syncho

syncho

syncho is a thin and fast wrapper around fibers for node.js. The API is very similar to node-sync but optimized to reduce overhead (see benchmarks) and in less than 100 lines of code.

Usage

npm install syncho

API

Sync(fn)

Pass a function fn to run in a fiber. This is just a shortcut to Fiber(fn).run(). Synchronized functions will throw in case of error so you can use standard try and catch to run your code. All syncho methods need to be run inside a fiber.

var Sync = require('syncho');

Sync(function (){
  try {
    // run code synchronously inside fiber
  }
  catch (e) {
    // handle error
  }
});

Function.prototype.sync(thisArg, args)

Execute an asynchronous function inside a fiber, and return the result or throw in case of error. Use the idiomatic Function.prototype.call signature where the first argument is the object you want to bind the function to, the others are the standard arguments of the function.

var Sync = require('syncho');
var redis = require('redis');
var db = redis.createClient();

function asyncEcho (text, cb) {
  process.nextTick(function () {
    cb(null, text);
  });
}

function asyncError (text, cb) {
  process.nextTick(function () {
    cb('Too bad');
  });
}

Sync(function (){
  try {

    // simple aync function

    console.log('Before sync');
    console.log(asyncEcho.sync(null, 'Hello!'));
    console.log('After sync');

    // async function binding object

    db.set.sync(db, 'foo', 'bar');
    console.log(db.get.sync(db, 'foo'));

    // this will throw an error

    asyncError.sync(null, 'Hello!');

  }
  catch (e) {
    console.error(e);
  }
});

Function.prototype.future(thisArg, args)

Execute an asynchronous function asynchronously and return a future, an object with a unique wait function, which will return the result or throw an error when wait is called. Behind the scene, Function.prototype.sync is just a shortcut to fn.future(...).wait(). Futures can be used to execute code in parallel. Note that the function starts executing as soon as the future is created. When wait is called, if the asynchronous function has not returned results yet, then and only then it will yield the fiber, otherwise it will just return the result or throw.

var fs = require('fs');

Sync(function (){
  try {

    var a = fs.stat.future(null, 'package.json');
    var b = fs.stat.future(null, '.gitignore');
    var c = fs.stat.future(null, 'README.md');

    console.log(a.wait().size);
    console.log(b.wait().size);
    console.log(c.wait().size);

  }
  catch (e) {
    console.error(e);
  }
});

Function.prototype.async()

Return a function that accepts a callback to run a synchronous function asynchronously. This can be useful to pass functions that use fibers to external APIs that are not fiber aware and need to run code asynchronously. You can also use this to make parallel executions of synchronous functions by calling the future method on the now asynchronous function.

var Sync = require('syncho');
var redis = require('redis');
var db = redis.createClient();

function syncGet (key) {
  return db.get.sync(db, key);
}

Sync(function (){
  try {

    // Just putting some data
    db.set.sync(db, 'foo', 'bar');
    db.set.sync(db, 'bar', 'baz');
    db.set.sync(db, 'baz', 'foo');

    // making our sync function async
    var asyncGet = syncGet.async();

    // execute in parallel
    var foo = asyncGet.future(db, 'foo');
    var bar = asyncGet.future(db, 'bar');
    var baz = asyncGet.future(db, 'baz');

    console.log(foo.wait(), bar.wait(), baz.wait());

  }
  catch (e) {
    console.error(e);
  }
});

Function.prototype.wrap()

Wrap the function in a Fiber. Useful for events emitters and other functions that expect a function as an argument.

   syncSubjectUnderTest().should.have.been.run.in.a.Fiber;

Function.prototype.wrapIt()

Wrap a function in a Fiber passing done callback for mocha tests.

In setup.js

var Sync = require('syncho');

In your tests:

  it('runs the test in a fiber', function (done) {
    syncSubjectUnderTest().should.have.been.run.in.a.Fiber;
    done();
  }.wrapIt());

Sync.Fiber(fn)

Expose the original Fiber object in case you need it.

