Define udf spark by reflection on String

Question

I am trying to define udf in spark (2.0) from a string containing the definition of a scala function. Here is a snippet:

val universe: scala.reflect.runtime.universe.type = scala.reflect.runtime.universe
import universe._
import scala.reflect.runtime.currentMirror
import scala.tools.reflect.ToolBox
val toolbox = currentMirror.mkToolBox()
val f = udf(toolbox.eval(toolbox.parse("(s:String) => 5")).asInstanceOf[String => Int])
sc.parallelize(Seq("1","5")).toDF.select(f(col("value"))).show

This gives me an error:

  Caused by: java.lang.ClassCastException: cannot assign instance of scala.collection.immutable.List$SerializationProxy to field org.apache.spark.rdd.RDD.org$apache$spark$rdd$RDD$$dependencies_ of type scala.collection.Seq in instance of org.apache.spark.rdd.MapPartitionsRDD
   at java.io.ObjectStreamClass$FieldReflector.setObjFieldValues(ObjectStreamClass.java:2133)
   at java.io.ObjectStreamClass.setObjFieldValues(ObjectStreamClass.java:1305)
   at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2024)
   at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1942)
   at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1808)
   at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
   at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2018)
   at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1942)
   at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1808)
   at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
   at java.io.ObjectInputStream.readObject(ObjectInputStream.java:373)
   at org.apache.spark.serializer.JavaDeserializationStream.readObject(JavaSerializer.scala:75)
   at org.apache.spark.serializer.JavaSerializerInstance.deserialize(JavaSerializer.scala:114)
   at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
   at org.apache.spark.scheduler.Task.run(Task.scala:85)
   at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:274)
   at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
   at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
   at java.lang.Thread.run(Thread.java:745)

However, when I define udf as:

val f = udf((s:String) => 5)

It works great. What is the problem? The ultimate goal is to take a string that has a defala function for scala and use it as udf.

Answer 1

As Giovanni noted, the problem is that the class loaders are different (you can examine this by calling .getClass.getClassLoaderon any object). Then, when workers try to deserialize their reflected function, all hell breaks.

, . , . , . , - node, , .

val f = udf {
  new Function1[String,Int] with Serializable {
    import scala.reflect.runtime.universe._
    import scala.reflect.runtime.currentMirror
    import scala.tools.reflect.ToolBox

    lazy val toolbox = currentMirror.mkToolBox()
    lazy val func = {
      println("reflected function") // triggered at every worker
      toolbox.eval(toolbox.parse("(s:String) => 5")).asInstanceOf[String => Int]
    }

    def apply(s: String): Int = func(s)
  }
}

sc.parallelize(Seq("1","5")).toDF.select(f(col("value"))).show .

println - , . spark-shell --master 'local', , spark-shell --master 'local[2]' .

UDF , , , toolbox func . , , - .

Answer 2

, ClassNotFoundException, JavaDeserializationStream , , - , , RDD/DataSet, ClassNotFoundError. , ( )

conf.setJars ( Array ("/fullpath/yourgeneratedjar.jar", "/fullpath/otherdependencies.jar") )

conf.set("spark.yarn.jars", "/fullpath/yourgeneratedjar.jar,/fullpath/otherdependencies.jar")