You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Eugene Koifman (JIRA)" <ji...@apache.org> on 2018/06/05 16:43:00 UTC
[jira] [Commented] (HIVE-19793) disable LLAP IO batch-to-row
wrapper for ACID deletes/updates
[ https://issues.apache.org/jira/browse/HIVE-19793?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16502077#comment-16502077 ]
Eugene Koifman commented on HIVE-19793:
---------------------------------------
I looked and I can see that it does what the Jira title says but I don't understand the implications beyond that. If you can just disable batch to row converter, what is it for?
What will happen if some query is run with vectorization off?
Perhaps [~mmccline] or [~teddy.choi] could comment
> disable LLAP IO batch-to-row wrapper for ACID deletes/updates
> -------------------------------------------------------------
>
> Key: HIVE-19793
> URL: https://issues.apache.org/jira/browse/HIVE-19793
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Priority: Major
> Attachments: HIVE-19793.patch
>
>
> 1) Batch to row converter doesn't propagate columns correctly because they are not in the schema.
> 2) Then, even if it did, the current VrbCtx model of ACID column propagation only works with VectorMapOperator. Regular MapOperator has no such context; the reader ends up storing the vector in some fake temporary ctx. I left a TODO that combined with a fix to (1) could fix this instead of disabling it.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)