1 Web Workers

Draft Recommendation — 14 January 2010

You can take part in this work. Join the working group's discussion list.

Web designers! We have a FAQ, a forum, and a help mailing list for you!

This version:
http://whatwg.org/ww
Version history:
Twitter messages (non-editorial changes only): http://twitter.com/WHATWG
Commit-Watchers mailing list: http://lists.whatwg.org/listinfo.cgi/commit-watchers-whatwg.org
Interactive Web interface: http://html5.org/tools/web-workers-tracker
Subversion interface: http://svn.whatwg.org/webworkers/
Issues:
To send feedback: whatwg@whatwg.org
To view and vote on feedback: http://www.whatwg.org/issues/
Editor:
Ian Hickson, Google, ian@hixie.ch

Abstract

This specification defines an API that allows Web application authors to spawn background workers running scripts in parallel to their main page. This allows for thread-like operation with message-passing as the coordination mechanism.

Status of this document

This is a work in progress! This document is changing on a daily if not hourly basis in response to comments and as a general part of its development process. Comments are very welcome, please send them to whatwg@whatwg.org. Thank you.

The current focus is in developing a first draft proposal.

Implementors should be aware that this specification is not stable. Implementors who are not taking part in the discussions are likely to find the specification changing out from under them in incompatible ways. Vendors interested in implementing this specification before it eventually reaches the call for implementations should join the WHATWG mailing list and take part in the discussions.

This specification is also being produced by the W3C Web Apps WG. The two specifications are identical from the table of contents onwards.

Table of contents


1.1 Introduction

1.1.1 Scope

This section is non-normative.

This specification defines an API for running scripts in the background independently of any user interface scripts.

This allows for long-running scripts that are not interrupted by scripts that respond to clicks or other user interactions, and allows long tasks to be executed without yielding to keep the page responsive.

Workers (as these background scripts are called herein) are relatively heavy-weight, and are not intended to be used in large numbers. For example, it would be inappropriate to launch one worker for each pixel of a four megapixel image. The examples below show some appropriate uses of workers.

Generally, workers are expected to be long-lived, have a high start-up performance cost, and a high per-instance memory cost.

1.1.2 Tutorial

This section is non-normative.

There are a variety of uses that workers can be put to. The following subsections show various examples of this use.

1.1.2.1 A background number-crunching worker

This section is non-normative.

The simplest use of workers is for performing a computationally expensive task without interrupting the user interface.

In this example, the main document spawns a worker to (naïvely) compute prime numbers, and progressively displays the most recently found prime number.

The main page is as follows:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Worker example: One-core computation</title>
 </head>
 <body>
  <p>The highest prime number discovered so far is: <output id="result"></output></p>
  <script>
   var worker = new Worker('worker.js');
   worker.onmessage = function (event) {
     document.getElementById('result').textContent = event.data;
   };
  </script>
 </body>
</html>

The Worker() constructor call creates a worker and returns a Worker object representing that worker, which is used to communicate with the worker. That object's onmessage event handler allows the code to receive messages from the worker.

The worker itself is as follows:

var n = 1;
search: while (true) {
  n += 1;
  for (var i = 2; i <= Math.sqrt(n); i += 1)
    if (n % i == 0)
     continue search;
  // found a prime!
  postMessage(n);
}

The bulk of this code is simply an unoptimized search for a prime number. To send a message back to the page, the postMessage() method is used to post a message when a prime is found.

View this example online.

1.1.2.2 A worker for updating a client-side database

This section is non-normative.

In this example, the main document spawns a worker whose only task is to listen for notifications from the server, and, when appropriate, either add or remove data from the client-side database.

Since no communication occurs between the worker and the main page, the main page can start the worker by just doing:

<script>
 new Worker('worker.js');
</script>

The worker itself is as follows:

var server = new WebSocket('ws://whatwg.org/database');
var database = openDatabase('demobase', '1.0', 'Demo Database', 10240);
server.onmessage = function (event) {
  // data is in the format "command key value"
  var data = event.data.split(' ');
  switch (data[0]) {
    case '+':
     database.transaction(function(tx) {
       tx.executeSql('INSERT INTO pairs (key, value) VALUES (?, ?)', data[1], data[2]);
     });
    case '-':
     database.transaction(function(tx) {
       tx.executeSql('DELETE FROM pairs WHERE key=? AND value=?', data[1], data[2]);
     });
  }
};

This connects to the server using the WebSocket mechanism and opens the local database (which, we presume, has been created earlier). The worker then just listens for messages from the server and acts on them as appropriate, forever (or until the main page is closed).

View this example online. (This example will not actually function, since the server does not actually exist and the database is not created by this sample code.)

1.1.2.3 Worker used for background I/O

This section is non-normative.

In this example, the main document uses two workers, one for fetching stock updates for at regular intervals, and one for fetching performing search queries that the user requests.

The main page is as follows:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Worker example: Stock ticker</title>
  <script>
   // TICKER
   var symbol = 'GOOG'; // default symbol to watch
   var ticker = new Worker('ticker.js');

   // SEARCHER
   var searcher = new Worker('searcher.js');
   function search(query) {
     searcher.postMessage(query);
   }

   // SYMBOL SELECTION UI
   function select(newSymbol) {
     symbol = newSymbol;
     ticker.postMessage(symbol);
   }
  </script>
 </head>
 <body onload="search('')">
  <p><output id="symbol"></output> <output id="value"></output></p>
  <script>
   ticker.onmessage = function (event) {
     var data = event.data.split(' ');
     document.getElementById('symbol').textContent = data[0];
     document.getElementById('value').textContent = data[1];
   };
   ticker.postMessage(symbol);
  </script>
  <p><label>Search: <input type="text" autofocus oninput="search(this.value)"></label></p>
  <ul id="results"></ul>
  <script>
   searcher.onmessage = function (event) {
     var data = event.data.split(' ');
     var results = document.getElementById('results');
     while (results.hasChildNodes()) // clear previous results
       results.removeChild(results.firstChild);
     for (var i = 0; i < data.length; i += 1) {
       // add a list item with a button for each result
       var li = document.createElement('li');
       var button = document.createElement('button');
       button.value = data[i];
       button.type = 'button';
       button.onclick = function () { select(this.value); };
       button.textContent = data[i];
       li.appendChild(button);
       results.appendChild(li);
     }
   };
  </script>
  <p>(The data in this example is not real. Try searching for "Google" or "Apple".)</p>
 </body>
</html>

