[ 
https://issues.apache.org/jira/browse/FLINK-7755?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16210847#comment-16210847
 ] 

ASF GitHub Bot commented on FLINK-7755:
---------------------------------------

GitHub user fhueske opened a pull request:

    https://github.com/apache/flink/pull/4858

    [FLINK-7755] [table] Fix NULL handling in batch joins.

    ## What is the purpose of the change
    
    This PR fixes a couple of issues with Table API / SQL batch joins:
    - Proper support for joining null values for inner and outer joins
    - Support for non-equi join predicates in outer joins (at least one 
equi-join predicate is required)
    - Support for local predicates on the outer input of outer joins (at least 
one equi-join predicate is required)
    
    ## Brief change log
    
    - Inner & Outer Joins: Evaluate all join predicates in a code-gen'd 
function (also equi-join predicates) for correct handling of three-value logic
    - Outer joins: translate outer joins into a sequence of GroupReduce -> 
OuterJoin -> GroupReduce. 
      - The first GroupReduce groups on the full input row and deduplicates the 
outer side(s) of the join. A count for the number of deduplicated rows is kept.
      - The OuterJoin evaluates the join predicate and computes possible join 
pairs of left and right rows. The non-outer element of the pair can be null if 
the join predicate does not match.
      - The second GroupReduce groups again on the full input row and computes 
for each outer row the join result. If it was not match with any inner row, it 
produces a null-padded result.
      - The plan for left and right outer joins requires only a single initial 
partitioning and sort of each input. The all operators can reuse the initial 
sort and produce a sorted result again. A full outer join requires an 
additional partitioning and sorting step.
    - Checks for outer join translation are removed to allow outer joins with 
non-equi and local predicates.
    
    ## Verifying this change
    
    - added ITCases for the new outer join features to `JoinITCase`
    - added plan tests for Table API and SQL for the new outer join features
    - updated validation tests
    
    ## Does this pull request potentially affect one of the following parts:
    
      - Dependencies (does it add or upgrade a dependency): **no**
      - The public API, i.e., is any changed class annotated with 
`@Public(Evolving)`: **no**
      - The serializers: **no**
      - The runtime per-record code paths (performance sensitive): **yes**
      - Anything that affects deployment or recovery: JobManager (and its 
components), Checkpointing, Yarn/Mesos, ZooKeeper: **no**
    
    ## Documentation
    
    The documentation does not need to be adjusted because the outer join 
limitation were not documented. 


You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/fhueske/flink tableBatchNullJoin

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/4858.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #4858
    
----
commit 1434d8d7debe207e4b8350199eded4e678885571
Author: Fabian Hueske <fhue...@apache.org>
Date:   2017-10-15T15:55:23Z

    [FLINK-7755] [table] Fix NULL handling in batch joins.
    
    Fixes [FLINK-5498] (Add support for non-equi join and local predicates to 
outer joins) as well.

----


> Null values are not correctly handled by batch inner and outer joins
> --------------------------------------------------------------------
>
>                 Key: FLINK-7755
>                 URL: https://issues.apache.org/jira/browse/FLINK-7755
>             Project: Flink
>          Issue Type: Bug
>          Components: Table API & SQL
>    Affects Versions: 1.4.0, 1.3.2
>            Reporter: Fabian Hueske
>            Assignee: Fabian Hueske
>            Priority: Blocker
>             Fix For: 1.4.0, 1.3.3
>
>
> Join predicates of batch joins are not correctly evaluated according to 
> three-value logic.
> This affects inner as well as outer joins.
> The problem is that some equality predicates are only evaluated by the 
> internal join algorithms of Flink which are based on {{TypeComparator}}. The 
> field {{TypeComparator}} for {{Row}} are implemented such that {{null == 
> null}} results in {{TRUE}} to ensure correct ordering and grouping. However, 
> three-value logic requires that {{null == null}} results to {{UNKNOWN}} (or 
> null). The code generator implements this logic correctly, but for equality 
> predicates, no code is generated.
> For outer joins, the problem is a bit tricker because these do not support 
> code-generated predicates yet (see FLINK-5520). FLINK-5498 proposes a 
> solution for this issue.
> We also need to extend several of the existing tests and add null values to 
> ensure that the join logic is correctly implemented. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to