You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@hive.apache.org by "Eric Wohlstadter (JIRA)" <ji...@apache.org> on 2018/05/02 20:44:00 UTC

[jira] [Comment Edited] (HIVE-19306) Arrow batch serializer

    [ https://issues.apache.org/jira/browse/HIVE-19306?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16461599#comment-16461599 ] 

Eric Wohlstadter edited comment on HIVE-19306 at 5/2/18 8:43 PM:
-----------------------------------------------------------------

[~teddy.choi]

I don't think it is necessary or useful to use FieldWriter for non-complex types. 

Should we change it so that non-complex types use "set" and "setNull" directly on the vector?

I believe this would remove some of the indirection which is making the code harder to debug.

If we are very close to making it work with FieldWriter on non-complex types, let's go ahead with that.

I'm just concerned we get stuck trying to fix something that isn't necessary. 


was (Author: ewohlstadter):
[~teddy.choi]

I don't think it is necessary or useful to use FieldWriter for non-complex types. 

Should we change it so that non-complex types use "set" and "setNull" directly on the vector?

I believe this would remove some of the indirection which is making the code harder to debug.

> Arrow batch serializer
> ----------------------
>
>                 Key: HIVE-19306
>                 URL: https://issues.apache.org/jira/browse/HIVE-19306
>             Project: Hive
>          Issue Type: Task
>          Components: Serializers/Deserializers
>            Reporter: Eric Wohlstadter
>            Assignee: Teddy Choi
>            Priority: Major
>         Attachments: HIVE-19306.2.patch
>
>
> Leverage the ThriftJDBCBinarySerDe code path that already exists in SemanticAnalyzer/FileSinkOperator to create a serializer that batches rows into Arrow vector batches.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)