dominictarr/scuttlebutt

scuttlebutt

A base-class for real-time replication.

This seems like a silly name, but I assure you, this is real science. Read this: http://www.cs.cornell.edu/home/rvr/papers/flowgossip.pdf

Or, if you're lazy: http://en.wikipedia.org/wiki/Scuttlebutt (laziness will get you nowhere, btw)

secure-scuttlebutt

Creating this module eventually lead me to secure-scuttlebutt it is based on a similar replication protocol as scuttlebutt, but puts security at the forefront.

Subclasses

Scuttlebutt is intended to be subclassed into a variety of data-models.

Two implementations are provided as examples scuttlebutt/model and scuttlebutt/events

subclasses:

• crdt higher-level, with sets and sequences.
• r-array Replicatable Array.
• r-edit Collaborative Text Editing.
• append-only news feed.
• scuttlebucket combine multiple scuttlebutts into one.
• expiry-model memory capped model with expiring keys.
• r-value replicate a single value.
• gossip-object like scuttlebutt/model, but supports nested objects.
• scuttleboat a dynamic scuttlebucket!
• r-string character-wise collaborative text editing.

Replication

Any Scuttlebutt subclass is replicated with createStream. Create a server and then connect a client to it:

var Model = require('scuttlebutt/model')
var net   = require('net')

var s = new Model()
var z = new Model()

net.createServer(function (stream) {

  stream.pipe(s.createStream()).pipe(stream)

}).listen(8000, function () {

  //then connect the client!
  var stream = net.connect(8000)
  stream.pipe(z.createStream()).pipe(stream)  

})

Gotchas

Scuttlebutt is always duplex. Scuttlebutt does a handshake on connecting to another scuttlebutt, and this won't work unless both sides are connected.

Right

stream.pipe(model.createStream()).pipe(stream)

WRONG!

wrongStream.pipe(model2.createStream())

Also, when creating a server, scuttlebutt needs a stream for EACH connection.

Right

net.createServer(function (stream) {
  stream.pipe(model.createStream()).pipe(stream)
}).listen(port)

WRONG!

this will use one stream for many connections!

var wrongStream = model.createStream()
net.createServer(function (stream) {
  stream.pipe(wrongStream).pipe(stream)
}).listen(port)

Errors and use in PRODUCTION

If have are using scuttlebutt in production, you must register on 'error' listener in case someone sends invalid data to it.

** Any stream that gets parsed should have an error listener! **

net.createServer(function (stream) {
  var ms = m.createStream()
  stream.pipe(ms).pipe(stream)
  ms.on('error', function () {
    stream.destroy()
  })
  stream.on('error', function () {
    ms.destroy()
  })
}).listen(9999)

Otherwise, if someone tries to connect to port 9999 with a different protocol (say, HTTP) this will emit an error. You must handle this and close the connection / log the error.

Also, you should handle errors on stream, stream may error if the client responsible for it crashes.

Persistence

Persist by saving to at least one writable stream.

var Model = require('scuttlebutt/model') //or some other subclass...
var fs = require('fs')
var m = new Model()

//stream FROM disk.
fs.createReadStream(file).pipe(m.createWriteStream())

//stream TO disk.
m.on('sync', function () {
  m.createReadStream().pipe(fs.createWriteStream(file))
})

Use on('sync',... to wait until the persisted state is in the file before writing to disk. (Make sure you rotate files, else there is an edge case where if the process crashes before the history has been written some data will be lost /this is where link to module for that will go/)

You may use kv to get streams to local storage.

read only mode.

Sometimes you want to use scuttlebutt to send data one way, from a master instance to a slave instance.

var s1 = master.createStream({writable: false, sendClock: true})
var s2 = slave.createStream({readable: false, sendClock: true})

master will emit updates, but not accept them, over this stream. This checking is per stream - so it's possible to attach master to another master node and have master nodes replicate each way.

Implementing Custom Scuttlebutts

A custom Scuttlebutt is a data model that inherits from Scuttlebutt. It must provide an implementation of history() and applyUpdate().

See r-value for a demonstration of the simplest possible Scuttlebutt, one that replicates a a single value.

