Conflux

Conflux is Redux for distributed systems.

Contents

• Motivation
• What Can I Build With Conflux?
• Quick Example
• API
  • Creating an instance
  • Performing Methods To Dispatch Actions
  • Reducing Actions Into State
  • Subscribing to changes
  • Getting Committed and Provisional State
  • Deconstructing an instance
• Correctness
• License

Motivation

Distributed systems are hard. Conflux is an attempt at making distributed systems understandable. It aims to do what Redux did for Flux, and what Raft did for Paxos.

Natrually, it does this by composing the two ideas.

What Can I Build With Conflux?

You can build serverless applications! I don't mean "serverless" as in "uses AWS Lambda". I mean serverless as in "there is no central server, just a cluster of nodes that coordinate with each other". Think Bittorent.

Conflux is very new, so I am working on a few example applications to demonstrate what's possible. If you build something with Conflux, send me a PR so I can add it to this list.

• Distributed Mutex

Quick Example

var node = conflux({
  id: uuid.v4()

  // how many nodes are in the cluster?
, clusterSize: 5

  // how should nodes communicate?
, channel: {
    name: 'redis'
  }

  // define your action creators
, methods: {
    append: function (thing) {
      return thing
    }
  }

  // define your reducer
, reduce: function (state, action) {
    state = state == null ? [] : state

    return state.concat(action)
  }
})

// subscribe to changes
node.subscribe(function () {
  console.log(nodeB.getState().log.join(' '))
})

// perform an action
node.perform('append', ['foo'])

If you've used Redux before, Conflux should look familiar. You subscribe() to a Conflux instance, and call getState() inside to get the current state. Instead of dispatching actions directly, you perform() Methods that dispatch() them. Methods are declared when you construct a Conflux instance, and are the equivalent of Action Creators in Redux.

API

Creating an instance

var conflux = require('conflux')
    // uuids are recommended, but you can use any string id
  , uuid = require('uuid')
  , c = conflux({
      /**
      * Required settings
      */

      id: uuid.v4()
    , clusterSize: 5
    , channel: {
        name: 'redis' // or 'memory', etc

        // ... additional keys are passed as options to the "redis" channel.
        // see Gaggle's Channel docs for available channels and options:
        // https://github.com/ben-ng/gaggle#channels
      }

      // these are "action creators" in redux parlance
    , methods: {
        foobar: function (foo, bar) {
          return {
            type: 'FOOBAR'
          , foo: foo
          , bar: bar
          }
        }
      }

      // this is the reducer function that redux stores are constructed with
    , reduce: function (state, action) {
        // return a new `state` using the information in `action`
      }

      /**
      * Optional, advanced settings. These control the parameters of the
      * underlying Raft algorithm, so you can optimize performance for the
      * network that you are on. You should set them lower on fast networks
      * and higher on slow networks.
      */

      // The range of random values that will be selected for the
      // Raft leader election timeout, in milliseconds
    , electionTimeout: {
        min: 300
      , max: 500
      }

      // The interval in milliseconds where the leader node will
      // send heartbeats to followers. Must be significantly shorter
      // than the electionTimeout. Should be longer than the average
      // round-trip message time.
    , heartbeatInterval: 50
    })

Conflux is built on top of Gaggle, and therefore supports any communication channel that Gaggle supports.

Performing Methods To Dispatch Actions

c.perform(String methodName, Array args, [Number timeout], [Function callback])

You never dispatch Actions directly in Conflux. Actions must be dispatched from the body of a Method. Methods must be synchronous. You declare Methods when constructing a Conflux instance, and call them with perform(). These Methods return the Action to be dispatched, null if nothing should be done, and an Error if the Action is invalid for the provisional state.

var opts = {
  methods: {
    foobar: function (foo, bar) {
      // The perform callback will be called with no error and this return
      // value as the second argument
      return {
        type: 'FOOBAR'
      , foo: foo
      , bar: bar
      }
    }
  , bonk: function () {
      return new Error('Whoops')
    }
  , noop: function () {
      // The perform callback will be called with no error or response
      return null
    }
  }

  // ... other Conflux options, like the channel to use, node id, etc...
}

var c = conflux(opts)

// Callback API
c.perform('noop', [], function (err) {
  console.log(err) // => null
})

c.perform('noop', [], 5000, function (err) {
  console.log(err) // => null
})

// Promise API
c.perform('foobar', ['a', 'b'])
.then(function (action) {
  console.log(action) // => {type: 'FOOBAR', foo: 'a', bar: 'b'}
})

c.perform('bonk', [], 5000)
.catch(function (err) {
  console.log(err) // => Error: Whoops
})

Reducing Actions Into State

// A starter template for your own reducer
function (state, action) {
  // Set initial state or clone existing state
  if (state == null) {
    state = {}
  }
  else {
    state = JSON.parse(JSON.stringify(state))
  }

  // Ignore unknown actions
  if (action == null) {
    return state
  }

  // Handle known actions
  switch (action.type) {
    case 'FOO':
      state.isFoo = true
    break
    case 'BAR':
      state.isBar = true
    break
  }

  // Return the new state
  return state
}

Reducers should obey a few rules:

1. Do not mutate state
2. Always return a new state
3. Be prepared to set an inital state if state is undefined
4. If the action is unrecognized (it might be null when Conflux initializes the state, for example), return the same state

Subscribing to changes

c.subscribe(function() callback)

Calls callback whenever an action is committed. Returns an unsubscribe function that when called, removes callback from the subscriptions.

Getting Committed and Provisional State

c.getState()
c.getProvisionalState()

Unlike Redux, Conflux has two types of state: committed state, and provisional state.

getState() gets you the committed state of the node. All nodes are guaranteed to enter this state at some point, but it does not contain the effects of uncommitted Actions.

getProvisionalState() gets you the state of the node if all currently uncommitted actions are committed. Since the leader might fail before these Actions are committed, it is possible that no nodes ever actually enter this state.

You should use the provisional state in your Methods to determine the validity of an Action. The committed state should be used just about everywhere else, like in your subscribe() callback.

Deconstructing an instance

c.close([function(Error) callback])

When you're done, call close to remove event listeners and disconnect the channel.

c.close(function (err) {})

c.close().then()

Correctness

Distributed systems are really difficult to prove and test, and Conflux is no exception. I am still working on formal proofs, but in the meantime here is an incomplete list of things that are being done in the name of correctness.

• Conflux has integration tests with full statement and branch coverage
• It is built on Gaggle, which has integration tests with full statement coverage
• My distributed mutex is built on Conflux and has full statement and branch coverage and a fuzz test you can run yourself

License

Copyright (c) 2016 Ben Ng [email protected]

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.