You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Prasanth Jayachandran (JIRA)" <ji...@apache.org> on 2015/09/03 19:50:46 UTC

[jira] [Commented] (HIVE-10191) ORC: Cleanup writer per-row synchronization

    [ https://issues.apache.org/jira/browse/HIVE-10191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14729463#comment-14729463 ] 

Prasanth Jayachandran commented on HIVE-10191:
----------------------------------------------

[~gopalv] We should document the impact of this change on the hive.exec.orc.memory.pool for multiple threaded writers writing separately to its own orc file. 

> ORC: Cleanup writer per-row synchronization
> -------------------------------------------
>
>                 Key: HIVE-10191
>                 URL: https://issues.apache.org/jira/browse/HIVE-10191
>             Project: Hive
>          Issue Type: Bug
>          Components: File Formats
>    Affects Versions: 1.2.0
>            Reporter: Gopal V
>            Assignee: Gopal V
>              Labels: TODOC1.3
>             Fix For: 1.3.0, 2.0.0
>
>         Attachments: HIVE-10191.1.patch, HIVE-10191.2.patch, HIVE-10191.3.patch, HIVE-10191.4.patch, HIVE-10191.5.patch
>
>
> ORC writers were originally meant to be thread-safe, but in the present day implementation each ORC writer is entirely share-nothing which converts most of the synchronized blocks in ORC as entirely uncontested locks.
> These uncontested locks prevent the JVM from inlining/optimizing these methods, while adding no extra thread-safety to the ORC writers.



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