[
https://issues.apache.org/jira/browse/SPARK-27582?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stanislav Bytsko updated SPARK-27582:
-------------------------------------
Description:
Currently we have
{code:java}
org.apache.spark.sql.Dataset[T]#joinWith[U](other: Dataset[U], condition:
Column, joinType: String): Dataset[(T, U)]
{code}
which explicitly excludes left_anti and left_semi joins, which is
understandable, because result would have different signature.
I think it's easily fixed drawback, which accepts 2 solutions I can think of:
- Extend current joinWith to return null for second (_2) item in the tuple. Not
ideal as no-one likes nulls, but workable, as client should be able to handle
that by doing {code}.map(_._1){code} immediately afterwards
- Add 2 new methods
{code}org.apache.spark.sql.Dataset[T]#joinSemiWith[U](other: Dataset[U],
condition: Column): Dataset[T]{code} and
{code}org.apache.spark.sql.Dataset[T]#joinAntiWith[U](other: Dataset[U],
condition: Column): Dataset[T]{code} which is much nicer, but adds 2 methods to
the API. names could be semiJoinWith and antiJoinWith, which is more logical,
but not sorted properly in the list of org.apache.spark.sql.Dataset methods
was:
Currently we have
{code:java}
org.apache.spark.sql.Dataset[T]#joinWith[U](other: Dataset[U], condition:
Column, joinType: String): Dataset[(T, U)]
{code}
which explicitly excludes left_anti and left_semi joins, which is
understandable, because result would have different signature.
I think it's easily fixed drawback, which accepts 2 solutions I can think of:
- Extend current joinWith to return null for second (_2) item in the tuple. Not
ideal as no-one likes nulls, but workable, as client should be able to handle
that by doing {code}.map(_._1){code} immediately afterwards
- Add 2 new methods
{code}org.apache.spark.sql.Dataset[T]#joinSemiWith[U](other: Dataset[U],
condition: Column): Dataset[T]{code} and
{code}org.apache.spark.sql.Dataset[T]#joinAntiWith[U](other: Dataset[U],
condition: Column): Dataset[T]{code} which is much nicer, but adds 2 methods to
the API.
> Add Dataset DSL for left_anti and left_semi joins
> -------------------------------------------------
>
> Key: SPARK-27582
> URL: https://issues.apache.org/jira/browse/SPARK-27582
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 2.4.2
> Reporter: Stanislav Bytsko
> Priority: Major
>
> Currently we have
> {code:java}
> org.apache.spark.sql.Dataset[T]#joinWith[U](other: Dataset[U], condition:
> Column, joinType: String): Dataset[(T, U)]
> {code}
> which explicitly excludes left_anti and left_semi joins, which is
> understandable, because result would have different signature.
> I think it's easily fixed drawback, which accepts 2 solutions I can think of:
> - Extend current joinWith to return null for second (_2) item in the tuple.
> Not ideal as no-one likes nulls, but workable, as client should be able to
> handle that by doing {code}.map(_._1){code} immediately afterwards
> - Add 2 new methods
> {code}org.apache.spark.sql.Dataset[T]#joinSemiWith[U](other: Dataset[U],
> condition: Column): Dataset[T]{code} and
> {code}org.apache.spark.sql.Dataset[T]#joinAntiWith[U](other: Dataset[U],
> condition: Column): Dataset[T]{code} which is much nicer, but adds 2 methods
> to the API. names could be semiJoinWith and antiJoinWith, which is more
> logical, but not sorted properly in the list of org.apache.spark.sql.Dataset
> methods
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
