[
https://issues.apache.org/jira/browse/HIVE-20279?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16572272#comment-16572272
]
Gopal V edited comment on HIVE-20279 at 8/7/18 8:23 PM:
--------------------------------------------------------
[~ashutoshc]: this patch is probably too fragile - I chased this bug down to
DruidQueryRecordReader not implementing ::getPos().
https://github.com/apache/hive/blob/master/druid-handler/src/java/org/apache/hadoop/hive/druid/serde/DruidQueryRecordReader.java#L145
was (Author: gopalv):
[~ashutoshc]: this patch is probably too fragile - I chased this bug down to
DruidQueryRecordReader not implementing ::getPos().
> HiveContextAwareRecordReader slows down Druid Scan queries.
> ------------------------------------------------------------
>
> Key: HIVE-20279
> URL: https://issues.apache.org/jira/browse/HIVE-20279
> Project: Hive
> Issue Type: Improvement
> Reporter: Nishant Bangarwa
> Assignee: Nishant Bangarwa
> Priority: Major
> Attachments: HIVE-20279.1.patch, HIVE-20279.patch, scan2.svg
>
>
> HiveContextAwareRecordReader add lots of overhead for Druid Scan Queries.
> See attached flame graph.
> Looks like the operations for checking for existence of footer/header buffer
> takes most of time For druid and other storage handlers that do not have
> footer buffer we should skip the logic for checking the existence for storage
> handlers atleast.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
