[
https://issues.apache.org/jira/browse/HIVE-10596?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14938987#comment-14938987
]
Ashutosh Chauhan commented on HIVE-10596:
-----------------------------------------
We can go farther than that by not launching job at all. Since query result can
entirely be computed without launching a job, once we obtain result set we can
attach that to 1) TS operator for next job or 2) to fetchtask if there are no
other jobs.
StatsOptimizer already does 2)
> Further optimize metadata only query
> -------------------------------------
>
> Key: HIVE-10596
> URL: https://issues.apache.org/jira/browse/HIVE-10596
> Project: Hive
> Issue Type: Improvement
> Components: Query Planning
> Reporter: Selina Zhang
> Assignee: Selina Zhang
>
> Currently with NullScan optimizer we skip the full table scan for metadata
> only query. However, further step can be taken: writing metadata info to a
> HDFS temp file and replace the original table info with the temp table info.
> In this way we will not calculate the mapper number according the original
> files. Users like to run metadata only query for facts table; calculates
> splits and allocates resources for this type of query seems unnecessary.
> We had this patch and it has been running for a while. Will upload it soon.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
