Reactive Streams

@kalmanb

Anatomy of an Actor (akka)

• Internally single threaded - easy to reason
• Actor systems - nice concurrency model
• Resource allocation (threads) in config at runtime
• Mailbox - be careful - unbounded - out of memory

... but wait

Why is it hard?

Options

Java8 Streams

• Eager evaluation - forEach :(
• Static push or pull :(
• Cannot reuse streams :(

RxJava

• Back pressure by blocking :(
• No back pressure for remote workers :(
• Push only :(

Work Pulling Pattern

pull <-----> push

• Push - to workers
• Pull - from coordinator
• Choose one or both?

Work Pulling Pattern - both

• Pull - when the coordinator has lots of work to do
• Push - when the workers have more capacity than the coordinator

http://www.reactive-streams.org

• Typesafe
• Netflix
• Oracle
• Pivotal
• Red Hat
• Twitter

Reactive Streams Goal

• A standard for asynchronous stream processing with non-blocking back pressure on the JVM
• Cool = Interop with RxJava, Vertx, Akka etc

Simple Interface

trait Publisher[T] {
  def subscribe(sub: Subscriber[T]): Unit
}

triat Subscription {
  def requestMore(n: Int): Unit
  def cancel(): Unit
}

trait Subscriber[T] {
  def onSubscribe(s: Subscription): Unit
  def onNext(elem: T): Unit
  def onError(thr: Throwable): Unit
  def onComplete(): Unit
}

Examples

• Data transfer
• Real-time data processing - stock market
• Batch processing - spark
• Monitoring - analytics

Akka Reactive Streams

• Implementation of the work pulling pattern
• All the complexity is hidden
• Non blocking back pressure :)
• Dynamic pull / push :)
• Statically Typed :)
• Remotable back pressure :)
• Reusable :)

Demo

@kalmanb