The two workers use a common library for performing the actual network calls. This library is as follows:

function get(url) {
  try {
    var xhr = new XMLHttpRequest();
    xhr.open('GET', url, false);
    xhr.send();
    return xhr.responseText;
  } catch (e) {
    return ''; // turn all errors into empty results
  }
}

The stock updater worker is as follows:

importScripts('io.js');
var timer;
var symbol;
function update() {
  postMessage(symbol + ' ' + get('stock.cgi?' + symbol));
  timer = setTimeout(update, 10000);
}
onmessage = function (event) {
  if (timer)
    clearTimeout(timer);
  symbol = event.data;
  update();
};

The search query worker is as follows:

importScripts('io.js');
onmessage = function (event) {
  postMessage(get('search.cgi?' + event.data));
};

View this example online.

1.1.2.4 Shared workers

This section is non-normative.

In this example, multiple windows (viewers) can be opened that are all viewing the same map. All the windows share the same map information, with a single worker coordinating all the viewers. Each viewer can move around independently, but if they set any data on the map, all the viewers are updated.

The main page isn't interesting, it merely provides a way to open the viewers:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Workers example: Multiviewer</title>
  <script>
   function openViewer() {
     window.open('viewer.html');
   }
  </script>
 </head>
 <body>
  <p><button type=button onclick="openViewer()">Open a new
  viewer</button></p>
  <p>Each viewer opens in a new window. You can have as many viewers
  as you like, they all view the same data.</p>
 </body>
</html>

The viewer is more involved:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Workers example: Multiviewer viewer</title>
  <script>
   var worker = new SharedWorker('worker.js', 'core');

   // CONFIGURATION
   function configure(event) {
     if (event.data.substr(0, 4) != 'cfg ') return;
     var name = event.data.substr(4).split(' ', 1);
     // update display to mention our name is name
     document.getElementsByTagName('h1')[0].textContent += ' ' + name;
     // no longer need this listener
     worker.port.removeEventListener('message', configure, false);
   }
   worker.port.addEventListener('message', configure, false);

   // MAP
   function paintMap(event) {
     if (event.data.substr(0, 4) != 'map ') return;
     var data = event.data.substr(4).split(',');
     // display tiles data[0] .. data[8]
     var canvas = document.getElementById('map');
     var context = canvas.getContext('2d');
     for (var y = 0; y < 3; y += 1) {
       for (var x = 0; x < 3; x += 1) {
         var tile = data[y * 3 + x];
         if (tile == '0')
           context.fillStyle = 'green';
         else 
           context.fillStyle = 'maroon';
         fillRect(x * 50, y * 50, 50, 50);
       }
     }
   }
   worker.port.addEventListener('message', paintMap, false);

   // PUBLIC CHAT
   function updatePublicChat(event) {
     if (event.data.substr(0, 4) != 'txt ') return;
     var name = event.data.substr(4).split(' ', 1);
     var message = event.data.substr(4 + length(name) + 1);
     // display "<name> message" in public chat
     var dialog = document.getElementById('public');
     var dt = document.createElement('dt');
     dt.textContent = name;
     dialog.appendChild(dt);
     var dd = document.createElement('dd');
     dd.textContent = message;
     dialog.appendChild(dd);
   }
   worker.port.addEventListener('message', updatePublicChat, false);

   // PRIVATE CHAT
   function startPrivateChat(event) {
     if (event.data.substr(0, 4) != 'msg ') return;
     var name = event.data.substr(4).split(' ', 1);
     var port = event.ports[0];
     // display a private chat UI
     var ul = document.getElementById('private');
     var li = document.createElement('li');
     var h3 = document.createElement('h3');
     h3.textContent = 'Private chat with ' + name;
     li.appendChild(h3);
     var dialog = document.createElement('dialog');
     var addMessage = function(name, message) {
       var dt = document.createElement('dt');
       dt.textContent = name;
       dialog.appendChild(dt);
       var dd = document.createElement('dd');
       dd.textContent = message;
       dialog.appendChild(dd);
     };
     port.onmessage = function (event) {
       addMessage(name, event.data);
     };
     li.appendChild(dialog);
     var form = document.createElement('form');
     var p = document.createElement('p');
     var input = document.createElement('input');
     input.size = 50;
     p.appendChild(input);
     p.appendChild(document.createTextNode(' '));
     var button = document.createElement('button');
     button.textContent = 'Post';
     p.appendChild(button);
     form.onsubmit = function () {
       port.postMessage(input.value);
       addMessage('me', input.value);
       input.value = '';
       return false;
     };
     form.appendChild(p);
     li.appendChild(form);
   }
   worker.port.addEventListener('message', startPrivateChat, false);

   worker.port.start();
  </script>
 </head>
 <body>
  <h1>Viewer</h1>
  <h2>Map</h2>
  <p><canvas id="map" height=150 width=150></canvas></p>
  <p>
   <button type=button onclick="worker.port.postMessage('mov left')">Left</button>
   <button type=button onclick="worker.port.postMessage('mov up')">Up</button>
   <button type=button onclick="worker.port.postMessage('mov down')">Down</button>
   <button type=button onclick="worker.port.postMessage('mov right')">Right</button>
   <button type=button onclick="worker.port.postMessage('set 0')">Set 0</button>
   <button type=button onclick="worker.port.postMessage('set 1')">Set 1</button>
  </p>
  <h2>Public Chat</h2>
  <dialog id="public"></dialog>
  <form onsubmit="worker.port.postMessage('txt ' + message.value); message.value = ''; return false;">
   <p>
    <input type="text" name="message" size="50">
    <button>Post</button>
   </p>
  </form>
  <h2>Private Chat</h2>
  <ul id="private"></ul>
 </body>
</html>

There are several key things worth noting about the way the viewer is written.

Multiple listeners. Instead of a single message processing function, the code here attaches multiple event listeners, each one performing a quick check to see if it is relevant for the message. In this example it doesn't make much difference, but if multiple authors wanted to collaborate using a single port to communicate with a worker, it would allow for independent code instead of changes having to all be made to a single event handling function.

