Spark-Scala recipes

Data Science Studio gives you the ability to write Spark recipes using Scala, Spark’s native language. Spark-Scala recipes can read and write datasets, even when their storage backend is not HDFS.

Spark-scala recipes can manipulate datasets by using SparkSQL’s DataFrames.

Reference documentation for the DSS Scala API can be found at:


Prior to writing Scala recipes, you need to ensure that DSS and Spark are properly configured together. See Setting up Spark integration.

Creating a Spark-Scala recipe

  • Create a new Spark-Scala recipe, either through a dataset’s Actions menu or in +Recipe > Hadoop & Spark > Spark-Scala
  • Add the input Datasets and/or Folders that will be used as source data in your recipes.
  • Select or create the output Datasets and/or Folder that will be filled by your recipe.
  • Click Create recipe.
  • You can now write your Spark code in Scala. A sample code is provided to get you started.


If the Spark-Scala icon is not enabled (greyed out), it can be because:

Basic Spark-Scala recipe

First of all, you will need to load the Dataiku API and entry points:

import com.dataiku.dss.spark._
import org.apache.spark.SparkContext
import org.apache.spark.sql.SQLContext
import org.apache.spark.sql.functions._

val sparkConf    = DataikuSparkContext.buildSparkConf()
val sparkContext = new SparkContext(sparkConf)
val sqlContext   = new SQLContext(sparkContext)
val dkuContext   = DataikuSparkContext.getContext(sparkContext)

You will then need to obtain DataFrames for your input datasets and directory handles for your input folders:

val inputDataset1 = dkuContext.getDataFrame(sqlContext, "KeyOfInputDataset")
val inputFolder1 =  dkuContext.getManagedFolderRoot("IdOfInputFolder")

These return a SparkSQL DataFrame and a Java File (pointing to the Folder’s root) respectively. You can then apply your transformations to the DataFrame and do what you need in the Folder.

Finally you can save the transformed DataFrame into the output dataset. By default the save method overwrites the dataset schema with that of the DataFrame:"KeyOfOutputDataset", transformedDataFrame)

If you declared the schema of the output dataset prior to running the Scala code, you can use save(…, writeSchema = false). However, it can be impractical to do so, especially if your code generates many columns (or columns that change often).

If you run your recipe on partitioned datasets, the above code will automatically load/save the partitions specified in the recipe parameters. You can forcibly load or save another partition (or load all partitions) of a dataset:

getDataFrame(sqlContext, "KeyOfInputDataset", partitions = Some(List("otherPartitionId")))
getDataFrame(sqlContext, "KeyOfInputDataset", partitions = Some(null)) // whole dataset
save("KeyOfOutputDataset", dataframe, targetPartition = Some("otherPartitionId"))

In the same vein, save will use the writing mode (append or overwrite) defined in the recipe’s configuration by default. You can also override this behavior, but please note that some dataset types do not support Append mode (e.g. HDFS):

save("KeyOfOutputDataset", dataframe, writeMode = Some(WriteMode.Append))