Apache Spark Task not Serializable when Class exends Serializable - scala

Question

I am consistently having errors regarding Task not Serializable.
I have made a small Class and it extends Serializable - which is what I believe is meant to be the case when you need values in it to be serialised.
class SGD(filePath : String) extends Serializable {
val rdd = sc.textFile(filePath)
val mappedRDD = rdd.map(x => x.split(" ")
.slice(0,3))
.map(y => Rating(y(0).toInt, y(1).toInt, y(2).toDouble))
.cache
val RNG = new Random(1)
val factorsRDD = mappedRDD(x => (x.user, (x.product, x.rating)))
.groupByKey
.mapValues(listOfItemsAndRatings =>
Vector(Array.fill(2){RNG.nextDouble}))
}
The final line always results in a Task not Serializable error. What I do not understand is: the Class is Serializable; and, the Class Random is also Serializable according to the API. So, what am I doing wrong? I consistently can't get stuff like this to work; therefore, I imagine my understanding is wrong. I keep being told the Class must be Serializable... well it is and it still doesn't work!?

Answer 1

scala.util.Random was not Serializable until 2.11.0-M2.
Most likely you are using an earlier version of Scala.
A class doesn't become Serializable until all its members are Serializable as well (or some other mechanism is provided to serialize them, e.g. transient or readObject/writeObject.)
I get the following stacktrace when running given example in spark-1.3:
Caused by: java.io.NotSerializableException: scala.util.Random
Serialization stack:
- object not serializable (class: scala.util.Random, value: scala.util.Random#52bbf03d)
- field (class: $iwC$$iwC$SGD, name: RNG, type: class scala.util.Random)
One way to fix it is to take instatiation of random variable within mapValues:
mapValues(listOfItemsAndRatings => { val RNG = new Random(1)
Vector(Array.fill(2)(RNG.nextDouble)) })