Skip to content

Latest commit

 

History

History
738 lines (509 loc) · 24.9 KB

README.md

File metadata and controls

738 lines (509 loc) · 24.9 KB

npm version Test

AnyCable JavaScript Client

AnyCable brings performance and scalability to real-time applications built with Ruby and Rails. It uses Action Cable protocol and its extensions for client-server communication.

This repository contains JavaScript packages to build AnyCable clients.

Motivation

Multiple reasons that forced us to implement an alternative client library for Action Cable / AnyCable:

📖 Read also the introductory post.

Usage: Web

Install

npm install @anycable/web

# or

yarn add @anycable/web

Initialization

First, you need to create a client (or consumer as it's called in Action Cable):

// cable.js
import { createCable } from '@anycable/web'

export default createCable()

By default, the connection URL is looked up in meta tags (action-cable-url or cable-url), and if none found, fallbacks to /cable. You can also specify the URL explicitly:

createCable('ws://cable.example.com/my_cable')

Pub/Sub

Important

This feature is backed by AnyCable signed streams (available since v1.5). See the documentation.

You can subscribe directly to data streams as follows:

const cable = createCable();

const chatChannel = cable.streamFrom('room/42');

chatChannel.on('message', (msg) => {
  // ...
});

// Publish transient events
chatChannel.whisper({event: 'typing', user: '<name>'});

In most cases, however, you'd prefer to use secured (signed) stream names generated by your backend:

const cable = createCable();
const signedName = await obtainSignedStreamNameFromWhenever();
const chatChannel = cable.streamFromSigned(signedName);
// ...

Channels

AnyCable client provides multiple ways to subscribe to channels: class-based subscriptions and headless subscriptions.

Tip

Read more about the concept of channels and how AnyCable uses it here.

Class-based subscriptions

Class-based APIs allows provides an abstraction layer to hide implementation details of subscriptions. You can add additional API methods, dispatch custom events, etc.

Let's consider an example:

import { Channel } from '@anycable/web'

// channels/chat.js
export default class ChatChannel extends Channel {
  // Unique channel identifier (channel class for Action Cable)
  static identifier = 'ChatChannel'

  async speak(message) {
    return this.perform('speak', { message })
  }

  receive(message) {
    if (message.type === 'typing') {
      // Emit custom event when message type is 'typing'
      return this.emit('typing', message)
    }

    // Fallback to the default behaviour
    super.receive(message)
  }
}
import cable from 'cable'
import { ChatChannel } from 'channels/chat'

// Build an instance of a ChatChannel class.
const channel = new ChatChannel({ roomId: '42' })

// Subscribe to the server channel via the client.
cable.subscribe(channel) // return channel itself for chaining

// Wait for subscription confirmation or rejection
// NOTE: it's not necessary to do that, you can perform actions right away,
// the channel would wait for connection automatically
await channel.ensureSubscribed()

// Perform an action
// NOTE: Action Cable doesn't implement a full-featured RPC with ACK messages,
// so return value is always undefined
let _ = await channel.speak('Hello')

// Handle incoming messages
channel.on('message', msg => console.log(`${msg.name}: ${msg.text}`))

// Handle custom typing messages
channel.on('typing', msg => console.log(`User ${msg.name} is typing`))

// Or subscription close events
channel.on('close', () => console.log('Disconnected from chat'))

// Or temporary disconnect
channel.on('disconnect', () => console.log('No chat connection'))

// Unsubscribe from the channel (results in a 'close' event)
channel.disconnect()

IMPORTANT: cable.subscribe(channel) is optimistic: it doesn't require the cable to be connected, and waits for it to connect before performing a subscription request. Even if the cable got disconnected before subscription was confirmed or rejected, a new attempt is made as soon as the connectivity restored.

Calling channel.disconnect() removes the subscription for this channel right away and send unsubscribe request asynchronously; if there is no connectivity, we assume that the server takes care of performing unsubscribe tasks, so we don't need to retry them.

