Skip to content
forked from archie/raft

Distributed consensus algorithm implemented in Scala and Akka

License

Notifications You must be signed in to change notification settings

pablosmedina/raft

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

93 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Raft in Scala

This is a partial implementation of Raft using Scala and Akka. As of now it has support for leader election and log replication. The outstanding issues are log compaction and cluster membership changes.

Build Status

Example

The following code snippet illustrates how to implement a client which periodically asks a Raft cluster for sequence numbers. Hence, the state machine that Raft executes commands against implements total ordering. Most of the magic is deferred to RaftClient which provides the interface to the Raft cluster.

class Sequencer extends Actor with RaftClient with ActorLogging {
  import context._

  def schedule = system.scheduler.scheduleOnce(50 millis, self, "sequence")

  override def preStart() = schedule
  override def postRestart(reason: Throwable) = {}

  def receive = {
    case "sequence" =>
      decide("get") onComplete {
        case Success(x) => log.info(s"Got $x")
        case Failure(t) => log.info(s"Error ${t.getMessage()}")
      }

      schedule
  }
}

object Main extends App {
  implicit val system = ActorSystem("raft")
  val members = Raft(3)
  val client = system.actorOf(Props[Sequencer], "client")

  println("Running raft demo - press enter key to exit")
  Console.readLine

  system.shutdown
}

This demo sets up a Raft cluster with five nodes on the same machine. In the future support for remote nodes will be added.

Background

What is Raft?

Raft is a distributed consensus algorithm developed by Diego Ongaro and John Ousterhout. Its original implementation was done by Diego in C++ and is called LogCabin.

What is consensus?

In a nutshell, consensus is about getting a set of computers to agree on a single value for a particular moment in time, and will continue to do so correctly as long as a majority of the machines remain available. Except for in politics, this may seem a trivial problem to solve. However, in 1985 a group of researchers proved that under specific conditions this is in theory impossible to solve. Luckily the world is not that rough and some randomness helps a long way.

The classroom algorithm for styding consensus in distributed systems has for a long time been Paxos developed by Leslie Lamport. It is also known for being notoriously difficult to implement.

Why implement Raft in Scala and Akka?

Together with Nick Rutherford, I implemented Paxos using Erlang as part of our master's programme. We have first-hand experience on the tricky implementation details of Paxos, fighting furiously to get the paper's definition to a working algorithm. Raft is designed to be easier to understand and I wanted to test that argument during my time at Hacker School.

Interestingly, there are quite a few open source implementations of Raft in a wide variety of languages, and at least one other in Scala using Finagle.

Moreover, Akka suitable for writing distributed applications and Scala is a pretty neat language that I've been wanting to explore further.

Install

Make sure you have at least Scala 2.10.2 and the build tool sbt installed.

  1. Clone this repo
  2. Run the tests sbt test. sbt will download the necessary dependencies automatically.
  3. Run the demo in the example above using sbt run

Configuration

The only configuration as of now is maintained in src/main/resources/application.conf.

Contributions

Pull requests, small and large, are more than welcome. Whether it is a suggestion for writing more idiomatic Scala, better test cases, or a bug in the implementation - feel free to fire away!

Todo

  • Log compaction
  • Membership changes
  • Remoting support
  • Better/more examples

Endnote

License: MIT

Made at Hacker School.

(c) Marcus Ljungblad, 2013

About

Distributed consensus algorithm implemented in Scala and Akka

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published