hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Seth (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-12856) LLAP: update (add/remove) the UDFs available in LLAP when they are changed (refresh periodically)
Date Fri, 19 Feb 2016 22:45:18 GMT

    [ https://issues.apache.org/jira/browse/HIVE-12856?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15155045#comment-15155045
] 

Siddharth Seth commented on HIVE-12856:
---------------------------------------

I'm assuming permanent UDFs can only be added by 'admins' when running with HiveServer2. When
this registration happens in HiveServer2 - could we make a call into the LLAP instances (or
register into the ZK registry for the LLAP instance). That can avoid calls out from LLAP to
the metastore. The HiveServer to LLAP / ZK connection would be secure.

> LLAP: update (add/remove) the UDFs available in LLAP when they are changed (refresh periodically)
> -------------------------------------------------------------------------------------------------
>
>                 Key: HIVE-12856
>                 URL: https://issues.apache.org/jira/browse/HIVE-12856
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Sergey Shelukhin
>            Assignee: Sergey Shelukhin
>         Attachments: HIVE-12856.01.nogen.patch, HIVE-12856.01.patch, HIVE-12856.02.nogen.patch,
HIVE-12856.02.patch, HIVE-12856.nogen.patch, HIVE-12856.patch
>
>
> I don't think re-querying the functions is going to scale, and the sessions obviously
cannot notify all LLAP clusters of every change. We should add global versioning to metastore
functions to track changes, and then possibly add a notification mechanism, potentially thru
ZK to avoid overloading the metastore itself.



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

Mime
View raw message