Mailing-List: contact dev-help@hive.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@hive.apache.org
Date: Tue, 29 Jan 2013 19:13:14 +0000 (UTC)
From: "Gang Tim Liu (JIRA)" <jira@apache.org>
To: hive-dev@hadoop.apache.org
Message-ID: <JIRA.12628388.1358561516861.210205.1359486794579@arcas>
In-Reply-To: <JIRA.12628388.1358561516861@arcas>
References: <JIRA.12628388.1358561516861@arcas>
Subject: [jira] [Commented] (HIVE-3917) Support fast operation for analyze
 command
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HIVE-3917?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13565667#comment-13565667 ] 

Gang Tim Liu commented on HIVE-3917:
------------------------------------

[~shreepadma] sorry to not fully understand your initial question.
[~ashutoshc] thank you very much for explaining it in more details and carrying the discussion. great! thanks.

Yes, partial scan is a great choice. Actually, we have thought about it. With https://cwiki.apache.org/Hive/rcfilecat.html, we can even achieve it for RCFile.

Yes, it will be faster than full scan but still slower than noscan. Consider a big data warehouse, partial scan is still magnitude slower than noscan. With potential speedup from simple MR, it will be sill slower than noscan.

Saying that, we can view all 3 as great choices for different use cases: noscan, partial scan and full scan (which is default).

I will create a follow up for partial scan.
                
> Support fast operation for analyze command
> ------------------------------------------
>
>                 Key: HIVE-3917
>                 URL: https://issues.apache.org/jira/browse/HIVE-3917
>             Project: Hive
>          Issue Type: Improvement
>          Components: Statistics
>    Affects Versions: 0.11.0
>            Reporter: Gang Tim Liu
>            Assignee: Gang Tim Liu
>         Attachments: HIVE-3917.patch.1
>
>
> hive supports analyze command to gather statistics from existing tables/partition https://cwiki.apache.org/confluence/display/Hive/StatsDev#StatsDev-ExistingTables
> It collects:
> 1. Number of Rows
> 2. Number of files
> 3. Size in Bytes
> If table/partition is big, the operation would take time since it will open all files and scan all data.
> It would be nice to support fast operation to gather statistics which doesn't require to open all files:
> 1. Number of files
> 2. Size in Bytes
> Potential syntax is 
> ANALYZE TABLE tablename [PARTITION(partcol1[=val1], partcol2[=val2], ...)] COMPUTE STATISTICS [noscan];
> In the future, all statistics without scan can be retrieved via this optional parameter.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira