[ 
https://issues.apache.org/jira/browse/HIVE-12227?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Sergey Shelukhin updated HIVE-12227:
------------------------------------
    Attachment: HIVE-12227.patch

Increasing the size of the pool. It's created inline and the vector types are 
only known at processing time and created by VRB utility methods inside 
TreeReader-s, so pre-creating CVBs would be needlessly complicated.

> LLAP: better column vector object pools
> ---------------------------------------
>
>                 Key: HIVE-12227
>                 URL: https://issues.apache.org/jira/browse/HIVE-12227
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Gopal V
>            Assignee: Sergey Shelukhin
>         Attachments: HIVE-12227.patch
>
>
> Vector allocations become a problem in sub-second cases. The pool of 8 per 
> request is too small. Needs to be bigger and potentially pre-populated if we 
> can do it off the main path.



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

Reply via email to