Registering event listeners in this way also allows you to unregister specific listeners when you are done with them, as is done with the configure() method in this example.

Finally, the worker:

var nextName = 0;
function getNextName() {
  // this could use more friendly names
  // but for now just return a number
  return nextName++;
}

var map = [
 [0, 0, 0, 0, 0, 0, 0],
 [1, 1, 0, 1, 0, 1, 1],
 [0, 1, 0, 1, 0, 0, 0],
 [0, 1, 0, 1, 0, 1, 1],
 [0, 0, 0, 1, 0, 0, 0],
 [1, 0, 0, 1, 1, 1, 1],
 [1, 1, 0, 1, 1, 0, 1],
];

function wrapX(x) {
  if (x < 0) return wrapX(x + map[0].length);
  if (x >= map[0].length) return wrapX(x - map[0].length);
  return x;
}

function wrapY(y) {
  if (y < 0) return wrapY(y + map.length);
  if (y >= map[0].length) return wrapY(y - map.length);
  return y;
}

function sendMapData(callback) {
  var data = '';
  for (var y = viewer.y-1; y <= viewer.y+1; y += 1) {
    for (var x = viewer.x-1; x <= viewer.x+1; x += 1) {
      if (data != '')
        data += ',';
      data += map[y][x];
    }
  }
  callback('map ' + data);
}

var viewers = {};
onconnect = function (event) {
  event.ports[0]._name = getNextName();
  event.ports[0]._data = { port: event.port, x: 0, y: 0, };
  viewers[event.ports[0]._name] = event.port._data;
  event.ports[0].postMessage('cfg ' + name);
  event.ports[0].onmessage = getMessage;
  sendMapData(event.ports[0].postMessage);
};

function getMessage(event) {
  switch (event.data.substr(0, 4)) {
    case 'mov ':
      var direction = event.data.substr(4);
      var dx = 0;
      var dy = 0;
      switch (direction) {
        case 'up': dy = -1; break;
        case 'down': dy = 1; break;
        case 'left': dx = -1; break;
        case 'right': dx = 1; break;
      }
      event.target._data.x = wrapX(event.target._data.x + dx);
      event.target._data.y = wrapY(event.target._data.y + dy);
      sendMapData(event.target.postMessage);
      break;
    case 'set ':
      var value = event.data.substr(4);
      map[event.target._data.y][event.target._data.x] = value;
      for (var viewer in viewers)
        sendMapData(viewers[viewer].port.postMessage);
      break;
    case 'txt ':
      var name = event.target._name;
      var message = event.data.substr(4);
      for (var viewer in viewers)
        viewers[viewer].port.postMessage('txt ' + name + ' ' + message);
      break;
    case 'msg ':
      var party1 = event._data;
      var party2 = viewers[event.data.substr(4).split(' ', 1)];
      if (party2) {
        var channel = new MessageChannel();
        party1.port.postMessage('msg ' + party2.name, [channel.port1]);
        party2.port.postMessage('msg ' + party1.name, [channel.port2]);
      }
      break;
  }
}

Connecting to multiple pages. The script uses the onconnect event listener to listen for multiple connections.

Direct channels. When the worker receives a "msg" message from one viewer naming another viewer, it sets up a direct connection between the two, so that the two viewers can communicate directly without the worker having to proxy all the messages.

View this example online.

1.1.2.5 Delegation

This section is non-normative.

With multicore CPUs becoming prevalent, one way to obtain better performance is to split computationally expensive tasks amongst multiple workers. In this example, a computationally expensive task that is to be performed for every number from 1 to 10,000,000 is farmed out to ten subworkers.

The main page is as follows, it just reports the result:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Worker example: Multicore computation</title>
 </head>
 <body>
  <p>Result: <output id="result"></output></p>
  <script>
   var worker = new Worker('worker.js');
   worker.onmessage = function (event) {
     document.getElementById('result').textContent = event.data;
   };
  </script>
 </body>
</html>

The worker itself is as follows:

// settings
var num_workers = 10;
var items_per_worker = 1000000;

// start the workers
var result = 0;
var pending_workers = num_workers;
for (var i = 0; i < num_workers; i += 1) {
  var worker = new Worker('core.js');
  worker.postMessage(i * items_per_worker);
  worker.postMessage((i+1) * items_per_worker);
  worker.onmessage = storeResult;
}

// handle the results
function storeResult(event) {
  result += 1*event.data;
  pending_workers -= 1;
  if (pending_workers <= 0)
    postMessage(result); // finished!
}

It consists of a loop to start the subworkers, and then a handler that waits for all the subworkers to respond.

The subworkers are implemented as follows:

var start;
onmessage = getStart;
function getStart(event) {
  start = 1*event.data;
  onmessage = getEnd;
}

var end;
function getEnd(event) {
  end = 1*event.data;
  onmessage = null;
  work();
}

function work() {
  var result = 0;
  for (var i = start; i < end; i += 1) {
    // perform some complex calculation here
    result += 1;
  }
  postMessage(result);
  close();
}

They receive two numbers in two events, perform the computation for the range of numbers thus specified, and then report the result back to the parent.

View this example online.

1.1.2.6 Providing libraries

This section is non-normative.

Suppose that a cryptography library is made available that provides three tasks:

Generate a public/private key pair
Takes a port, on which it will send two messages, first the public key and then the private key.
Given a plaintext and a public key, return the corresponding cyphertext
Takes a port, to which any number of messages can be sent, the first giving the public key, and the remainder giving the plaintext, each of which is encrypted and then sent on that same channel as the cyphertext. The user can close the port when it is done encrypting content.
Given a cyphertext and a private key, return the corresponding plaintext
Takes a port, to which any number of messages can be sent, the first giving the private key, and the remainder giving the cyphertext, each of which is decrypted and then sent on that same channel as the plaintext. The user can close the port when it is done decrypting content.

