[
https://issues.apache.org/jira/browse/HIVE-9603?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14310252#comment-14310252
]
Gopal V commented on HIVE-9603:
-------------------------------
[~prasanth_j]: Reading up.
>From a high level, the bursty windowed back-pressure throttling is exactly
>what I was looking for (to run 1 pipeline off a row from any split in the
>group).
The most natural model for the hive operator pipeline is the call-stack
backpressure via processOp() + pushRow() off the inputs. But the same model
with a ring-buffer or an LMAX disruptor is the windowed approach.
Will read the code and see if it actually implements the logical concepts
performantly without object copies or abstract/interface thunks.
> LLAP: (IO) consider using reactive framework to string together parts of code
> -----------------------------------------------------------------------------
>
> Key: HIVE-9603
> URL: https://issues.apache.org/jira/browse/HIVE-9603
> Project: Hive
> Issue Type: Sub-task
> Reporter: Sergey Shelukhin
>
> Hand-rolled async consumers-producers are somewhat confusing. [~prasanth_j]
> proposed using RxJava; Observable seems to be a natural fit for data producer
> and provides built-in error handling and backpressure.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
