Merge two spark sql columns of type Array[string] into a new Array[string] column

Question

I have two columns in a Spark SQL DataFrame with each entry in either column as an array of strings.

val  ngramDataFrame = Seq(
  (Seq("curious", "bought", "20"), Seq("iwa", "was", "asj"))
).toDF("filtered_words", "ngrams_array")

I want to merge the arrays in each row to make a single array in a new column. My code is as follows:

def concat_array(firstarray: Array[String], 
                 secondarray: Array[String]) : Array[String] = 
                                     { (firstarray ++ secondarray).toArray }
val concatUDF = udf(concat_array _)
val concatFrame = ngramDataFrame.withColumn("full_array", concatUDF($"filtered_words", $"ngrams_array"))

I can successfully use the concat_array function on two arrays. However when I run the above code, I get the following exception:

org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 16.0 failed 1 times, most recent failure: Lost task 0.0 in stage 16.0 (TID 12, localhost): org.apache.spark.SparkException: Failed to execute user defined function(anonfun$1: (array, array) => array) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source) at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43) at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$8$$anon$1.hasNext(WholeStageCodegenExec.scala:370) at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:389) at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408) at org.apache.spark.shuffle.sort.BypassMergeSortShuffleWriter.write(BypassMergeSortShuffleWriter.java:125) at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:79) at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:47) at org.apache.spark.scheduler.Task.run(Task.scala:86) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:274) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748) Caused by: java.lang.ClassCastException: scala.collection.mutable.WrappedArray$ofRef cannot be cast to [Ljava.lang.String; at $line80.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$anonfun$1.apply(:76) ... 13 more Driver stacktrace:

Answer 1

Arjun there is an error in the udf you had created.when you are passing the array type columns .data type is not Array[String] it is WrappedArray[String].below i am pasting the modified udf along with output.

val SparkCtxt = new SparkContext(sparkConf)

val sqlContext = new SQLContext(SparkCtxt)

import sqlContext.implicits

import org.apache.spark.sql.functions._
val temp=SparkCtxt.parallelize(Seq(Row(Array("String1","String2"),Array("String3","String4"))))
val df= sqlContext.createDataFrame(temp,
  StructType(List(
    StructField("Col1",ArrayType(StringType),true),
    StructField("Col2",ArrayType(StringType),true)
  )
  )    )

def concat_array(firstarray: mutable.WrappedArray[String],
                 secondarray: mutable.WrappedArray[String]) : mutable.WrappedArray[String] =
{
 (firstarray ++ secondarray)
}
val concatUDF = udf(concat_array _)
val df2=df.withColumn("udftest",concatUDF(df.col("Col1"), df.col("Col2")))
df2.select("udftest").foreach(each=>{println("***********")
println(each(0))})
df2.show(true)

OUTPUT:

+------------------+------------------+--------------------+
|              Col1|              Col2|             udftest|
+------------------+------------------+--------------------+
|[String1, String2]|[String3, String4]|[String1, String2...|
+------------------+------------------+--------------------+

WrappedArray(String1, String2, String3, String4)

Answer 2

In Spark 2.4 or later you can use concat (if you want to keep duplicates):

ngramDataFrame.withColumn(
  "full_array", concat($"filtered_words", $"ngrams_array")
).show

+--------------------+---------------+--------------------+
|      filtered_words|   ngrams_array|          full_array|
+--------------------+---------------+--------------------+
|[curious, bought,...|[iwa, was, asj]|[curious, bought,...|
+--------------------+---------------+--------------------+

or array_union (if you want to drop duplicates):

ngramDataFrame.withColumn(
  "full_array",
   array_union($"filtered_words", $"ngrams_array")
)

These can be also composed from the other higher order functions, for example

ngramDataFrame.withColumn(
   "full_array",
   flatten(array($"filtered_words", $"ngrams_array"))
)

with duplicates, and

ngramDataFrame.withColumn(
   "full_array",
   array_distinct(flatten(array($"filtered_words", $"ngrams_array")))
)

without.

On a side note, you shouldn't use WrappedArray when working with ArrayType columns. Instead you should expect the guaranteed interface, which is Seq. So the udf should use function with following signature:

(Seq[String], Seq[String]) => Seq[String]

Please refer to SQL Programming Guide for details.