Headless subscriptions

Headless subscriptions are very similar to Action Cable client-side subscriptions except from the fact that no mixins are allowed (you classes in case you need them).

Let's rewrite the same example using headless subscriptions:

import cable from 'cable'

const subscription = cable.subscribeTo('ChatChannel', { roomId: '42' })

const _ = await subscription.perform('speak', { msg: 'Hello' })

subscription.on('message', msg => {
  if (msg.type === 'typing') {
    console.log(`User ${msg.name} is typing`)
  } else {
    console.log(`${msg.name}: ${msg.text}`)
  }
})

Action Cable compatibility mode

We provide an Action Cable compatible APIs for smoother migrations.

All you need is to change the imports:

- import { createConsumer } from "@rails/actioncable";
+ import { createConsumer } from "@anycable/web";

 // createConsumer accepts all the options available to createCable
 export default createConsumer();

Then you can use consumer.subscriptions.create as before (under the hood a headless channel would be create).

Lifecycle events

Both cables and channels allow you to subscribe to various lifecycle events for better observability.

Learn more from the dedicated documentation.

Handling connection failures, or automatic reconnects

AnyCable client provides automatic reconnection on network failure out-of-the-box. Under the hood, it uses the exponential backoff with jitter algorithm to make reconnection attempts non-deterministic (and, thus, prevent thundering herd attacks on the server). You can read more about it in the blog post.

The component responsible for reconnection is called Monitor, and it's created automatically, if you use the createCable (or createConsumer) function.

Sometimes it might be useful to disable reconnection. In that case, you MUST pass the monitor: false to the createCable function:

cable = createCable({monitor: false})

TypeScript support

You can make your channels more strict by adding type constraints for parameters, incoming message types and custom events:

// ChatChannel.ts
import { Channel, ChannelEvents } from '@anycable/web'

type Params = {
  roomId: string | number
}

type TypingMessage = {
  type: 'typing'
  username: string
}

type ChatMessage = {
  type: 'message'
  username: string
  userId: string
}

type Message = TypingMessage | ChatMessage

interface Events extends ChannelEvents<Message> {
  typing: (msg: TypingMessage) => void
}

// Which actions can be performed on this channel
interface Actions {
  speak: (msg: ChatMessage) => void
  typing: () => void
}

export class ChatChannel extends Channel<Params,Message,Events,Actions> {
  static identifier = 'ChatChannel'

  receive(message: Message) {
    if (message.type === 'typing') {
      return this.emit('typing', message)
    }

    super.receive(message)
  }

  sendMessage(msg: ChatMessage) {
    return this.perform('speak', msg)
  }

  sendTyping() {
    return this.perform('typing')
  }
}

Now this typings information would help you to provide params or subscribe to events:

let channel: ChatChannel

channel = new ChatChannel({roomId: '2021'}) //=> OK

channel = new ChatChannel({room_id: '2021'}) //=> NOT OK: incorrect params key
channel = new ChatChannel() //=> NOT OK: missing params

channel.on('typing', (msg: TypingMessage) => {}) //=> OK

channel.on('typing', (msg: string) => {}) //=> NOT OK: 'msg' type mismatch
channel.on('types', (msg: TypingMessage) => {}) //=> NOT OK: unknown event

channel.perform('speak', {type: 'message', username: 'John', userId: '42'}) //=> OK
channel.perform('speak', {body: 'hello!'}) //=> NOT OK: incorrect message type

channel.perform('typing') //=> OK
channel.perform('typing', {type: 'typing', username: 'John'}) //=> NOT OK: no payload expected

channel.perform('type') //=> NOT OK: unknown action

Supported protocols

By default, when you call createCable() we use the actioncable-v1-json protocol (supported by Action Cable).

