[
https://issues.apache.org/jira/browse/HIVE-12391?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jesus Camacho Rodriguez updated HIVE-12391:
-------------------------------------------
Description:
SkewJoinOptimizer will not kick in if the columns are just renamed after the TS
e.g. by the creation of a derived table.
To reproduce, consider the following example:
{code}
set hive.optimize.skewjoin.compiletime = true;
CREATE TABLE T1(key STRING, val STRING)
SKEWED BY (key) ON ((2)) STORED AS TEXTFILE;
CREATE TABLE T2(key STRING, val STRING)
SKEWED BY (key) ON ((3)) STORED AS TEXTFILE;
{code}
For this query, SkewJoinOptimizer kicks in:
{code}
SELECT a.*, b.*
FROM T1 a JOIN T2 b
ON a.key = b.key
{code}
For this one, it does not:
{code}
SELECT a.*, b.*
FROM
(SELECT key as k, val as v FROM T1) a
JOIN
(SELECT key as k, val as v FROM T2) b
ON a.k = b.k;
{code}
The reason is that SkewJoinOptimizer does not backtrack the origin of the
column. Instead it just uses its name to know if it is produced by a certain TS.
> SkewJoinOptimizer might not kick in if columns are renamed after TS
> -------------------------------------------------------------------
>
> Key: HIVE-12391
> URL: https://issues.apache.org/jira/browse/HIVE-12391
> Project: Hive
> Issue Type: Bug
> Components: Logical Optimizer
> Affects Versions: 1.3.0, 2.0.0
> Reporter: Jesus Camacho Rodriguez
> Assignee: Jesus Camacho Rodriguez
>
> SkewJoinOptimizer will not kick in if the columns are just renamed after the
> TS e.g. by the creation of a derived table.
> To reproduce, consider the following example:
> {code}
> set hive.optimize.skewjoin.compiletime = true;
> CREATE TABLE T1(key STRING, val STRING)
> SKEWED BY (key) ON ((2)) STORED AS TEXTFILE;
> CREATE TABLE T2(key STRING, val STRING)
> SKEWED BY (key) ON ((3)) STORED AS TEXTFILE;
> {code}
> For this query, SkewJoinOptimizer kicks in:
> {code}
> SELECT a.*, b.*
> FROM T1 a JOIN T2 b
> ON a.key = b.key
> {code}
> For this one, it does not:
> {code}
> SELECT a.*, b.*
> FROM
> (SELECT key as k, val as v FROM T1) a
> JOIN
> (SELECT key as k, val as v FROM T2) b
> ON a.k = b.k;
> {code}
> The reason is that SkewJoinOptimizer does not backtrack the origin of the
> column. Instead it just uses its name to know if it is produced by a certain
> TS.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
