Don’t collect large RDDs

Apache Spark - Best Practices and Tuning

Search…

When a collect operation is issued on a RDD, the dataset is copied to the driver, i.e. the master node. A memory exception will be thrown if the dataset is too large to fit in memory; takeor takeSamplecan be used to retrieve only a capped number of elements instead.
Another way has been showed in [8] where you can get the array of partition indexes:
val parallel = sc.parallelize(1 to 9)
val parts = parallel.partitions
and then create a smaller rdd filtering out everything but a single partition. Collect the data from smaller rdd and iterate over values of a single partition:
for(p <- parts){
  val idx = p.index
  val partRDD = parallel.mapPartitionsWithIndex((index: Int, it: Iterator[Int]) => if(index == idx) it else Iterator(), true)
  val data = partRDD.collect
  // Data contains all values from a single partition in the form of array.
  // Now you can do with the data whatever you want: iterate, save to a file, etc.
}
​
// You can use also the foreachPartition operation
parallel.foreachPartition(partition => {
  partition.toArray
  // Your code
})
Of cause, it will work only if the partitions are small enough. If they aren't, you can always increase the number of partitions with rdd.coalesce(numParts, true).

Introduction

Next - RDD

Don't use count() when you don't need to return the exact number of rows

Last modified 2yr ago

Copy link