[
https://issues.apache.org/jira/browse/HIVE-9561?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rui Li updated HIVE-9561:
-------------------------
Attachment: HIVE-9561.1-spark.patch
The patch can only partially solve the problem. For some cases, we still can
use SHUFFLE_SORT when we don't have to. For example:
{code}
explain select * from (select pagerank from tiny sort by pagerank) t join
(select * from skewed order by pagerank) s on t.pagerank=s.pagerank;
{code}
We have a sort-by and an order-by. However, the patch can't distinguish these
two and we'll use SHUFFLE_SORT for both of them.
[~xuefuz] any ideas?
> SHUFFLE_SORT should only be used for order by query [Spark Branch]
> ------------------------------------------------------------------
>
> Key: HIVE-9561
> URL: https://issues.apache.org/jira/browse/HIVE-9561
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Rui Li
> Assignee: Rui Li
> Attachments: HIVE-9561.1-spark.patch
>
>
> The {{sortByKey}} shuffle launches probe jobs. Such jobs can hurt performance
> and are difficult to control. So we should limit the use of {{sortByKey}} to
> order by query only.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
