You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Sergey Shelukhin (JIRA)" <ji...@apache.org> on 2018/04/03 23:14:00 UTC

[jira] [Comment Edited] (HIVE-17970) MM LOAD DATA with OVERWRITE doesn't use base_n directory concept

    [ https://issues.apache.org/jira/browse/HIVE-17970?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424726#comment-16424726 ] 

Sergey Shelukhin edited comment on HIVE-17970 at 4/3/18 11:13 PM:
------------------------------------------------------------------

[~ekoifman] can you take a look? thanks

The test results didn't change; the old approach removed the delta directories not matching the current one, I removed that and it broke the test; handling overwrite properly with base fixed the test.


was (Author: sershe):
[~ekoifman] can you take a look? thanks

> MM LOAD DATA with OVERWRITE doesn't use base_n directory concept
> ----------------------------------------------------------------
>
>                 Key: HIVE-17970
>                 URL: https://issues.apache.org/jira/browse/HIVE-17970
>             Project: Hive
>          Issue Type: Sub-task
>          Components: Transactions
>    Affects Versions: 3.0.0
>            Reporter: Eugene Koifman
>            Assignee: Sergey Shelukhin
>            Priority: Major
>              Labels: mm-gap-2
>         Attachments: HIVE-17970.patch
>
>
> Judging by 
> {code:java}
> Hive.loadTable(Path loadPath, String tableName, LoadFileType loadFileType, boolean isSrcLocal,
>       boolean isSkewedStoreAsSubdir, boolean isAcid, boolean hasFollowingStatsTask,
>       Long txnId, int stmtId, boolean isMmTable)
> {code}
> LOAD DATA with OVERWRITE will delete all existing data then write new data into the table.  This logic makes sense for non-acid tables but for Acid/MM it should work like INSERT OVERWRITE statement and write new data to base_n/. This way the lock manager can be used to either get an X lock for IOW and thus block all readers or let it run with SemiShared and let readers continue and make the system more concurrent.



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