The library itself is as follows:

function handleMessage(e) {
  if (e.data == "genkeys")
    genkeys(e.ports[0]);
  else if (e.data == "encrypt")
    encrypt(e.ports[0]);
  else if (e.data == "decrypt")
    decrypt(e.ports[0]);
}

function genkeys(p) {
  var keys = _generateKeyPair();
  p.postMessage(keys[0]);
  p.postMessage(keys[1]);
}

function encrypt(p) {
  var key, state = 0;
  p.onmessage = function (e) {
    if (state == 0) {
      key = e.data;
      state = 1;
    } else {
      p.postMessage(_encrypt(key, e.data));
    }
  };
}

function decrypt(p) {
  var key, state = 0;
  p.onmessage = function (e) {
    if (state == 0) {
      key = e.data;
      state = 1;
    } else {
      p.postMessage(_decrypt(key, e.data));
    }
  };
}

// support being used as a shared worker as well as a dedicated worker
if ('onmessage' in this) // dedicated worker
  onmessage = handleMessage;
else // shared worker
  onconnect = function (e) { e.port.onmessage = handleMessage; }


// the "crypto" functions:

function _generateKeyPair() {
  return [Math.random(), Math.random()];
}

function _encrypt(k, s) {
  return 'encrypted-' + k + ' ' + s;
}

function _decrypt(k, s) {
  return s.substr(s.indexOf(' ')+1);
}

Note that the crypto functions here are just stubs and don't do real cryptography.

This library could be used as follows:

<!DOCTYPE HTML>
<html>
 <head>
  <title>Worker example: Crypto library</title>
  <script>
   var crytoLib = new Worker('libcrypto-v1.js'); // or could use 'libcrypto-v2.js'
   function getKeys() {
     var state = 0;
     cryptoLib.startConversation("genkeys").onmessage = function (e) {
       if (state == 0)
         document.getElementById('public').value = e.data;
       else if (state == 1)
         document.getElementById('private').value = e.data;
       state += 1;
     };
   }
   function enc() {
     var port = cryptoLib.startConversation("encrypt");
     port.postMessage(document.getElementById('public').value);
     port.postMessage(document.getElementById('input').value);
     port.onmessage = function (e) {
       document.getElementById('input').value = e.data;
       port.close();
     };
   }
   function dec() {
     var port = cryptoLib.startConversation("decrypt");
     port.postMessage(document.getElementById('private').value);
     port.postMessage(document.getElementById('input').value);
     port.onmessage = function (e) {
       document.getElementById('input').value = e.data;
       port.close();
     };
   }
  </script>
  <style>
   textarea { display: block; }
  </style>
 </head>
 <body onload="getKeys()">
  <fieldset>
   <legend>Keys</legend>
   <p><label>Public Key: <textarea id="public"></textarea></label></p>
   <p><label>Private Key: <textarea id="private"></textarea></label></p>
  </fieldset>
  <p><label>Input: <textarea id="input"></textarea></label></p>
  <p><button onclick="enc()">Encrypt</button> <button onclick="dec()">Decrypt</button></p>
 </body>
</html>

A later version of the API, though, might want to offload all the crypto work onto subworkers. This could be done as follows:

function handleMessage(e) {
  if (e.data == "genkeys")
    genkeys(e.ports[0]);
  else if (e.data == "encrypt")
    encrypt(e.ports[0]);
  else if (e.data == "decrypt")
    decrypt(e.ports[0]);
}

function genkeys(p) {
  var generator = new Worker('libcrypto-v2-generator.js');
  generator.postMessage('', [p]);
}

function encrypt(p) {
  p.onmessage = function (e) {
    var key = e.data;
    var encryptor = new Worker('libcrypto-v2-encryptor.js');
    encryptor.postMessage(key, [p]);
  };
}

function encrypt(p) {
  p.onmessage = function (e) {
    var key = e.data;
    var decryptor = new Worker('libcrypto-v2-decryptor.js');
    decryptor.postMessage(key, [p]);
  };
}

// support being used as a shared worker as well as a dedicated worker
if ('onmessage' in this) // dedicated worker
  onmessage = handleMessage;
else // shared worker
  onconnect = function (e) { e.ports[0].onmessage = handleMessage };

The little subworkers would then be as follows.

For generating key pairs:

onmessage = function (e) {
  var k = _generateKeyPair();
  e.ports[0].postMessage(k[0]);
  e.ports[0].postMessage(k[1]);
  close();
}

function _generateKeyPair() {
  return [Math.random(), Math.random()];
}

For encrypting:

onmessage = function (e) {
  var key = e.data;
  e.ports[0].onmessage = function (e) {
    var s = e.data;
    postMessage(_encrypt(key, s));
  }
}

function _encrypt(k, s) {
  return 'encrypted-' + k + ' ' + s;
}

For decrypting:

onmessage = function (e) {
  var key = e.data;
  e.ports[0].onmessage = function (e) {
    var s = e.data;
    postMessage(_decrypt(key, s));
  }
}

function _decrypt(k, s) {
  return s.substr(s.indexOf(' ')+1);
}

Notice how the users of the API don't have to even know that this is happening — the API hasn't changed; the library can delegate to subworkers without changing its API, even though it is accepting data using message channels.

View this example online.

1.2 Conformance requirements

All diagrams, examples, and notes in this specification are non-normative, as are all sections explicitly marked non-normative. Everything else in this specification is normative.

The key words "MUST", "MUST NOT", "REQUIRED", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in the normative parts of this document are to be interpreted as described in RFC2119. For readability, these words do not appear in all uppercase letters in this specification. [RFC2119]

Requirements phrased in the imperative as part of algorithms (such as "strip any leading space characters" or "return false and abort these steps") are to be interpreted with the meaning of the key word ("must", "should", "may", etc) used in introducing the algorithm.

Some conformance requirements are phrased as requirements on attributes, methods or objects. Such requirements are to be interpreted as requirements on user agents.

Conformance requirements phrased as algorithms or specific steps may be implemented in any manner, so long as the end result is equivalent. (In particular, the algorithms defined in this specification are intended to be easy to follow, and not intended to be performant.)

The only conformance class defined by this specification is user agents.

User agents may impose implementation-specific limits on otherwise unconstrained inputs, e.g. to prevent denial of service attacks, to guard against running out of memory, or to work around platform-specific limitations.

1.2.1 Dependencies

This specification relies on several other underlying specifications.

HTML5

Many fundamental concepts from HTML5 are used by this specification. [HTML5]

WebIDL

The IDL blocks in this specification use the semantics of the WebIDL specification. [WEBIDL]

1.3 Terminology

The construction "a Foo object", where Foo is actually an interface, is sometimes used instead of the more accurate "an object implementing the interface Foo".

