workerpool

workerpool offers an easy way to create a pool of workers for both dynamically offloading computations as well as managing a pool of dedicated workers. workerpool basically implements a thread pool pattern. There is a pool of workers to execute tasks. New tasks are put in a queue. A worker executes one task at a time, and once finished, picks a new task from the queue. Workers can be accessed via a natural, promise based proxy, as if they are available straight in the main application.

workerpool runs on node.js, Chrome, Firefox, Opera, Safari, and IE10+.

Features

• Easy to use
• Runs in the browser and on node.js
• Dynamically offload functions to a worker
• Access workers via a proxy
• Cancel running tasks
• Set a timeout on tasks
• Handles crashed workers
• Small: 5 kB minified and gzipped

Why

JavaScript is based upon a single event loop which handles one event at a time. Jeremy Epstein explains this clearly:

In Node.js everything runs in parallel, except your code. What this means is that all I/O code that you write in Node.js is non-blocking, while (conversely) all non-I/O code that you write in Node.js is blocking.

This means that CPU heavy tasks will block other tasks from being executed. In case of a browser environment, the browser will not react to user events like a mouse click while executing a CPU intensive task (the browser "hangs"). In case of a node.js server, the server will not respond to any new request while executing a single, heavy request.

For front-end processes, this is not a desired situation. Therefore, CPU intensive tasks should be offloaded from the main event loop onto dedicated workers. In a browser environment, Web Workers can be used. In node.js, child processes and worker_threads are available. An application should be split in separate, decoupled parts, which can run independent of each other in a parallelized way. Effectively, this results in an architecture which achieves concurrency by means of isolated processes and message passing.

Install

Install via npm:

npm install workerpool

Load

To load workerpool in a node.js application (both main application as well as workers):

1const workerpool = require('workerpool');

To load workerpool in the browser:

1<script src="workerpool.js"></script>

To load workerpool in a web worker in the browser:

1importScripts('workerpool.js');

Use

Offload functions dynamically

In the following example there is a function add, which is offloaded dynamically to a worker to be executed for a given set of arguments.

myApp.js

1const workerpool = require('workerpool');
2const pool = workerpool.pool();
3
4function add(a, b) {
5  return a + b;
6}
7
8pool.exec(add, [3, 4])
9    .then(function (result) {
10      console.log('result', result); // outputs 7
11    })
12    .catch(function (err) {
13      console.error(err);
14    })
15    .then(function () {
16      pool.terminate(); // terminate all workers when done
17    });

Note that both function and arguments must be static and stringifiable, as they need to be sent to the worker in a serialized form. In case of large functions or function arguments, the overhead of sending the data to the worker can be significant.

Dedicated workers

A dedicated worker can be created in a separate script, and then used via a worker pool.

myWorker.js

1const workerpool = require('workerpool');
2
3// a deliberately inefficient implementation of the fibonacci sequence
4function fibonacci(n) {
5  if (n < 2) return n;
6  return fibonacci(n - 2) + fibonacci(n - 1);
7}
8
9// create a worker and register public functions
10workerpool.worker({
11  fibonacci: fibonacci
12});

This worker can be used by a worker pool:

myApp.js

1const workerpool = require('workerpool');
2
3// create a worker pool using an external worker script
4const pool = workerpool.pool(__dirname + '/myWorker.js');
5
6// run registered functions on the worker via exec
7pool.exec('fibonacci', [10])
8    .then(function (result) {
9      console.log('Result: ' + result); // outputs 55
10    })
11    .catch(function (err) {
12      console.error(err);
13    })
14    .then(function () {
15      pool.terminate(); // terminate all workers when done
16    });
17
18// or run registered functions on the worker via a proxy:
19pool.proxy()
20    .then(function (worker) {
21      return worker.fibonacci(10);
22    })
23    .then(function (result) {
24      console.log('Result: ' + result); // outputs 55
25    })
26    .catch(function (err) {
27      console.error(err);
28    })
29    .then(function () {
30      pool.terminate(); // terminate all workers when done
31    });

