#review149276
-----------------------------------------------------------


Ship it!




Finally found time to understand what you were changing :) yes. We should not 
block in 2 places for the same cause. +1 !

- Navina Ramesh


On Sept. 16, 2016, 5:56 p.m., Xinyu Liu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/51962/
> -----------------------------------------------------------
> 
> (Updated Sept. 16, 2016, 5:56 p.m.)
> 
> 
> Review request for samza, Boris Shkolnik, Jake Maes, and Yi Pan (Data 
> Infrastructure).
> 
> 
> Bugs: SAMZA-1021
>     https://issues.apache.org/jira/browse/SAMZA-1021
> 
> 
> Repository: samza
> 
> 
> Description
> -------
> 
> The waiting inside AsyncRunLoop is redundent given systemconsumers will block 
> on polling for empty partition. Empty partitions happens when chooser returns 
> null envelope and at the same time some task worker is avaliable to pick up 
> messages. In this case we should rely on the systemconsumers.poll() to block 
> instead of blocking in scheduling.
> 
> The redundent waiting has been seen in some our jobs where both choose_ns and 
> block_ns are high at the same time.
> 
> 
> Diffs
> -----
> 
>   samza-core/src/main/java/org/apache/samza/task/AsyncRunLoop.java 
> a510bb0c5914c772438930d27f100b4d360c1296 
> 
> Diff: https://reviews.apache.org/r/51962/diff/
> 
> 
> Testing
> -------
> 
> Tested by yarn job.
> 
> 
> Thanks,
> 
> Xinyu Liu
> 
>

Reply via email to