The term DOM is used to refer to the API set made available to scripts in Web applications, and does not necessarily imply the existence of an actual Document object or of any other Node objects as defined in the DOM Core specifications. [DOMCORE]

An IDL attribute is said to be getting when its value is being retrieved (e.g. by author script), and is said to be setting when a new value is assigned to it.

The term "JavaScript" is used to refer to ECMA262, rather than the official term ECMAScript, since the term JavaScript is more widely known. [ECMA262]

1.4 Infrastructure

There are two kinds of workers; dedicated workers, and shared workers. Dedicated workers, once created, and are linked to their creator; but message ports can be used to communicate from a dedicated worker to multiple other browsing contexts or workers. Shared workers, on the other hand, are named, and once created any script running in the same origin can obtain a reference to that worker and communicate with it.

1.4.1 The global scope

The global scope is the "inside" of a worker.

1.4.1.1 The WorkerGlobalScope abstract interface

interface WorkerGlobalScope {
  readonly attribute WorkerGlobalScope self;
  readonly attribute WorkerLocation location;

  void close();
           attribute Function onerror;
};
WorkerGlobalScope implements WorkerUtils;
WorkerGlobalScope implements EventTarget;

The self attribute must return the WorkerGlobalScope object itself.

The location attribute must return the WorkerLocation object created for the WorkerGlobalScope object when the worker was created. It represents the absolute URL of the script that was used to initialize the worker, after any redirects.


When a script invokes the close() method on a WorkerGlobalScope object, the user agent must run the following steps (atomically):

  1. Discard any tasks that have been added to the event loop's task queues.

  2. Set the worker's WorkerGlobalScope object's closing flag to true. (This prevents any further tasks from being queued.)

  3. Disentangle all the ports in the list of the worker's ports.

The following are the event handlers (and their corresponding event handler event types) that must be supported, as IDL attributes, by objects implementing the WorkerGlobalScope interface:

Event handler Event handler event type
onerror error

The WorkerGlobalScope interface must not exist if the interface's relevant namespace object is a Window object. [WEBIDL]

1.4.1.2 Dedicated workers and the DedicatedWorkerGlobalScope interface

[Supplemental, NoInterfaceObject]
interface DedicatedWorkerGlobalScope : WorkerGlobalScope {
  void postMessage(in any message, in optional MessagePortArray ports);
           attribute Function onmessage;
};

DedicatedWorkerGlobalScope objects act as if they had an implicit MessagePort associated with them. This port is part of a channel that is set up when the worker is created, but it is not exposed. This object must never be garbage collected before the DedicatedWorkerGlobalScope object.

All messages received by that port must immediately be retargeted at the DedicatedWorkerGlobalScope object.

The postMessage() method on DedicatedWorkerGlobalScope objects must act as if, when invoked, it immediately invoked the method of the same name on the port, with the same arguments, and returned the same return value.

The following are the event handlers (and their corresponding event handler event types) that must be supported, as IDL attributes, by objects implementing the DedicatedWorkerGlobalScope interface:

Event handler Event handler event type
onmessage message

For the purposes of the application cache networking model, a dedicated worker is an extension of the cache host from which it was created.

1.4.1.3 Shared workers and the SharedWorkerGlobalScope inteface

[Supplemental, NoInterfaceObject]
interface SharedWorkerGlobalScope : WorkerGlobalScope {
  readonly attribute DOMString name;
  readonly attribute ApplicationCache applicationCache;
           attribute Function onconnect;
};

Shared workers receive message ports through connect events on their global object for each connection.

The name attribute must return the value it was assigned when the SharedWorkerGlobalScope object was created by the "run a worker" algorithm. Its value represents the name that can be used to obtain a reference to the worker using the SharedWorker constructor.

The following are the event handlers (and their corresponding event handler event types) that must be supported, as IDL attributes, by objects implementing the SharedWorkerGlobalScope interface:

Event handler Event handler event type
onconnect connect

For the purposes of the application cache networking model, a shared worker is its own cache host. The run a worker algorithm takes care of associating the worker with an application cache.

The applicationCache attribute returns the ApplicationCache object for the worker.

1.4.2 Origins of workers

Both the origin and effective script origin of scripts running in workers are the origin of the absolute URL given in that the worker's location attribute represents.

1.4.3 The event loop

Each WorkerGlobalScope object has an event loop distinct from those defined for units of related similar-origin browsing contexts. This event loop has no associated browsing context, and its task queues only have events, callbacks, and networking activity as tasks. The processing model of these event loops is defined below in the run a worker algorithm.

Each WorkerGlobalScope object also has a closing flag, which must initially be false, but which can get set to true by the algorithms in the processing model section below.

Once the WorkerGlobalScope's closing flag is set to true, the event loop's task queues must discard any further tasks that would be added to them (tasks already on the queue are unaffected except where otherwise specified). Effectively, once the closing flag is true, timers stop firing, notifications for all pending asynchronous operations are dropped, etc.

1.4.4 The worker's lifetime

Workers communicate with other workers and with browsing contexts through message channels and their MessagePort objects.

Each WorkerGlobalScope worker global scope has a list of the worker's ports, which consists of all the MessagePort objects that are entangled with another port and that have one (but only one) port owned by worker global scope. This list includes the implicit MessagePort in the case of dedicated workers.

Each WorkerGlobalScope also has a list of the worker's workers. Initially this list is empty; it is populated when the worker creates or obtains further workers.

Finally, each WorkerGlobalScope also has a list of the worker's Documents. Initially this list is empty; it is populated when the worker is created.

Whenever a Document d is added to the worker's Documents, the user agent must, for each worker in the list of the worker's workers whose list of the worker's Documents does not contain d, add d to q's WorkerGlobalScope owner's list of the worker's Documents.

Whenever a Document object is discarded, it must be removed from the list of the worker's Documents of each worker whose list contains that Document.

Given a script's global object o when creating or obtaining a worker, the list of relevant Document objects to add depends on the type of o. If o is a WorkerGlobalScope object (i.e. if we are creating a nested worker), then the relevant Documents are the Documents that are in o's own list of the worker's Documents. Otherwise, o is a Window object, and the relevant Document is just the Document that is the active document of the Window object o.


A worker is said to be a permissible worker if its list of the worker's Documents is not empty.