Worker can also initialize asynchronously:

myAsyncWorker.js

1define(['workerpool/dist/workerpool'], function(workerpool) {
2
3  // a deliberately inefficient implementation of the fibonacci sequence
4  function fibonacci(n) {
5    if (n < 2) return n;
6    return fibonacci(n - 2) + fibonacci(n - 1);
7  }
8
9  // create a worker and register public functions
10  workerpool.worker({
11    fibonacci: fibonacci
12  });
13
14});

Examples

Examples are available in the examples directory:

https://github.com/josdejong/workerpool/tree/master/examples

API

The API of workerpool consists of two parts: a function workerpool.pool to create a worker pool, and a function workerpool.worker to create a worker.

pool

A workerpool can be created using the function workerpool.pool:

workerpool.pool([script: string] [, options: Object]) : Pool

When a script argument is provided, the provided script will be started as a dedicated worker. When no script argument is provided, a default worker is started which can be used to offload functions dynamically via Pool.exec. Note that on node.js, script must be an absolute file path like __dirname + '/myWorker.js'. In a browser environment, script can also be a data URL like 'data:application/javascript;base64,...'. This allows embedding the bundled code of a worker in your main application. See examples/embeddedWorker for a demo.

The following options are available:

• minWorkers: number | 'max'. The minimum number of workers that must be initialized and kept available. Setting this to 'max' will create maxWorkers default workers (see below).
• maxWorkers: number. The default number of maxWorkers is the number of CPU's minus one. When the number of CPU's could not be determined (for example in older browsers), maxWorkers is set to 3.
• maxQueueSize: number. The maximum number of tasks allowed to be queued. Can be used to prevent running out of memory. If the maximum is exceeded, adding a new task will throw an error. The default value is Infinity.
• workerType: 'auto' | 'web' | 'process' | 'thread'.
  • In case of 'auto' (default), workerpool will automatically pick a suitable type of worker: when in a browser environment, 'web' will be used. When in a node.js environment, worker_threads will be used if available (Node.js >= 11.7.0), else child_process will be used.
  • In case of 'web', a Web Worker will be used. Only available in a browser environment.
  • In case of 'process', child_process will be used. Only available in a node.js environment.
  • In case of 'thread', worker_threads will be used. If worker_threads are not available, an error is thrown. Only available in a node.js environment.

Important note on 'workerType': when sending and receiving primitive data types (plain JSON) from and to a worker, the different worker types ('web', 'process', 'thread') can be used interchangeably. However, when using more advanced data types like buffers, the API and returned results can vary. In these cases, it is best not to use the 'auto' setting but have a fixed 'workerType' and good unit testing in place.

A worker pool contains the following functions:

• Pool.exec(method: Function | string, params: Array | null) : Promise.<*, Error>
  Execute a function on a worker with given arguments.

  • When method is a string, a method with this name must exist at the worker and must be registered to make it accessible via the pool. The function will be executed on the worker with given parameters.
  • When method is a function, the provided function fn will be stringified, send to the worker, and executed there with the provided parameters. The provided function must be static, it must not depend on variables in a surrounding scope.

• Pool.proxy() : Promise.<Object, Error>
  Create a proxy for the worker pool. The proxy contains a proxy for all methods available on the worker. All methods return promises resolving the methods result.

• Pool.stats() : Object
  Retrieve statistics on workers, and active and pending tasks.

  Returns an object containing the following properties:

  {
    totalWorkers: 0,
    busyWorkers: 0,
    idleWorkers: 0,
    pendingTasks: 0,
    activeTasks: 0
  }
  

• Pool.terminate([force: boolean [, timeout: number]])

  If parameter force is false (default), workers will finish the tasks they are working on before terminating themselves. Any pending tasks will be rejected with an error 'Pool terminated'. When force is true, all workers are terminated immediately without finishing running tasks. If timeout is provided, worker will be forced to terminate when the timeout expires and the worker has not finished.

