Stateful operations in Spark Streaming

PairDStreamFunctions.updateStateByKey

updateState by key can be used to create a stateful DStream based on upcoming data. It requires a function:

object UpdateStateFunctions {
  def updateState(current: Seq[Double], previous: Option[StatCounter]) = {
    previous.map(s => s.merge(current)).orElse(Some(StatCounter(current)))
  }
}

which takes a sequence of the current values, an Option of previous state and returns an Option of the updated state. Putting this all together:

import org.apache.spark._
import org.apache.spark.streaming.dstream.DStream
import scala.collection.mutable.Queue
import org.apache.spark.util.StatCounter
import org.apache.spark.streaming._

object UpdateStateByKeyApp {
  def main(args: Array[String]) {

    val sc = new SparkContext("local", "updateStateByKey", new SparkConf())
    val ssc = new StreamingContext(sc, Seconds(10))
    ssc.checkpoint("/tmp/chk")

    val queue = Queue(
      sc.parallelize(Seq(("foo", 5.0), ("bar", 1.0))),
      sc.parallelize(Seq(("foo", 1.0), ("foo", 99.0))),
      sc.parallelize(Seq(("bar", 22.0), ("foo", 1.0))),
      sc.emptyRDD[(String, Double)],
      sc.emptyRDD[(String, Double)],
      sc.emptyRDD[(String, Double)],
      sc.parallelize(Seq(("foo", 1.0), ("bar", 1.0)))
    )

    val inputStream: DStream[(String, Double)] = ssc.queueStream(queue)

    inputStream.updateStateByKey(UpdateStateFunctions.updateState _).print()

    ssc.start()
    ssc.awaitTermination()
    ssc.stop()
  }
}

PairDStreamFunctions.mapWithState

mapWithState, similarly to updateState, can be used to create a stateful DStream based on upcoming data. It requires StateSpec:

import org.apache.spark.streaming._

object StatefulStats {
  val state = StateSpec.function(
    (key: String, current: Option[Double], state: State[StatCounter]) => {
      (current, state.getOption) match {
        case (Some(x), Some(cnt)) => state.update(cnt.merge(x))
        case (Some(x), None) => state.update(StatCounter(x))
        case (None, None) => state.update(StatCounter())
        case _ =>
      }
      
      (key, state.get)
    }
  )
}

which takes key key, current value and accumulated State and returns new state. Putting this all together:

import org.apache.spark._
import org.apache.spark.streaming.dstream.DStream
import scala.collection.mutable.Queue
import org.apache.spark.util.StatCounter

object MapStateByKeyApp {
  def main(args: Array[String]) {
    val sc = new SparkContext("local", "mapWithState", new SparkConf())

    val ssc = new StreamingContext(sc, Seconds(10))
    ssc.checkpoint("/tmp/chk")

    val queue = Queue(
      sc.parallelize(Seq(("foo", 5.0), ("bar", 1.0))),
      sc.parallelize(Seq(("foo", 1.0), ("foo", 99.0))),
      sc.parallelize(Seq(("bar", 22.0), ("foo", 1.0))),
      sc.emptyRDD[(String, Double)],
      sc.parallelize(Seq(("foo", 1.0), ("bar", 1.0)))
    )

    val inputStream: DStream[(String, Double)] = ssc.queueStream(queue)

    inputStream.mapWithState(StatefulStats.state).print()

    ssc.start()
    ssc.awaitTermination()
    ssc.stop()
  }
}

Finally expected output:

-------------------------------------------
Time: 1469923280000 ms
-------------------------------------------
(foo,(count: 1, mean: 5.000000, stdev: 0.000000, max: 5.000000, min: 5.000000))
(bar,(count: 1, mean: 1.000000, stdev: 0.000000, max: 1.000000, min: 1.000000))

-------------------------------------------
Time: 1469923290000 ms
-------------------------------------------
(foo,(count: 3, mean: 35.000000, stdev: 45.284287, max: 99.000000, min: 1.000000))
(foo,(count: 3, mean: 35.000000, stdev: 45.284287, max: 99.000000, min: 1.000000))

-------------------------------------------
Time: 1469923300000 ms
-------------------------------------------
(bar,(count: 2, mean: 11.500000, stdev: 10.500000, max: 22.000000, min: 1.000000))
(foo,(count: 4, mean: 26.500000, stdev: 41.889736, max: 99.000000, min: 1.000000))

-------------------------------------------
Time: 1469923310000 ms
-------------------------------------------

-------------------------------------------
Time: 1469923320000 ms
-------------------------------------------
(foo,(count: 5, mean: 21.400000, stdev: 38.830916, max: 99.000000, min: 1.000000))
(bar,(count: 3, mean: 8.000000, stdev: 9.899495, max: 22.000000, min: 1.000000))