A worker is said to be a protected worker if it is a permissible worker and either it has outstanding timers, database transactions, or network connections, or its list of the worker's ports is not empty, or its WorkerGlobalScope is actually a SharedWorkerGlobalScope object (i.e. the worker is a shared worker).

A worker is said to be an active needed worker if any of the Document objects in the worker's Documents are fully active.

A worker is said to be a suspendable worker if it is not an active needed worker but it is a permissible worker.

1.4.5 Processing model

When a user agent is to run a worker for a script with URL url, a browsing context owner browsing context, an origin owner origin, and with global scope worker global scope, it must run the following steps:

  1. Create a completely separate and parallel execution environment (i.e. a separate thread or process or equivalent construct), and run the rest of these steps asynchronously in that context.

  2. If worker global scope is actually a SharedWorkerGlobalScope object (i.e. the worker is a shared worker), and there are any relevant application caches that are identified by a manifest URL with the same origin as url and that have url as one of their entries, not excluding entries marked as foreign, then associate the worker global scope with the most appropriate application cache of those that match.

  3. Attempt to fetch the resource identified by url, from the owner origin.

    If the attempt fails, or if the attempt involves any redirects to URIs that do not have the same origin as url (even if the final URI is at the same origin as the original url), then for each Worker or SharedWorker object associated with worker global scope, queue a task to fire a simple event named error at that object. Abort these steps.

    If the attempt succeeds, then convert the script resource to Unicode by assuming it was encoded as UTF-8, to obtain its source.

    Let language be JavaScript.

    As with script elements, the MIME type of the script is ignored. Unlike with script elements, there is no way to override the type. It's always assumed to be JavaScript.

  4. A new script is now created, as follows.

    Create a new script execution environment set up as appropriate for the scripting language language.

    Parse/compile/initialize source using that script execution environment, as appropriate for language, and thus obtain a list of code entry-points; set the initial code entry-point to the entry-point for any executable code to be immediately run.

    Set the script's global object to worker global scope.

    Set the script's browsing context to owner browsing context.

    Set the script's URL character encoding to UTF-8. (This is just used for encoding non-ASCII characters in the query component of URLs.)

    Set the script's base URL to url.

  5. Closing orphan workers: Start monitoring the worker such that no sooner than it stops being either a protected worker or a suspendable worker, and no later than it stops being a permissible worker, worker global scope's closing flag is set to true.

  6. Suspending workers: Start monitoring the worker, such that whenever worker global scope's closing flag is false and the worker is a suspendable worker, the user agent suspends execution of script in that worker until such time as either the closing flag switches to true or the worker stops being a suspendable worker.

  7. Jump to the script's initial code entry-point, and let that run until it either returns, fails to catch an exception, or gets prematurely aborted by the "kill a worker" or "terminate a worker" algorithms defined below.

  8. If worker global scope is actually a DedicatedWorkerGlobalScope object (i.e. the worker is a dedicated worker), then enable the port message queue of the worker's implicit port.

  9. Event loop: Wait until either there is a task in one of the event loop's task queues or worker global scope's closing flag is set to true.

  10. Run the oldest task on one of the event loop's task queues, if any. The user agent may pick any task queue.

    The handling of events or the execution of callbacks might get prematurely aborted by the "kill a worker" or "terminate a worker" algorithms defined below.

  11. Remove the task run in the previous step, if any, from its task queue.

  12. If there are any more events in the event loop's task queues or if worker global scope's closing flag is set to false, then jump back to the step above labeled event loop.

  13. If there are any outstanding transactions that have callbacks that involve scripts whose global object is the worker global scope, roll them back (without invoking any of the callbacks).

  14. Empty the worker global scope's list of active timeouts and its list of active intervals.


When a user agent is to kill a worker it must run the following steps in parallel with the worker's main loop (the "run a worker" processing model defined above):

  1. Set the worker's WorkerGlobalScope object's closing flag to true.

  2. If there are any tasks queued in the event loop's task queues, discard them without processing them.

  3. Wait a user-agent-defined amount of time.

  4. Abort the script currently running in the worker.

User agents may invoke the "kill a worker" processing model on a worker at any time, e.g. in response to user requests, in response to CPU quota management, or when a worker stops being an active needed worker if the worker continues executing even after its closing flag was set to true.


When a user agent is to terminate a worker it must run the following steps in parallel with the worker's main loop (the "run a worker" processing model defined above):

  1. Set the worker's WorkerGlobalScope object's closing flag to true.

  2. If there are any tasks queued in the event loop's task queues, discard them without processing them.

  3. Abort the script currently running in the worker.

  4. If the worker's WorkerGlobalScope object is actually a DedicatedWorkerGlobalScope object (i.e. the worker is a dedicated worker), then empty the port message queue of the port that the worker's implicit port is entangled with.


The task source for the tasks mentioned above is the DOM manipulation task source.

1.4.6 Runtime script errors

Whenever an uncaught runtime script error occurs in one of the worker's scripts, if the error did not occur while handling a previous script error, the user agent must report the error using the WorkerGlobalScope object's onerror attribute. [HTML5]

For shared workers, if the error is still not handled afterwards, or if the error occurred while handling a previous script error, the error may be reported to the user. [HTML5]

For dedicated workers, if the error is still not handled afterwards, or if the error occurred while handling a previous script error, the user agent must queue a task to fire a worker error event at the Worker object associated with the worker.

When the user agent is to fire a worker error event at a Worker object, it must dispatch an event that uses the ErrorEvent interface, with the name error, that doesn't bubble and is cancelable, with its message, filename, and lineno attributes set appropriately. The default action of this event depends on whether the Worker object is itself in a worker. If it is, and that worker is also a dedicated worker, then the user agent must again queue a task to fire a worker error event at the Worker object associated with that worker. Otherwise, then the error may be reported to the user.

The task source for the tasks mentioned above is the DOM manipulation task source.


interface ErrorEvent : Event {
  readonly attribute DOMString message;
  readonly attribute DOMString filename;
  readonly attribute unsigned long lineno;
  void initErrorEvent(in DOMString typeArg, in boolean canBubbleArg, in boolean cancelableArg, in DOMString messageArg, in DOMString filenameArg, in unsigned long linenoArg);
};

The initErrorEvent() method must initialize the event in a manner analogous to the similarly-named method in the DOM Events interfaces. [DOMEVENTS]

The message attribute represents the error message.

