Web Workers

Web Workers are a widely supported technology (Chrome, Firefox, Safari, Edge and IE11) that allows JavaScript to execute in a different thread, maximizing the usage of multiple CPUs; but most importantly not blocking the main thread.

The main thread is where JavaScript runs by default and has access to the document, window and other DOM APIs. The problem is that long-running JS prevents the browser from running smooth animations (CSS animations, transitions, canvas, svg...), making your site look frozen. That's why if your application needs to run CPU-intensive tasks, Web Workers are a great help.

When to use Web Workers?

The first thing to understand is when to use a Web Workers, and when not to use them since they come with a set of costs and limitations:

• There is no access to the DOM. This means you cannot interact with document, window or any elements in the page.
• There is no access to any of the @rindo/core APIs. For example, you cannot declare and use a component in a Web Worker, for the same reasons there is no access to the DOM.
• A Web Worker has its own isolated state since each worker has their own memory space. For example, a variable declared on the main thread cannot be directly referenced from a worker.
• There is an overhead when passing data between workers and the main thread. As a general rule, it's best to minimize the amount of data sent to and from the worker and be mindful if the work to send your data takes more time than doing it on the main thread.
• Communication is always asynchronous. Luckily Promises and async/await makes this relatively easy, but it's important to understand that communication between threads is always asynchronous.
• You can only pass primitives and objects that implement the structured clone algorithm. Best way to think of it is any data that can be serialized to JSON is safe to use.

In short, it's generally a good idea to use workers to move logic that is thread-blocking -- or UI-blocking, preventing users from interacting with the page -- into a Web Worker, such as real-time code syntax highlighting.

Best Practices when using Web Workers

• Use pure and functional algorithms in workers. (input1, input2) => output.
• The worker logic itself can be as complex as it has to be, however, the input and output data should stay fairly simple.
• Look for ways to reduce passing data between the main thread and worker thread.
• Class instances cannot be passed as data. Instead, only work with data can be JSON serializable.
• Minimize state within the worker, or better yet, completely avoid maintaining any state (e.g., don't put redux inside a worker).
• The cost of a worker should be easily amortized because it would be doing some CPU-intensive jobs.

How vanilla Web Workers "work"?

The browser comes with a Worker API, that works the following way:

const worker = new Worker('/my-worker.js');
worker.postMessage(['send message to worker']);
worker.onmessage = (ev) => {
  console.log('data from worker', ev.data);
};

This API, while powerful, is very low level and makes it tedious to write complex apps, since the event-driven paradigm leads easily to spaghetti-code, and quickly misses out on strongly-typed functions and data.

For further information, check out this fantastic tutorial by our friends at HTML5Rocks.

A Web Worker also requires the generation of a separate JavaScript bundle, such as the my-worker.js file in the example above. This means you usually need extra build scripts and tooling that transpiles and bundles the worker entry point into another .js file. Additionally, the main bundle must be able to reference the worker bundle's file location, which is oftentimes a challenge after transpiling, bundling, minifying, filename hashing and deploying to production servers.

Fortunately, Rindo can help you solve these two problems:

• Tooling: Transpiling, bundling, hashing, worker url path referencing
• Communication: Converting event-based communication to Promises, while still maintaining types.

Web Workers with Rindo

As we already mention, Rindo's compiler can help you to use workers in production seamlessly. Any TypeScript file within the src directory that ends with .worker.ts will automatically use a worker. For example:

src/stuff.worker.ts:

export const sum = async (a: number, b: number) => {
  return a + b;
};

export const expensiveTask = async (buffer: ArrayBuffer) => {
  for (let i = 0; i < buffer.length; i++) {
    // do a lot of processing
  }
  return buffer;
};

src/my-app/my-app.tsx:

import { Component } from '@rindo/core';

// Import the worker directly.
// Rindo will automatically create
// a proxy and run the module in a worker.
// IDEs and TypeScript will treat this import
// no differently than any other ESM import.
import { sum, expensiveTask } from '../../stuff.worker';

@Component({
  tag: 'my-cmp'
}
export class MyApp {

  async componentWillLoad() {
    // sum() will run inside a worker! and the result is a Promise<number>
    const result = await sum(1, 2);
    console.log(result); // 3

    // expensiveTask() will not block the main thread,
    // because it runs in parallel inside the worker.
    // Note that the functions must be async.
    const newBuffer = await expensiveTask(buffer);
    console.log(newBuffer);
  }
}

Under the hood, Rindo compiles a worker file and uses the standard new Worker() API to instantiate the worker. Then it creates proxies for each of the exported functions, so developers can interact with it using structured programming constructs instead of event-based ones.

note

Workers are already placed in a different chunk, and dynamically loaded using new Worker(). You should avoid using a dynamic import() to load them, as this will cause two network requests. Instead, use ES module imports as it's only importing the proxies for communicating with the worker.

