hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eugene Koifman (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HIVE-16722) Converting bucketed non-acid table to acid should perform validation
Date Fri, 13 Oct 2017 19:38:00 GMT

    [ https://issues.apache.org/jira/browse/HIVE-16722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16204067#comment-16204067
] 

Eugene Koifman edited comment on HIVE-16722 at 10/13/17 7:37 PM:
-----------------------------------------------------------------

patch 1 will validate the names of the files when transactional=true is set for existing table.
The work in HIVE-17204 ensures that non-acid to acid conversion can handle original files
(_OrcRawRecordMerger.OriginalReaderPair_)  which can be in subdirectories of table/partition.


was (Author: ekoifman):
patch 1 will validate the names of the files when transactional=true is set for existing table.
The work in HIVE-17204 ensures that non-acid to acid conversion can handle original files
(_OrcRawRecordMerger_)  which can be in subdirectories of table/partition.

> Converting bucketed non-acid table to acid should perform validation
> --------------------------------------------------------------------
>
>                 Key: HIVE-16722
>                 URL: https://issues.apache.org/jira/browse/HIVE-16722
>             Project: Hive
>          Issue Type: Sub-task
>          Components: Transactions
>    Affects Versions: 1.0.0
>            Reporter: Eugene Koifman
>            Assignee: Eugene Koifman
>         Attachments: HIVE-16722.01.patch, HIVE-16722.WIP.patch
>
>
> Converting a non acid table to acid only performs metadata validation (in _TransactionalValidationListener_).
> The data read code path only understands certain directory layouts and file names and
ignores (generally) files that don't match the expected format.
> In Hive, directory layout and bucket file naming (especially older releases) is poorly
enforced.
> Need to add a validation step on 
> {noformat}
> alter table T SET TBLPROPERTIES ('transactional'='true')
> {noformat}
> to 
> scan the file system and report any possible data loss scenarios.
> Currently Acid understands bucket files name like "00000_0" and (with HIVE-16177) 00000_0_copy1"
etc at the root of the partition.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message