I am working on a Spark app in which an RDD is first calculated, then need to be stored to disk, and then loaded again into Spark. To this end, I am looking for a minimal working example of saving an RDD to a local file and then loading it.
The file format is not suitable for text conversion, so saveAsTextFile won't fly.
The RDD can either be a plain RDD or Pair RDD, it is not crucial. The file format can be either of HDFS or not.
The example can be either in Java or Scala.
Thanks!
解决方案
As long as values in the RDD are serializable you can try to use RDD.saveAsObjectFile / SparkContext.objectFile:
case class Foobar(foo: Int, bar: Map[String, Int])
val rdd = sc.parallelize(Seq(
Foobar(1, Map("foo" -> 0)),
Foobar(-1, Map("bar" -> 3))
))
rdd.saveAsObjectFile("foobar")
sc.objectFile[Foobar]("foobar")