Imports within a worker

Normal ESM imports are possible when building workers in Rindo. Under the hood, the compiler bundles all the dependencies of a worker into a single file that becomes the worker's entry-point, a dependency-free file that can run without problems.

src/loader.worker.ts:

import upngjs from 'upng-js';
import { Images } from './materials';

export const loadTexture = async (imagesSrcs: Images) => {
  const images = await Promise.all(imagesSrcs.map(loadOriginalImage));
  return images;
};

async function loadOriginalImage(src: string) {
  const res = await fetch(src);
  const png = upngjs.decode(await res.arrayBuffer());
  return png;
}

In this example, we are building a worker called loader.worker.ts that imports an NPM dependency (upngjs, used to parse png files), and a local module (./materials). Rindo will use Rollup to bundle all dependencies and remove all imports at runtime. Be aware that code will be duplicated if imported inside and outside a worker.

Dynamic imports

In order to load scripts dynamically inside of a worker, Web Workers come with a handy API, importScript().

Here's an example of how to use typescript directly from a CDN with importScript().

importScripts('https://cdn.jsdelivr.net/npm/typescript@latest/lib/typescript.js');

note

Do not use importScript() to import NPM dependencies you have installed using npm or yarn. Use normal ES module imports as usual, so the bundler can understand it.

Worker Callbacks

In most cases, waiting for a Promise to resolve with the output data is all we'll need. However, a limitation with native Promises is that it provides only one returned value. Where a traditional callback still shines is that it can be called numerous times with different data.

Let's say that we have a long running process that may take a few seconds to complete. With a Promise, we're unable to periodically receive the progress of the task, since all we can do is wait for Promise to resolve.

A feature with Rindo's worker is the ability to pass a callback to the method, and within the worker, execute the callback as much as it's needed before the task resolves.

In the example below, the task is given a number that it counts down from the number provided, and the task completes when it gets to 0. During the count down, however, the main thread will still receive an update every second. This example will console log from 5 to 0

src/countdown.worker.ts:

export const countDown = (num: number, progress: (p: number) => void) => {
  return new Promise((resolve) => {
    const tmr = setInterval(() => {
      num--;
      if (num > 0) {
        progress(num);
      } else {
        clearInterval(tmr);
        resolve(num);
      }
    }, 1000);
  });
};

src/my-app/my-app.tsx:

import { Component } from '@rindo/core';
import { countDown } from '../countdown.worker';

@Component({
  tag: 'my-cmp'
}
export class MyApp {

  componentWillLoad() {
    const startNum = 5;
    console.log('start', startNum);

    countDown(startNum, (p) => {
      console.log('progress', p);
    }).then(result => {
      console.log('finish', result);
    });
  }
}

When executed, the result would take 5 seconds and would log:

start 5
progress 4
progress 3
progress 2
progress 1
finish 0

Advanced cases

Sometimes it might be necessary to access the actual Worker instance, because manual usage of the postMessage() and onmessage is desired. However, there's still a tooling challenge in having to bundle the worker, and have the main bundle correctly reference the worker bundle url path. In that case, Rindo also has an API that exposes the worker directly so it can be used instead of the proxies mentioned early.

For a direct worker reference, add ?worker at the end of an ESM import. This virtual ES module will export:

• worker: The actual Worker instance.
• workerPath: The path to the worker's entry-point (usually a path to a .js file).
• workerName: The name of the worker, useful for debugging purposes.

src/my-app/my-app.tsx:

import { Component } from '@rindo/core';
import { sum } from '../../stuff.worker';

// Using the ?worker query, allows to access the worker instance directly.
import { worker } from '../../stuff.worker.ts?worker';

@Component({
  tag: 'my-cmp'
}
export class MyApp {

  componentWillLoad() {
    // Use worker api directly
    worker.postMessage(['send data manually']);

    // Use the proxy
    const result = await sum(1, 2);
    console.log(result); // 3
  }
}

You can even use this feature you create multiple Worker manually:

import { workerPath } from '../../stuff.worker.ts?worker';

const workerPool = [new Worker(workerPath), new Worker(workerPath), new Worker(workerPath), new Worker(workerPath)];

In this example, we exclusively take advantage of the bundling performed by the compiler to obtain the workerPath to the worker's entry point, then manually create a pool of workers.

note

Rindo will not instantiate a worker if it's unused, it takes advantage of tree-shaking to do this.

Worker Termination

Any Web Workers can be terminated using the Worker.terminate() API, but since Rindo creates one worker shared across all the proxied methods, it's not recommended to terminate it manually. If you have a use-case for using terminate and rebuilding workers, then we recommend using the workerPath and creating a new Worker directly:

import { workerPath } from '../../stuff.worker.ts?worker';
const worker = new Worker(workerPath);
// ...
worker.terminate();