The function Pool.exec and the proxy functions all return a Promise. The promise has the following functions available:

• Promise.then(fn: Function.<result: *>)
  Get the result of the promise once resolve.
• Promise.catch(fn: Function.<error: Error>)
  Get the error of the promise when rejected.
• Promise.cancel()
  A running task can be cancelled. The worker executing the task is enforced to terminate immediately. The promise will be rejected with a Promise.CancellationError.
• Promise.timeout(delay: number)
  Cancel a running task when it is not resolved or rejected within given delay in milliseconds. The timer will start when the task is actually started, not when the task is created and queued. The worker executing the task is enforced to terminate immediately. The promise will be rejected with a Promise.TimeoutError.

Example usage:

1const workerpool = require('workerpool');
2
3function add(a, b) {
4  return a + b;
5}
6
7const pool1 = workerpool.pool();
8
9// offload a function to a worker
10pool1.exec(add, [2, 4])
11    .then(function (result) {
12      console.log(result); // will output 6
13    })
14    .catch(function (err) {
15      console.error(err);
16    });
17
18// create a dedicated worker
19const pool2 = workerpool.pool(__dirname + '/myWorker.js');
20
21// supposed myWorker.js contains a function 'fibonacci'
22pool2.exec('fibonacci', [10])
23    .then(function (result) {
24      console.log(result); // will output 55
25    })
26    .catch(function (err) {
27      console.error(err);
28    });
29
30// create a proxy to myWorker.js
31pool2.proxy()
32    .then(function (myWorker) {
33      return myWorker.fibonacci(10)
34    })
35    .then(function (result) {
36      console.log(result); // will output 55
37    })
38    .catch(function (err) {
39      console.error(err);
40    });
41
42// create a pool with a specified maximum number of workers
43const pool3 = workerpool.pool({maxWorkers: 7});

worker

A worker is constructed as:

workerpool.worker([methods: Object.<String, Function>])

Argument methods is optional can can be an object with functions available in the worker. Registered functions will be available via the worker pool.

Example usage:

1// file myWorker.js
2const workerpool = require('workerpool');
3
4function add(a, b) {
5  return a + b;
6}
7
8function multiply(a, b) {
9  return a * b;
10}
11
12// create a worker and register functions
13workerpool.worker({
14  add: add,
15  multiply: multiply
16});

Asynchronous results can be handled by returning a Promise from a function in the worker:

1// file myWorker.js
2const workerpool = require('workerpool');
3
4function timeout(delay) {
5  return new Promise(function (resolve, reject) {
6    setTimeout(resolve, delay)
7  });
8}
9
10// create a worker and register functions
11workerpool.worker({
12  timeout: timeout
13});

Utilities

Following properties are available for convenience:

• platform: The Javascript platform. Either node or browser
• isMainThread: Whether the code is running in main thread or not (Workers)
• cpus: The number of CPUs/cores available

Roadmap

• Implement functions for parallel processing: map, reduce, forEach, filter, some, every, ...
• Implement graceful degradation on old browsers not supporting webworkers: fallback to processing tasks in the main application.
• Implement session support: be able to handle a series of related tasks by a single worker, which can keep a state for the session.

Related libraries

• https://github.com/learnboost/cluster
• https://github.com/adambom/parallel.js
• https://github.com/padolsey/operative
• https://github.com/calvinmetcalf/catiline
• https://github.com/Unitech/pm2
• https://github.com/godaddy/node-cluster-service
• https://github.com/ramesaliyev/EasyWebWorker
• https://github.com/rvagg/node-worker-farm

Build

First clone the project from github:

git clone git://github.com/josdejong/workerpool.git
cd workerpool

Install the project dependencies:

npm install

Then, the project can be build by executing the build script via npm:

npm run build

This will build the library workerpool.js and workerpool.min.js from the source files and put them in the folder dist.

Test

To execute tests for the library, install the project dependencies once:

npm install

Then, the tests can be executed:

npm test

To test code coverage of the tests:

npm run coverage

To see the coverage results, open the generated report in your browser:

./coverage/lcov-report/index.html

Publish

• Describe changes in HISTORY.md
• Update version in package.json, run npm install to update it in package-lock.json too.
• Push to github
• Deploy to npm via npm publish
• Add a git tag with the version number like:

  git tag v1.2.3
  git push --tags
  

License

Copyright (C) 2014-2020 Jos de Jong wjosdejong@gmail.com

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.