An implementation of the actor model for Node.js, aiming for high scale, low-latency and clustering
Switch branches/tags
Nothing to show
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
.gitignore initial commit Jan 23, 2014
Solution.sln initial commit Jan 23, 2014
index.js improved documentation Jan 24, 2014
package.json improved documentation Jan 24, 2014 added todos Jan 24, 2014
test.js added bloom filters Jan 24, 2014
towtruck.njsproj initial commit Jan 23, 2014


Experimental actor model for Node.js, attempting to build a high-scale low latency system, by maintaining a pool of in-memory objects (actors).


npm install towtruck


// create an instance of towtruck, and choose a port number
var towtruck = require('towtruck');

// you need to supply a function which will create actors (objects)
function createActor(id, client){
  return {
    // the object can have methods, which can be called over http, or from other actors
    hello: function(data, cb){
	  cb(null, {hello: "from actor " + id});

When the application is running, you access actors over http:


{"hello":"from actor 1"}

Any JSON that you post in the body will appear as the data argument.

Actors are created when they are requested, and will stay in memory until they are garbage collected. The GC interval can be controlled with configuration.

Other commands are available over http.

Register another towtruck instance (this must be reciprocated):

Query the registered servers:

Remove a server:

Get the health of a server:

Calling other actors from inside an actor:

// the client function allows you to make calls from within twotruck
function createActor(id, client){
  return {
    hello: function(data, cb){

	  // this will invoke the goodbye function on actor 1
	  client.invoke(1, "goodbye", { data:"to send"}, function(err, dataBack){
		cb(null, {"hello from bar": dataBack});
	goodbye: function(data, cb){
	  cb(err, {farewell: true});


  • Support all HTTP verbs on cluster (everything is post)
  • Forward request body to cluster
  • Implement bloom filters
  • Write some documentation
  • Observe head size when garbage collecting
  • Think about clustering a bit more
  • Optimise
  • Test at scale
  • Support node failure
  • Vastly improve error reporting
  • Record metrics