[
https://issues.apache.org/jira/browse/SPARK-21402?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16646337#comment-16646337
]
Vladimir Kuriatkov commented on SPARK-21402:
--------------------------------------------
Working on this issue.
The problem can be solved by using unresolved expressions instead of resolved
expressions for ArrayType and MapType when building java bean deserializer.
In this case struct data type of array elements or map's keys/values is defined
during "analysis" and is based on actual data, not on java bean.
The same approach is currently used to define struct data types of nested
structs. In case of array elements or map's keys/values struct data types are
currently extracted from java beans, hence we get incorrect field order.
> Java encoders - switch fields on collectAsList
> ----------------------------------------------
>
> Key: SPARK-21402
> URL: https://issues.apache.org/jira/browse/SPARK-21402
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.1.1
> Environment: mac os
> spark 2.1.1
> Using Scala version 2.11.8, Java HotSpot(TM) 64-Bit Server VM, 1.8.0_121
> Reporter: Tom
> Priority: Major
>
> I have the following schema in a dataset -
> root
> |-- userId: string (nullable = true)
> |-- data: map (nullable = true)
> | |-- key: string
> | |-- value: struct (valueContainsNull = true)
> | | |-- startTime: long (nullable = true)
> | | |-- endTime: long (nullable = true)
> |-- offset: long (nullable = true)
> And I have the following classes (+ setter and getters which I omitted for
> simplicity) -
>
> {code:java}
> public class MyClass {
> private String userId;
> private Map<String, MyDTO> data;
> private Long offset;
> }
> public class MyDTO {
> private long startTime;
> private long endTime;
> }
> {code}
> I collect the result the following way -
> {code:java}
> Encoder<MyClass> myClassEncoder = Encoders.bean(MyClass.class);
> Dataset<MyClass> results = raw_df.as(myClassEncoder);
> List<MyClass> lst = results.collectAsList();
> {code}
>
> I do several calculations to get the result I want and the result is correct
> all through the way before I collect it.
> This is the result for -
> {code:java}
> results.select(results.col("data").getField("2017-07-01").getField("startTime")).show(false);
> {code}
> |data[2017-07-01].startTime|data[2017-07-01].endTime|
> +-----------------------------+--------------+
> |1498854000 |1498870800 |
> This is the result after collecting the reuslts for -
> {code:java}
> MyClass userData = results.collectAsList().get(0);
> MyDTO userDTO = userData.getData().get("2017-07-01");
> System.out.println("userDTO startTime: " + userDTO.getStartTime());
> System.out.println("userDTO endTime: " + userDTO.getEndTime());
> {code}
> --
> data startTime: 1498870800
> data endTime: 1498854000
> I tend to believe it is a spark issue. Would love any suggestions on how to
> bypass it.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
