hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sankar Hariappan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-19750) Initialize NEXT_WRITE_ID. NWI_NEXT on converting an existing table to full acid
Date Fri, 01 Jun 2018 13:44:00 GMT

    [ https://issues.apache.org/jira/browse/HIVE-19750?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16498007#comment-16498007
] 

Sankar Hariappan commented on HIVE-19750:
-----------------------------------------

[~ekoifman],

I just have one comment. "Need to normalize (change to lower case) the db and table name when
insert to NEXT_WRITE_ID. Not sure if we get it in lower case from Table object".

Also, one question,

- What will happen to non-ACID tables which were previously converted to ACID in 3.0.0 and
then upgraded to 3.1.0? It will have original files but the write_id starts with 1.

 

 

> Initialize NEXT_WRITE_ID. NWI_NEXT on converting an existing table to full acid
> -------------------------------------------------------------------------------
>
>                 Key: HIVE-19750
>                 URL: https://issues.apache.org/jira/browse/HIVE-19750
>             Project: Hive
>          Issue Type: Bug
>          Components: Transactions
>    Affects Versions: 3.0.0
>            Reporter: Eugene Koifman
>            Assignee: Eugene Koifman
>            Priority: Critical
>             Fix For: 3.1.0
>
>         Attachments: HIVE-19750.01.patch
>
>
> Need to set this to a reasonably high value the the table.
> This will reserve a range of write IDs that will be treated by the system as committed.
> This is needed so that we can assign unique ROW__IDs to each row in files that already
exist in the table.  For example, if the value is initialized to the number of files currently
in the table, we can think of each file as written by a separate transaction and thus a free
to assign bucketProperty (BucketCodec) of ROW_ID in whichever way is convenient.
> it's guaranteed that all rows get unique ROW_IDs this way.



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

Mime
View raw message