Scuttlebutt#history(sources)

sources is a hash of source_ids: timestamps. History must return an array of all known events from all sources That occur after the given timestamps for each source.

The array MUST be in order by timestamp.

Scuttlebutt#applyUpdate (update)

Each update is of the form [change, timestamp, source] (see Protocol below).

Possibly apply a given update to the subclasses model. Return 'true' if the update was applied. (See scuttlebutt/model.js for an example of a subclass that does not apply every update.)

Scuttlebutt#createStream (opts)

Create a duplex stream to replicate with a remote endpoint.

The stream returned here emits a special 'header' event with the id of the local and remote nodes and the vector clock. You can set metadata on the header object using opts.meta.

Examples

Connect two Model scuttlebutts locally.

var Model = require('scuttlebutt/model')

var a = new Model()
var b = new Model()

a.set(key, value)

b.on('update', console.log)

var s = a.createStream()
s.pipe(b.createStream()).pipe(s)

scuttlebutt/events

A reliable event emmitter. Multiple instances of an emitter may be connected to each other and will remember events, so that they may be present after a disconnection or crash.

With this approach it is also possible to persist events to disk, making them durable over crashes.

var Emitter = require('scuttlebutt/events')
var emitter = new Emitter()

emit (event, data)

Emit an event. Only one argument is permitted.

on (event, listener)

Add an event listener.

scuttlebutt/model

A replicateable Model object, a simple key-value store.

var Model = require('scuttlebutt/model')
var model = new Model()

get (key)

Get a property.

set (key, value)

Set a property.

on('update', function ([key, value], source, updateId))

Emmitted when a property changes. If source !== this.id then it was a remote update.

Protocol

Messages are sent in this format:

[change, timestamp, source]

source is the id of the node which originated this message. Timestamp is the time when the message was created. This message is created using Scuttlebutt#localUpdate(key, value).

When two Scuttlebutts are piped together, they both exchange their current list of sources. This is an object of {source_id: latest_timestamp_for_source_id} After receiving this message, Scuttlebutt sends any messages not yet known by the other end. This is the heart of Scuttlebutt Reconciliation.

Security

Scuttlebutt has an (optional) heavy duty security model using public keys. This enables a high level of security even in peer-to-peer applications. You can be sure that a given message is from the node that sent it, even if you did not receive the messasge from them directly.

Enabling Security

var model = require('scuttlebutt/model')
var security = require('scuttlebutt/security')
var keys = {}
var m = new Model(security(keys, PRIVATE, PUBLIC))

Security API

When security is enabled, each scuttlebutt message is signed with a private key. It is then possible for any scuttlebutt instance to be confident about the authenticity of the message by verifying it against the source's public key.

This is possible even if the verifying node received the message from an intermediate node.

Security is activated by passing in a security object to the contructor of a scuttlebutt subclass.

Use the included implementation:

var security = require('scuttlebutt/security')(keys, PRIVATE, PUBLIC)
var Model = require('scuttlebutt/model')

var m = new Model(security)

See scuttlebutt/security.js for a simple example implementation.

sign(update) should sign the update with the instance's private key. verify(update, cb) should verify the update, using public key associated with the source field in the update. Verification may be asynchronous. verify must callback cb(err, boolean) where boolean indicates whether or not the signature is valid. Only callback in error in the most extreme circumstances. If there was no known key for the required source then that should be treated as a verification failure. If it is not possible to reach the key database (or whatever) then the request should be retried until it is available.

Note: although the API supports asynchronous verification, it's probably a good idea to load keys into memory so that messages can be verified and signed synchronously.

createId() returns a new id for the current node. This is used in the example security implementation to return a id that is a hash of the public key. This makes it impossible for rogue nodes to attempt to associate a old node id with a new public key.

Generating Keys.

Generate an ssh private key, and a PEM encoded public key.

ssh-keygen -f $KEYNAME -b $LENGTH -N $PASSWORD -q
ssh-keygen -e -f $KEYNAME.pub -m PEM > $KEYNAME.pem

$LENGTH must be >= 786, shorter is faster but less secure. password may be empty ''.

$KEYNAME is the private key, and $KEYNAME.pem is the public key to use with Scuttlebutt.