The filename attribute represents the absolute URL of the script in which the error originally occurred.

The lineno attribute represents the line number where the error occurred in the script.

1.4.7 Creating workers

1.4.7.1 The AbstractWorker abstract interface

[Supplemental, NoInterfaceObject]
interface AbstractWorker {
           attribute Function onerror;

};
AbstractWorker implements EventTarget;

The following are the event handlers (and their corresponding event handler event types) that must be supported, as IDL attributes, by objects implementing the AbstractWorker interface:

Event handler Event handler event type
onerror error

1.4.7.2 Dedicated workers and the Worker interface

[Constructor(in DOMString scriptURL)]
interface Worker : AbstractWorker {
  void terminate();

  void postMessage(in any message, in optional MessagePortArray ports);
           attribute Function onmessage;
};

The terminate() method, when invoked, must cause the "terminate a worker" algorithm to be run on the worker with with the object is associated.

Worker objects act as if they had an implicit MessagePort associated with them. This port is part of a channel that is set up when the worker is created, but it is not exposed. This object must never be garbage collected before the Worker object.

All messages received by that port must immediately be retargeted at the Worker object.

The postMessage() method on Worker objects must act as if, when invoked, it immediately invoked the method of the same name on the port, with the same arguments, and returned the same return value.

The following are the event handlers (and their corresponding event handler event types) that must be supported, as IDL attributes, by objects implementing the Worker interface:

Event handler Event handler event type
onmessage message

When the Worker(scriptURL) constructor is invoked, the user agent must run the following steps:

  1. Resolve the scriptURL argument relative to the entry script's base URL, when the method is invoked.

  2. If this fails, throw a SYNTAX_ERR exception.

  3. If the origin of the resulting absolute URL is not the same as the origin of the entry script, then throw a SECURITY_ERR exception.

    Thus, scripts must be external files with the same scheme as the original page: you can't load a script from a data: URL or javascript: URL, and a https: page couldn't start workers using scripts with http: URLs.

  4. Create a new DedicatedWorkerGlobalScope object. Let worker global scope be this new object.

  5. Create a new Worker object, associated with worker global scope. Let worker be this new object.

  6. Create a new MessagePort object owned by the global object of the script that invoked the constructor. Let this be the outside port.

  7. Associate the outside port with worker.

  8. Create a new MessagePort object owned by worker global scope. Let inside port be this new object.

  9. Associate inside port with worker global scope.

  10. Entangle outside port and inside port.

  11. Return worker, and run the following steps asynchronously.

  12. Enable outside port's port message queue.

  13. Let docs be the list of relevant Document objects to add given the global object of the script that invoked the constructor.

  14. Add to worker global scope's list of the worker's Documents the Document objects in docs.

  15. If the global object of the script that invoked the constructor is a WorkerGlobalScope object (i.e. we are creating a nested worker), add worker global scope to the list of the worker's workers of the WorkerGlobalScope object that is the global object of the script that invoked the constructor.

  16. Run a worker for the resulting absolute URL, with the script's browsing context of the script that invoked the method as the owner browsing context, with the origin of the entry script as the owner origin, and with worker global scope as the global scope.

This constructor must be visible when the script's global object is either a Window object or an object implementing the WorkerUtils interface.

1.4.7.3 Shared workers and the SharedWorker interface

[Constructor(in DOMString scriptURL, in optional DOMString name)]
interface SharedWorker : AbstractWorker {
  readonly attribute MessagePort port;
};

The port attribute must return the value it was assigned by the object's constructor. It represents the MessagePort for communicating with the shared worker.

When the SharedWorker(scriptURL, name) constructor is invoked, the user agent must run the following steps:

  1. Resolve the scriptURL argument.

  2. If this fails, throw a SYNTAX_ERR exception.

  3. Otherwise, let scriptURL be the resulting absolute URL.

  4. Let name be the value of the second argument, or the empty string if the second argument was omitted.

  5. If the origin of scriptURL is not the same as the origin of the entry script, then throw a SECURITY_ERR exception.

    Thus, scripts must be external files with the same scheme as the original page: you can't load a script from a data: URL or javascript: URL, and a https: page couldn't start workers using scripts with http: URLs.

  6. Let docs be the list of relevant Document objects to add given the global object of the script that invoked the constructor.

  7. Execute the following substeps atomically:

    1. Create a new SharedWorker object, which will shortly be associated with a SharedWorkerGlobalScope object. Let this SharedWorker object be worker.

    2. Create a new MessagePort object owned by the global object of the script that invoked the method. Let this be the outside port.

    3. Assign outside port to the port attribute of worker.

    4. Let worker global scope be null.

    5. If name is not the empty string and there exists a SharedWorkerGlobalScope object whose closing flag is false, whose name attribute is exactly equal to name, and whose location attribute represents an absolute URL with the same origin as scriptURL, then let worker global scope be that SharedWorkerGlobalScope object.

    6. Otherwise, if name is the empty string and there exists a SharedWorkerGlobalScope object whose closing flag is false, and whose location attribute is exactly equal to scriptURL, then let worker global scope be that SharedWorkerGlobalScope object.

    7. If worker global scope is not null, then run these steps:

      1. If worker global scope's location attribute represents an absolute URL that is not exactly equal to scriptURL, then throw a URL_MISMATCH_ERR exception and abort all these steps.

      2. Associate worker with worker global scope.

      3. Create a new MessagePort object owned by worker global scope. Let this be the inside port.

      4. Entangle outside port and inside port.

      5. Return worker and perform the next step asynchronously.

      6. Create an event that uses the MessageEvent interface, with the name connect, which does not bubble, is not cancelable, has no default action, has a data attribute whose value is the empty string and has a ports attribute whose value is an array containing only the newly created port, and queue a task to dispatch the event at worker global scope.

      7. Add to worker global scope's list of the worker's Documents the Document objects in docs.

      8. If the global object of the script that invoked the constructor is a WorkerGlobalScope object, add worker global scope to the list of the worker's workers of the WorkerGlobalScope object that is the global object of the script that invoked the constructor.

