CODEWITHSUNDEEP
scalaapache-sparkrdd
LearningBigData
LearningBigData

Reputation: 13

How do I join two rdds based on a common field?

I am very new to Scala and learning to work with RDDs. I have two csv files which have the following headers and data: csv1.txt

id,"location", "zipcode" 
1, "a", "12345"
2, "b", "67890"
3, "c" "54321"

csv2.txt

"location_x", "location_y", trip_hrs
"a", "b", 1
"a", "c", 3
"b", "c", 2
"a", "b", 1
"c", "b", 2

Basically, csv1 data is a distinct set of locations and zip codes, whereas csv2 data has the trip duration between location_x and location_y.

The common piece of information in these two data sets is location in csv1 and location_x in csv2 even though they have different header names.

I would like to create two RDDs with one containing the data from csv1 and the other from csv2.

Then I would like to join these two RDDs and return the location, zipcode, and sum of all trip times from that location as shown below:

("a", "zipcode", 5)
("b", "zipcode", 2)
("c", "zipcode", 2)

I was wondering if one of you can assist me with this problem. Thanks.

Upvotes: 1

Views: 592

Answers (2)

Chema
Chema

Reputation: 2828

I will give you the code (a complete app in IntelliJ) with some explanations. I hope it can be helpful.

Please read the Spark documentation for the explicit details.

working-with-key-value-pairs

This problem can be done with Spark Dataframes, you can try for yourself.

import org.apache.log4j.{Level, Logger}
import org.apache.spark.sql.SparkSession

object Joining {

  val spark = SparkSession
    .builder()
    .appName("Joining")
    .master("local[*]")
    .config("spark.sql.shuffle.partitions", "4") //Change to a more reasonable default number of partitions for our data
    .config("spark.app.id", "Joining")  // To silence Metrics warning
    .getOrCreate()

  val sc = spark.sparkContext

  val path = "/home/cloudera/files/tests/"

  def main(args: Array[String]): Unit = {

    Logger.getRootLogger.setLevel(Level.ERROR)

    try {

      // read the files
      val file1 = sc.textFile(s"${path}join1.csv")
      val header1 = file1.first // extract the header of the file
      val file2 = sc.textFile(s"${path}join2.csv")
      val header2 = file2.first // extract the header of the file

      val rdd1 = file1
        .filter(line => line != header1) // to leave out the header
        .map(line => line.split(",")) // split the lines => Array[String]
        .map(arr => (arr(1).trim,arr(2).trim)) // to make up a pairRDD with arr(1) as key and zipcode

      val rdd2 = file2
          .filter(line => line != header2)
          .map(line => line.split(",")) // split the lines => Array[String]
          .map(arr => (arr(0).trim, arr(2).trim.toInt)) // to make up a pairRDD with arr(0) as key and trip_hrs

      val joined = rdd1 // join the pairRDD by its keys
          .join(rdd2)
          .cache()  // cache joined in memory

      joined.foreach(println) // checking data
      println("**************")

//      ("c",("54321",2))
//      ("b",("67890",2))
//      ("a",("12345",1))
//      ("a",("12345",3))
//      ("a",("12345",1))

      val result = joined.reduceByKey({ case((zip, time), (zip1, time1) ) => (zip, time + time1) })

      result.map({case( (id,(zip,time)) ) => (id, zip, time)}).foreach(println) // checking output

//      ("b","67890",2)
//      ("c","54321",2)
//      ("a","12345",5)

      // To have the opportunity to view the web console of Spark: http://localhost:4041/
      println("Type whatever to the console to exit......")
      scala.io.StdIn.readLine()
    } finally {
      sc.stop()
      println("SparkContext stopped")
      spark.stop()
      println("SparkSession stopped")
    }
  }
}

Upvotes: 1

pasha701
pasha701

Reputation: 7207

If you can read CSV into RDD already, Trips can be summarized, and then joined with Locations:

val tripsSummarized = trips
  .map({ case (location, _, hours) => (location, hours) })
  .reduceByKey((hoursTotal, hoursIncrement) => hoursTotal + hoursIncrement)

val result = locations
  .map({ case (_, location, zipCode) => (location, zipCode) })
  .join(tripsSummarized)
  .map({case (location, (zipCode, hoursTotal)) => (location, zipCode, hoursTotal) })

If locations without trips are required, "leftOuterJoin" can be used.

Upvotes: 0

Related Questions