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

Posted to issues@hive.apache.org by "Matt McCline (JIRA)" <ji...@apache.org> on 2018/02/01 14:22:00 UTC

[jira] [Updated] (HIVE-18600) Vectorization: Top-Level Vector Expression Scratch Column Deallocation

     [ https://issues.apache.org/jira/browse/HIVE-18600?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Matt McCline updated HIVE-18600:
--------------------------------
    Status: Patch Available  (was: Open)

> Vectorization: Top-Level Vector Expression Scratch Column Deallocation
> ----------------------------------------------------------------------
>
>                 Key: HIVE-18600
>                 URL: https://issues.apache.org/jira/browse/HIVE-18600
>             Project: Hive
>          Issue Type: Bug
>          Components: Hive
>            Reporter: Matt McCline
>            Assignee: Matt McCline
>            Priority: Critical
>             Fix For: 3.0.0
>
>         Attachments: HIVE-18600.01.patch
>
>
> The operators create various vector expression *arrays* for predicates, SELECT clauses, key expressions, etc.  We could have those be marked as special "top level" vector expression then we could defer deallocation until the top level expression is complete.  This could be a simple solution that avoids trying fix our current eager deallocation that tries to reuse scratch columns as soon as possible.  It *isn't optimal*, but it *shouldn't be too bad*. This solution is much better than not deallocating at all - especially for queries that SELECT a large number of columns or have a lot of expressions in the operator tree.



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