You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by "Charles Chen (JIRA)" <ji...@apache.org> on 2011/08/03 03:16:27 UTC

[jira] [Assigned] (HIVE-2337) Predicate pushdown erroneously conservative with outer joins

     [ https://issues.apache.org/jira/browse/HIVE-2337?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Charles Chen reassigned HIVE-2337:
----------------------------------

    Assignee: Charles Chen

> Predicate pushdown erroneously conservative with outer joins
> ------------------------------------------------------------
>
>                 Key: HIVE-2337
>                 URL: https://issues.apache.org/jira/browse/HIVE-2337
>             Project: Hive
>          Issue Type: Bug
>          Components: Query Processor
>            Reporter: Charles Chen
>            Assignee: Charles Chen
>
> The predicate pushdown filter is not applying left associativity of joins correctly in determining possible aliases for pushing predicates.
> In hive.ql.ppd.OpProcFactory.JoinPPD.getQualifiedAliases, the criteria for pushing aliases is specified as:
> {noformat}
>     /**
>      * Figures out the aliases for whom it is safe to push predicates based on
>      * ANSI SQL semantics For inner join, all predicates for all aliases can be
>      * pushed For full outer join, none of the predicates can be pushed as that
>      * would limit the number of rows for join For left outer join, all the
>      * predicates on the left side aliases can be pushed up For right outer
>      * join, all the predicates on the right side aliases can be pushed up Joins
>      * chain containing both left and right outer joins are treated as full
>      * outer join. [...]
>      *
>      * @param op
>      *          Join Operator
>      * @param rr
>      *          Row resolver
>      * @return set of qualified aliases
>      */
> {noformat}
> Since hive joins are left associative, something like "a RIGHT OUTER JOIN b LEFT OUTER JOIN cĀ INNER JOIN d" should be interpreted as "((a RIGHT OUTER JOIN b) LEFT OUTER JOIN c) INNER JOIN d", so there would be cases where joins with both left and right outer joins can have aliases that can be pushed.  Here, aliases a and d are eligible to be pushed up while the current criteria provide that none are eligible.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira