concurrent.js

Non-blocking Concurrent Computation for JavaScript RTEs (Web Browsers, Node.js & Deno & Bun)
GitHub
384
Created 2 years ago, last commit 10 months ago
Number of contributors not available
22 commits
Stars added on GitHub, month by month
0
0
12
1
2
3
4
5
6
7
8
9
10
11
2023
2024
Stars added on GitHub, per day, on average
Yesterday
=
Last week
0.0
/day
Last month
0.0
/day
Last 12 months
+0.1
/day
npmPackage on NPM
Monthly downloads on NPM
12
1
2
3
4
5
6
7
8
9
10
11
2023
2024
No dependencies
README

Concurrent.js

Change Log | Star History | Community

Concurrent.js is a library that enables non-blocking computation on JavaScript RTEs by dynamically loading a module into a background thread.

Features

  • Sharing workers
  • Parallel execution
  • Reactive concurrency
  • Isolation

Technical Facts

  • Built upon web workers (a.k.a. worker threads).
  • Simplifies the complexity of worker usage by providing a minimal API.
  • Automatically creates and terminates workers.
  • Automatically cleans up a worker's memory.
  • Has no third-party runtime dependency.
  • Written in TypeScript with the strictest ESNext config.
  • Strictly designed to support strongly-typed programming.
  • Packaged as platform-specific bundles that target ES2020.

Hello World!

Save and run the hello world script to see Concurrent.js in action:

bash hello_world.sh

Installation

npm i @bitair/concurrent.js

Usage

At its highest level of design, Concurrent.js is a dynamic module importer that loads a module into a web worker:

import { concurrent } from '@bitair/concurrent.js'
// In Deno
// import { concurrent } from 'https://deno.land/x/concurrentjs@v0.8.2/mod.ts'

// Import a module
const MyModule = concurrent.import(new URL('./sample_module.js', import.meta.url))
// In a CommonJS module
// const MyModule = concurrent.import(path.join(__dirname, 'sample_module.js'))

// Load it into a web worker
const { SampleObject, sampleFunction } = await MyModule.load()
// Load it into another web worker 
// const { SampleObject: SampleObject2, sampleFunction: sampleFunction2 } = await MyModule.load()

// Run a function
const result = await sampleFunction(/*...args*/)

// Run a class (instance members)
const obj = await new SampleObject(/*...args*/) // Instantiate
const value = await obj.sampleProp // Get a field or getter
await ((obj.sampleProp = 1), obj.sampleProp) // Set a field or setter
const result = await obj.sampleMethod(/*...args*/) // Call a method

// Run a class (static members)
const value = await SampleObject.sampleStaticProp // Get a static field or getter
await ((SampleObject.sampleStaticProp = 1), SampleObject.sampleStaticProp) // Set a static field or setter
const result = await SampleObject.sampleStaticMethod(/*...args*/) // Call a static method

// Terminate Concurrent.js
await concurrent.terminate()

Samples

Benchmark

The following results demonstrate the average execution time and CPU usage of running 10 concurrent calculations (10 iterations) of the factorial of 50,000 on various JavaScript runtime environments (RTEs). These calculations were performed on a Quad-core AMD APU with a base clock rate of 2.2GHz within a freshly installed isolated Ubuntu VM.

(There are 213,237 digits in the factorial of 50,000)

RTE JS Engine Execution Time CPU Usage
1 Deno (v1.40) V8 7.9168s 100%
2 Chrome* (v121.0) V8 7.919s 100%
3 Node (v20.11) V8 8.117s 100%
4 Servo (v0.0.1-c94d584) SpiderMonkey 31.267s 99%
5 LibreWolf (122.0) SpiderMonkey 35.417s 92%
6 Firefox* (v125.0) SpiderMonkey 49.061s 95%
7 Bun (v1.0.26) JavaScriptCore 51.502s 99%
8 GNOME Web (v45.2) JavaScriptCore 59.058s 75%
  • A headless environment was used for benchmarking.

To benchmark Node, Deno, Bun RTEs as well as Chrome and Firefox browsers use the benchmarking app:

git clone https://github.com/bitair-org/concurrent.js.git
cd concurrent.js/apps/benchmark
npm i
npm start # This command starts a web server required by the headless browsers. Do not open the http://127.0.0.1:8080 address
npm run benchmark 

For benchmarking other browsers, use the browser basic usage sample

git clone https://github.com/bitair-org/concurrent.js.git
cd concurrent.js/apps/sample/browser
npm i
npm start # Open the http://127.0.0.1:8080 address in the target browser

Parallelism

To run each function call or object instance on a separate CPU core, the load method of the imported module must be called for each function call or object instance individually:

import { concurrent } from '@bitair/concurrent.js'

const extraBigint = concurrent.import('extra-bigint')

concurrent.config({ maxThreads: 16 }) // Instead of a hardcoded value, use os.availableParallelism() in Node.js v19.4.0 or later

const tasks = []
for (let i = 0; i <= 100; i++) {
  const { factorial } = await extraBigint.load()
  tasks.push(factorial(i))
}

const results = await Promise.all(tasks)
// ...rest of the code

await concurrent.terminate()

Reactive Concurrency

The reactive concurrency feature provides a bidirectional channel for messaging. A message can be replied to by returning a value:

services/index.mjs

// import type { IChannel } from '@bitair/concurrent.js'

export async function reactiveAdd(channel /*: IChannel */) {
  let done = false
  let sum = 0
  let i = 0

  channel.onmessage(name => {
    if (name === 'done') done = true
  })

  do {
    sum += await channel.postMessage('next', i++)
  } while (!done)

  return sum
}

index.mjs

import { concurrent, Channel } from '@bitair/concurrent.js'

const { reactiveAdd } = await concurrent.import(new URL('./services/index.mjs', import.meta.url)).load()

const channel = new Channel((onmessage, postMessage) => {
  const arr = [1, 2, 3, 4]

  onmessage(async (name, ...data) => {
    if (name === 'next') {
      const [i] = data
      if (i === arr.length - 1) await postMessage('done')
      return arr[i]
    }
  })
})

const result = await reactiveAdd(channel)
// ...rest of the code

await concurrent.terminate()

API

concurrent.import<T>(src: URL | string): IConcurrentModule<T>

Prepares a module to be loaded into workers. Note that only functions and classes can be imported.

  • src: URL | string

    Source of the module. Must be either a URL or a package name. Note that passing a package name is only applicable in Node.js.

IConcurrentModule<T>.load() : Promise<T>

Loads the module into a worker.

concurrent.config(settings: ConcurrencySettings): void

Configures the global settings of Concurrent.js.

  • settings: ConcurrencySettings

    • settings.maxThreads: number [default=1]

      The maximum number of available threads to be spawned.

    • settings.threadIdleTimeout: number | typeof Infinity [default=Infinity]

      Number of minutes before Concurrent.js terminates an idle thread.

    • settings.minThreads: number [default=0]

      The number of threads created when Concurrent.js starts and kept alive to avoid thread recreation overhead.

concurrent.terminate(force?: boolean): Promise<void>

Terminates Concurrent.js.

  • force?: boolean [Not implemented] Forces Concurrent.js to exit immediately without waiting for workers to finish their tasks.
class Channel implements IChannel

Used to send/receive messages to/from functions and methods (instance or static). Note that a function or method can only have one channel argument and it must be the last argument. The channel object cannot be reused to call another function or method.

  • constructor(listener: (onmessage: Channel['onmessage'], postMessage: Channel['postMessage']) => void)
  • onmessage(handler: (name: string | number, ...data: unknown[]) => unknown): void

    Sets the event handler for receiving a message. The handler should return a value if a reply is required for the message.

  • postMessage(name: string | number, ...data: unknown[]): Promise<unknown>

    Sends a message to the other end and returns its reply.

License

MIT License