[
https://issues.apache.org/jira/browse/SPARK-1239?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Josh Rosen updated SPARK-1239:
------------------------------
Assignee: Josh Rosen (was: Kostas Sakellis)
I'm re-assigning this to me since I've been working in this code recently and
plan to completely re-write MapOutputStatusTracker to address this and a few
other issues.
I like the idea of piggybacking the status on the task launching RPC itself; if
we took the approach of having reducers specify their reduce ids when fetching
the statuses, then this would add an extra RPC for every task launch, which
could result in a big latency increase.
> Don't fetch all map output statuses at each reducer during shuffles
> -------------------------------------------------------------------
>
> Key: SPARK-1239
> URL: https://issues.apache.org/jira/browse/SPARK-1239
> Project: Spark
> Issue Type: Improvement
> Components: Shuffle, Spark Core
> Affects Versions: 1.0.2, 1.1.0
> Reporter: Patrick Wendell
> Assignee: Josh Rosen
>
> Instead we should modify the way we fetch map output statuses to take both a
> mapper and a reducer - or we should just piggyback the statuses on each task.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
