[
https://issues.apache.org/jira/browse/HIVE-8042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14129001#comment-14129001
]
Gunther Hagleitner commented on HIVE-8042:
------------------------------------------
[~xuefuz] I'm torn about this one. I don't like the large number of configs we
have, but this is changing the behavior of hive.exec.parallel and I wanted to
make sure folks can fall back to the old behavior. Bug defense and maybe
situations where you don't want to do that much work in parallel on the client
machine (e.g.: HS2).
> Optionally allow move tasks to run in parallel
> ----------------------------------------------
>
> Key: HIVE-8042
> URL: https://issues.apache.org/jira/browse/HIVE-8042
> Project: Hive
> Issue Type: Bug
> Reporter: Gunther Hagleitner
> Assignee: Gunther Hagleitner
> Attachments: HIVE-8042.1.patch
>
>
> hive.exec.parallel allows one to run different stages of a query in parallel.
> However that applies only to map-reduce tasks. When using large multi insert
> queries there are many MoveTasks that are all executed in sequence on the
> client. There's no real reason for that - they could be run in parallel as
> well (i.e.: the stage graph captures the dependencies and knows which tasks
> can happen in parallel)
> Proposal here is to create a new config var:
> hive.exec.parallel.movetask
> When set to true it run movetasks in parallel also.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
