[jira] [Commented] (HIVE-20046) remove NUM_FILES check

["Sergey Shelukhin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-20046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16551395#comment-16551395 ] 

Sergey Shelukhin commented on HIVE-20046:
-----------------------------------------

Confirmed by adding a test that this needs to be removed. TRUNCATE sets numFiles to 0, but stats can still be invalid for specific columns.

> remove NUM_FILES check
> ----------------------
>
>                 Key: HIVE-20046
>                 URL: https://issues.apache.org/jira/browse/HIVE-20046
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Sergey Shelukhin
>            Assignee: Sergey Shelukhin
>            Priority: Major
>         Attachments: HIVE-20046.01.patch
>
>
> {noformat}
>     // Since newly initialized empty table has 0 for the parameter.
>     if (Long.parseLong(statsParams.get(StatsSetupConst.NUM_FILES)) == 0) {
>       return true;
>     }
> {noformat}
> This doesn't look safe; # of files could be set to 0 by an invalid update, or potentially a parallel update that we cannot see (not sure if this is possible; there's some code in metastore that updates basic stats outside of the scope of the query).
> It would be better to remove this, and see if it breaks some tests. If we do need this, there should be a negative test at some point



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