hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sergey Shelukhin (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HIVE-15279) map join dummy operators are not set up correctly in certain cases with merge join
Date Mon, 05 Dec 2016 23:01:00 GMT

    [ https://issues.apache.org/jira/browse/HIVE-15279?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15723650#comment-15723650
] 

Sergey Shelukhin edited comment on HIVE-15279 at 12/5/16 11:00 PM:
-------------------------------------------------------------------

1) As suggested above, we now take dummy ops from all the works, not just the main work. However,
I am not sure if it's possible for the same dummy to be added to multiple works given the
code (someone might have a picture in their head of how they are propagated, but to me it
looks like we just put them in all potentially affected works and hope for the best; hence
all the bugs with them). Therefore, to handle this possibility, we use hashset to make sure
we only init each dummy operator once.
2) The main goal of the log line is to tell whether work has any dummy operators at all. IDs
can be used in conjunction with operator tree that is output elsewhere.
3) Will remove the todo


was (Author: sershe):
1) As suggested above, we now take dummy ops from all the works, not just the main work. However,
I am not sure if it's possible for the same dummy to be added to multiple works given the
code (someone might have a picture in their head of how they are propagated, but to me it
looks like we just put them in all potentially affected works and hope for the best; hence
all the bugs with them). Therefore, to handle this possibility, we use hashset to make sure
we only init each dummy operator once.
2) The main goal of the log line is to tell whether work has any dummy operators at all. IDs
can be used in conjunction with operator tree that is output elsewhere.
3) Will remove.

> map join dummy operators are not set up correctly in certain cases with merge join
> ----------------------------------------------------------------------------------
>
>                 Key: HIVE-15279
>                 URL: https://issues.apache.org/jira/browse/HIVE-15279
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Sergey Shelukhin
>            Assignee: Sergey Shelukhin
>         Attachments: HIVE-15279.01.patch, HIVE-15279.02.patch, HIVE-15279.patch
>
>
> As a result, MapJoin is not initialized and there's NPE later.
> Tez-specific.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message