      9. Abort all these steps.

    8. Create a new SharedWorkerGlobalScope object. Let worker global scope be this new object.

    9. Associate worker with worker global scope.

    10. Set the name attribute of worker global scope to name.

    11. Create a new MessagePort object owned by worker global scope. Let inside port be this new object.

    12. Entangle outside port and inside port.

  8. Return worker and perform the remaining steps asynchronously.

  9. Create an event that uses the MessageEvent interface, with the name connect, which does not bubble, is not cancelable, has no default action, has a data attribute whose value is the empty string and has a ports attribute whose value is an array containing only the newly created port, and queue a task to dispatch the event at worker global scope.

  10. Add to worker global scope's list of the worker's Documents the Document objects in docs.

  11. If the global object of the script that invoked the constructor is a WorkerGlobalScope object, add worker global scope to the list of the worker's workers of the WorkerGlobalScope object that is the global object of the script that invoked the constructor.

  12. Run a worker for scriptURL, with the script's browsing context of the script that invoked the method as the owner browsing context, with the origin of the entry script as the owner origin, and with worker global scope as the global scope.

This constructor must be visible when the script's global object is either a Window object or an object implementing the WorkerUtils interface.

The task source for the tasks mentioned above is the DOM manipulation task source.

1.5 APIs available to workers

[Supplemental, NoInterfaceObject]
interface WorkerUtils {
  void importScripts(in DOMString... urls);
  readonly attribute WorkerNavigator navigator;
};
WorkerUtils implements WindowTimers;

The DOM APIs (Node objects, Document objects, etc) are not available to workers in this version of this specification.

1.5.1 Importing scripts and libraries

When a script invokes the importScripts(urls) method on a WorkerGlobalScope object, the user agent must run the following steps:

  1. If there are no arguments, return without doing anything. Abort these steps.

  2. Resolve each argument.

  3. If any fail, throw a SYNTAX_ERR exception.

  4. Attempt to fetch each resource identified by the resulting absolute URLs, from the entry script's origin.

  5. For each argument in turn, in the order given, starting with the first one, run these substeps:

    1. Wait for the fetching attempt for the corresponding resource to complete.

      If the fetching attempt failed, throw a NETWORK_ERR exception and abort all these steps.

      If the attempt succeeds, then convert the script resource to Unicode by assuming it was encoded as UTF-8, to obtain its source.

      Let language be JavaScript.

      As with the worker's script, the script here is always assumed to be JavaScript, regardless of the MIME type.

    2. Create a script, using source as the script source and language as the scripting language, using the same global object, browsing context, URL character encoding, base URL, and script group as the script that was created by the worker's run a worker algorithm.

      Let the newly created script run until it either returns, fails to parse, fails to catch an exception, or gets prematurely aborted by the "kill a worker" or "terminate a worker" algorithms defined above.

      If it failed to parse, then throw an ECMAScript SyntaxError exception and abort all these steps. [ECMA262]

      If an exception was raised or if the script was prematurely aborted, then abort all these steps, letting the exception or aborting continue to be processed by the script that called the importScripts() method.

      If the "kill a worker" or "terminate a worker" algorithms abort the script then abort all these steps.

1.5.2 The WorkerNavigator object

The navigator attribute of the WorkerUtils interface must return an instance of the WorkerNavigator interface, which represents the identity and state of the user agent (the client):

interface WorkerNavigator {};
WorkerNavigator implements NavigatorID;
WorkerNavigator implements NavigatorOnLine;

Objects implementing the WorkerNavigator interface also implement the NavigatorID and NavigatorOnLine interfaces. [HTML5]

This WorkerNavigator interface must not exist if the interface's relevant namespace object is a Window object. [WEBIDL]

1.5.3 APIs defined in other specifications

The openDatabase() and openDatabaseSync() methods are defined in the Web SQL Database specification. [WEBSQL]

1.5.4 Interface objects and constructors

There must be no interface objects and constructors available in the global scope of scripts whose script's global object is a WorkerGlobalScope object except for the following:

These requirements do not override the requirements defined by the Web IDL specification, in particular concerning the visibility of interfaces annotated with the [NoInterfaceObject] extended attribute.

1.5.5 Worker locations

interface WorkerLocation {
  readonly attribute DOMString href;
  readonly attribute DOMString protocol;
  readonly attribute DOMString host;
  readonly attribute DOMString hostname;
  readonly attribute DOMString port;
  readonly attribute DOMString pathname;
  readonly attribute DOMString search;
  readonly attribute DOMString hash;
};

A WorkerLocation object represents an absolute URL set at its creation.

The href attribute must return the absolute URL that the object represents.

The WorkerLocation interface also has the complement of URL decomposition IDL attributes, protocol, host, port, hostname, pathname, search, and hash. These must follow the rules given for URL decomposition IDL attributes, with the input being the absolute URL that the object represents (same as the href attribute), and the common setter action being a no-op, since the attributes are defined to be readonly. [HTML5]

The WorkerLocation interface must not exist if the interface's relevant namespace object is a Window object. [WEBIDL]

References

All references are normative unless marked "Non-normative".

[DOMCORE]
Document Object Model (DOM) Level 3 Core Specification, A. Le Hors, P. Le Hegaret, L. Wood, G. Nicol, J. Robie, M. Champion, S. Byrnes. W3C, April 2004.
[DOMEVENTS]
Document Object Model (DOM) Level 3 Events Specification, D. Schepers. W3C, July 2009.
[ECMA262]
ECMAScript Language Specification. ECMA, December 2009.
[HTML5]
HTML5, I. Hickson. WHATWG, August 2009.
[RFC2119]
Key words for use in RFCs to Indicate Requirement Levels, S. Bradner. IETF, March 1997.
[WEBSQL]
Web SQL Database, I. Hickson. W3C, July 2009.
[WEBIDL]
Web IDL, C. McCormack. W3C, July 2009.
[XHR]
XMLHttpRequest, A. van Kesteren. W3C, June 2009.

Acknowledgements

Thanks to Aaron Boodman, Алексей Проскуряков (Alexey Proskuryakov), Anne van Kesteren, Ben Turner, Dmitry Titov, Drew Wilson, Jeremy Orlow, Jonas Sicking, Justin James, Kevin Hakanson, Maciej Stachowiak, Michael Nordman, Mike Smith, and Philip Taylor for their useful and substantial comments.

Huge thanks to the whole Gears team, who pioneered this technology and whose experience has been a huge influence on this specification.