使用Spark Scala检查一个数据框列中的值是否在另一数据框列中存在

Question

I have 2 dataframes df1 and df2,

df1 has column Name with values like a,b,c etc df2 has column Id with values like a,b

If Name column in df1 has a match in Id column in df2 , then we need to have match status as 0. If there is no match then we need to have match status as 1. I know that I can put df2 ID column in a collection using collect and then check if Name column in df1 has matching entry.

val df1 = Seq(“Rey”, “John”).toDF(“Name”)
val df2 = Seq(“Rey”).toDF(“Id”)

val collect = df2.select("Id").map(r => r.getString(0)).collect.toList 

something like,

    val df3 = 
    df1.withColumn("match_sts",when(df1("Name").isin(collect).then(0).else(1)))

Expected output
+ — — + — -+
|Name|match_sts|
+ — — + — -+
| Rey| 0  |
|John| 1  |
+ — — + — -+

But I don't want to use collect here. Is there any alternate approach available.

Answer 1

With collect is not what you want, but is a well -known issue for DF col --> list conversion. If not a huge list, then you can do - this works actually, you can also broadcast the inlist:

import org.apache.spark.sql.functions._

val df1 = Seq("Rey", "John", "Donald", "Trump").toDF("Name")
val df2 = Seq("Rey", "Donald").toDF("Id")

val inlist = df2.select("Id").map(r => r.getString(0)).collect.toList

val df3 = df1.withColumn("match_status", when(df1("Name").isin(inlist: _*),1).otherwise(0))
df3.show(false)

Even in the classical examples that use the stopwords from a file for filtering output, they do this:

val stopWords = stopWordsInput.flatMap(x => x.split(" ")).map(_.trim).collect.toSet

and broadcast if too big to the Workers. But I am not sure what 1 lakh is!!!

Another approach is with Spark SQL, relying on Catalyst to optimize SQL when EXISTS is used:

import spark.implicits._ 
import org.apache.spark.sql.functions._

val df1 = Seq("Rey", "John", "Donald", "Trump").toDF("Name")
val df2 = Seq("Rey", "Donald").toDF("Id") // This can be read from file and split etc.

// Optimizer converts to better physical plan for performance in general
df1.createOrReplaceTempView("searchlist") 
df2.createOrReplaceTempView("inlist")    
val df3 = spark.sql("""SELECT Name, 1 
                     FROM searchlist A
                    WHERE EXISTS (select B.Id from inlist B WHERE B.Id = A.Name )
                                   UNION
                   SELECT Name, 0 
                     FROM searchlist A
                    WHERE NOT EXISTS (select B.Id from inlist B WHERE B.Id = A.Name )
                """)
df3.show(false)