How to force DataFrame evaluation in Spark

Question

Sometimes (e.g. for testing and bechmarking) I want force the execution of the transformations defined on a DataFrame. AFAIK calling an action like count does n

Answer 1

It appears that df.cache.count is the way to go:

scala> val myUDF = udf((i:Int) => {if(i==1000) throw new RuntimeException;i})
myUDF: org.apache.spark.sql.expressions.UserDefinedFunction = UserDefinedFunction(,IntegerType,Some(List(IntegerType)))

scala> val df = sc.parallelize(1 to 1000).toDF("id")
df: org.apache.spark.sql.DataFrame = [id: int]

scala> df.withColumn("test",myUDF($"id")).show(10)
[rdd_51_0]
+---+----+
| id|test|
+---+----+
|  1|   1|
|  2|   2|
|  3|   3|
|  4|   4|
|  5|   5|
|  6|   6|
|  7|   7|
|  8|   8|
|  9|   9|
| 10|  10|
+---+----+
only showing top 10 rows

scala> df.withColumn("test",myUDF($"id")).count
res13: Long = 1000

scala> df.withColumn("test",myUDF($"id")).cache.count
org.apache.spark.SparkException: Failed to execute user defined function($anonfun$1: (int) => int)
        at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
.
.
.
Caused by: java.lang.RuntimeException

Source