[
https://issues.apache.org/jira/browse/SPARK-13333?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15151746#comment-15151746
]
Xiao Li commented on SPARK-13333:
---------------------------------
Yeah, you are right. This part is an issue. That is why I did not remove the
existing rand/randn function with a seed argument. I am unable to find a
solution for controlling data partitioning and task scheduling when users call
rand/randn functions with a specific seed number.
At least, we have to let users realize these results are not deterministic.
Actually, I think this is also an issue in many of our test cases. Many test
cases are using seeds with an assumption that rand/rand(seed) can provide a
deterministic result.
> DataFrame filter + randn + unionAll has bad interaction
> -------------------------------------------------------
>
> Key: SPARK-13333
> URL: https://issues.apache.org/jira/browse/SPARK-13333
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 1.4.2, 1.6.1, 2.0.0
> Reporter: Joseph K. Bradley
>
> Buggy workflow
> * Create a DataFrame df0
> * Filter df0
> * Add a randn column
> * Create a copy of the DataFrame
> * unionAll the two DataFrames
> This fails, where randn produces the same results on the original DataFrame
> and the copy before unionAll but fails to do so after unionAll. Removing the
> filter fixes the problem.
> The bug can be reproduced on master:
> {code}
> import org.apache.spark.sql.functions.randn
> val df0 = sqlContext.createDataFrame(Seq(0, 1).map(Tuple1(_))).toDF("id")
> // Removing the following filter() call makes this give the expected result.
> val df1 = df0.filter(col("id") === 0).withColumn("b", randn(12345))
> println("DF1")
> df1.show()
> val df2 = df1.select("id", "b")
> println("DF2")
> df2.show() // same as df1.show(), as expected
> val df3 = df1.unionAll(df2)
> println("DF3")
> df3.show() // NOT two copies of df1, which is unexpected
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