Sync.sleep(ms)

Sleep for number of milliseconds (default 1000).

var Sync = require('syncho');

Sync(function (){
  try {

    console.time('sleep');
    Sync.sleep(100);
    console.timeEnd('sleep');

  }
  catch (e) {
    console.error(e);
  }
});

Sync.middleware()

Middleware for express. Run each request in a fiber so that you can easily use sync methods.

var Sync = require('syncho');
var express = require('express'), app = express();
var fs = require('fs');

app.use(express.query());
app.use(Sync.middleware());

app.get('/', function (req, res) {
  var file = fs.readFile.sync(null, req.query.file);
  res.end(file);
});

app.listen(3000, function () {
  console.log('Express server listening on port 3000');
});

Why syncho and comparison with other fibers abstraction modules

Fibers are awesome. I am not going to discussed their value here as it has been thoroughly documented but going to focus on why a new module was needed when there is already a ton of existing abstractions.

I really loved the simplicity of the sync module from @0ctave compared to the original Future abstraction and other Fiber modules but there are some performance issues especially with Function.prototype.future in the sync module and code like statistics that I think unnecessary (500 loc for sync vs 100 loc for syncho). As this isn't a fork of the sync module at all (code is written entirely from scratch) and is not 100% compatible with its API [1], it was better to create a brand new module.

Compare the different APIs (based on code already running in a fiber for simplicity)

// Raw fibers
var Fiber = require('fibers');
var fiber = Fiber.current;
db.get('foo', function (err, res) {
  if (err) fiber.throw(err);
  else fiber.run(res);
});
Fiber.yield();

// Future
var Future = require('fibers/future');
var future = Future.wrap(db.get.bind(db));
var res = future('foo').wait();

// synchronize
var sync = require('synchronize');
sync(db, 'get');
var res = db.get('foo');

// syncho
var res = db.get.sync(db, 'foo'); // Sync only need to be required once in your project / module to run the code in the fiber

Also note that syncho does not decorate the fiber or any of the objects or functions that you sync and the only additions are the methods added to Function.prototype.

(1) API for futures is not compatible with the sync module due to the creation of multiple properties on each future which has quite an impact on performance. I chose a simple wait function as with Marcel's original future implementation.

ES6 generators and coroutines

With ES6 generators and coroutines, Fibers may become obsolete but until there is a stable node version released supporting them, the only way to run asynchronous code synchronously is using fibers. Also the yield and * keywords are a bit intrusive and the sync API is still the more elegant IMO. It will be interesting to see benchmarks with the stable version of node for comparison. At the moment (0.11.9), 'syncho' with fibers is faster than co with ES6 generators.

syncho-simple: 12647ms
syncho-object: 12793ms
co-simple: 13850ms
co-object: 14643ms

Benchmarks

NB: all the existing Fibers modules are in general pretty fast and the cost of fibers is marginal. The longer the IO takes, the more marginal it becomes to the point of being insignificant.

TL;DR syncho is the fastest (from 5% to 50 times faster than sync) of all fibers modules on every benchmark, and almost as fast as using asynchronous code except when running synchronous functions in parallel. It is even faster than using the async module (I haven't tested other asynchronous control-flow modules) and often faster than using raw fibers directly.

Simple function (10,000)

synchronize: 12905ms
syncho: 12750ms
syncho-future: 12712ms
sync: 13265ms
fibrous: 14171ms
fibers: 13074ms
future: 14212ms
async.series: 13181ms
asyncFn: 12872ms

Binding to an object (10,000)

synchronize: 12939ms
syncho: 12684ms
sync: 13502ms
fibrous: 14449ms
fibers: 13381ms
future: 14497ms
async.series: 13322ms
asyncFn: 12990ms

Parallel processing with futures (100,000)

sync-future: 8697ms
syncho-future: 755ms
async.parallel: 2331ms
future: 2004ms

Parallel processing of synced functions with async and futures (10,000)

syncho: 486ms
async.parallel: 307ms
sync: 23981ms

Contributions

Please open issues for bugs and suggestions in github. Pull requests with tests are welcome.

Author

Jerome Touffe-Blin, @jtblin, About me

License

syncho is copyright 2013 Jerome Touffe-Blin and contributors. It is licensed under the BSD license. See the include LICENSE file for details.