flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-3697) keyBy() with nested POJO computes invalid field position indexes
Date Tue, 05 Apr 2016 12:51:26 GMT

    [ https://issues.apache.org/jira/browse/FLINK-3697?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15226186#comment-15226186
] 

ASF GitHub Bot commented on FLINK-3697:
---------------------------------------

GitHub user rmetzger opened a pull request:

    https://github.com/apache/flink/pull/1851

    [FLINK-3697] Properly access type information for nested POJO key selection

    

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

    $ git pull https://github.com/rmetzger/flink flink3697

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

    https://github.com/apache/flink/pull/1851.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 #1851
    
----
commit 614e59f619ed64e20c7970640e8a22f7883a50af
Author: Robert Metzger <rmetzger@apache.org>
Date:   2016-04-05T12:48:40Z

    [FLINK-3697] Properly access type information for nested POJO key selection

----


> keyBy() with nested POJO computes invalid field position indexes
> ----------------------------------------------------------------
>
>                 Key: FLINK-3697
>                 URL: https://issues.apache.org/jira/browse/FLINK-3697
>             Project: Flink
>          Issue Type: Bug
>          Components: DataStream API
>    Affects Versions: 1.0.0
>         Environment: MacOS X 10.10
>            Reporter: Ron Crocker
>            Priority: Critical
>              Labels: pojo
>
> Using named keys in keyBy() for nested POJO types results in failure. The iindexes for
named key fields are used inconsistently with nested POJO types. In particular, {{PojoTypeInfo.getFlatFields()}}
returns the field's position after (apparently) flattening the structure but is referenced
in the unflattened version of the POJO type by {{PojoTypeInfo.getTypeAt()}}.
> In the example below, getFlatFields() returns positions 0, 1, and 14. These positions
appear correct in the flattened structure of the Data class. However, in {{KeySelector<X,
Tuple> getSelectorForKeys(Keys<X> keys, TypeInformation<X> typeInfo, ExecutionConfig
executionConfig)}}, a call to {{compositeType.getTypeAt(logicalKeyPositions[i])}} for the
third key results {{PojoTypeInfo.getTypeAt()}} declaring it out of range, as it compares the
length of the directly named fields of the object vs the length of flattened version of that
type.
> Concrete Example:
> Consider this graph:
> {code}
> DataStream<TimesliceData> dataStream = see.addSource(new FlinkKafkaConsumer08<>(timesliceConstants.topic,
new DataDeserialzer(), kafkaConsumerProperties));
> dataStream
>       .flatMap(new DataMapper())
>       .keyBy("aaa", "abc", "wxyz")
> {code}
> {{DataDeserialzer}} returns a "NativeDataFormat" object; {{DataMapper}} takes this NativeDataFormat
object and extracts individual Data objects: {code}
> public class Data {
>     public int aaa;
>     public int abc;
>     public long wxyz;
>     public int t1;
>     public int t2;
>     public Policy policy;
>     public Stats stats;
>     public Data() {}
> {code}
> A {{Policy}} object is an instance of this class:
> {code}
> public class Policy {
>     public short a;
>     public short b;
>     public boolean c;
>     public boolean d;
>     public Policy() {}
> }
> {code}
> A {{Stats}} object is an instance of this class:
> {code}
> public class Stats {
>     public long count;
>     public float a;
>     public float b;
>     public float c;
>     public float d;
>     public float e;
>     public Stats() {}
> }
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message