You can also use Msgpack and Protobuf (soon) protocols supported by AnyCable Pro:

// cable.js
import { createCable } from '@anycable/web'
import { MsgpackEncoder } from '@anycable/msgpack-encoder'

export default createCable({protocol: 'actioncable-v1-msgpack', encoder: new MsgpackEncoder()})

// or for protobuf
import { createCable } from '@anycable/web'
import { ProtobufEncoder } from '@anycable/protobuf-encoder'

export default createCable({protocol: 'actioncable-v1-protobuf', encoder: new ProtobufEncoder()})

NOTE: You MUST install the corresponding encoder package yourself, e.g., yarn add @anycable/msgpack-encoder or yarn add @anycable/protobuf-encoder.

Extended Action Cable protocol

AnyCable client also supports an extended version of the Action Cable protocol (actioncable-v1-ext-json) implemented by AnyCable server (v1.4+).

This version provides additional functionality to improve data consistency:

  • Session recovery mechanism to restore subscriptions without re-subscribing.
  • History support for streams, or automatic retrieval of missing messages during short-term disconnects.

The features are implemented by the protocol itself, no need to update any existing channels code. All you need is to specify the protocol version when creating a client:

import { createCable } from '@anycable/web'
// or for non-web projects
// import { createCable } from '@anycable/core'

export default createCable({protocol: 'actioncable-v1-ext-json'})

Using with Protobuf and Msgpack

You can use the extended protocol with Protobuf and Msgpack encoders as follows:

// cable.js
import { createCable } from '@anycable/web'
import { MsgpackEncoder } from '@anycable/msgpack-encoder'

export default createCable({protocol: 'actioncable-v1-ext-msgpack', encoder: new MsgpackEncoder()})

// or for protobuf
import { createCable } from '@anycable/web'
import { ProtobufEncoderV2 } from '@anycable/protobuf-encoder'

export default createCable({protocol: 'actioncable-v1-ext-protobuf', encoder: new ProtobufEncoderV2()})

Loading initial history on client initialization

To catch up messages broadcasted during the initial page load (or client-side application initialization), you can specify the historyTimestamp option to retrieve messages after the specified time along with subscription requests. The value must be a UTC timestamp (the number of seconds). For example:

export default createCable({
  protocol: 'actioncable-v1-ext-json',
  protocolOptions: {
    historyTimestamp: 1614556800 // 2021-03-01 00:00:00 UTC
  }
})

By default, we use the current time (Date.now() / 1000). For web applications, you can specify the value using a meta tag with the name "action-cable-history-timestamp" (or "cable-history-timestamp"). For example, in Rails, you can add the following to your application layout

<!DOCTYPE html>
<html>
  <head>
    <!-- ... -->
    <%= action_cable_meta_tag %>
    <meta name="action-cable-history-timestamp" content="<%= Time.now.to_i %>">
  </head>
  <!-- ... -->
</html>

This is a recommended way to use this feature with Hotwire applications, where initial state is included in the HTML response.

IMPORTANT: For later subscriptions (not during the initial page initialization), the value of the historyTimestamp is automatically adjusted to the last time a "ping" message has been received.

You can also disable retrieving history since the specified time completely by setting the historyTimestamp option to false.

Handling history retrieval failures

AnyCable reliable streams store history for a finite period of time and also have an upper size limit. Thus, in some cases, clients may fail to retrieve the missed messages (e.g., after a long-term disconnect). To gracefully handle this situation, you may decide to fallback to a full state reset (e.g., a browser page reload). You can use the specific "info" event to react on various protocol-level events not exposed to the generic Channel interface:

import { createCable, Channel } from '@anycable/web'

const cable = createCable({protocol: 'actioncable-v1-ext-json'});

class ChatChannel extends Channel {
  static identifier = 'ChatChannel'

  constructor(params) {
    super(params)

    this.on("info", (evt) => {
      if (evt.type === "history_not_found") {
        // Restore state by performing an action
        this.perform("resetState")
      }

      // Successful history retrieval is also notified
      if (evt.type === "history_received") {
        // ...
      }
    })
  }
}

PONGs support

The extended protocol also support sending pong commands in response to ping messages. A server (AnyCable-Go) keeps track of pongs and disconnect the client if no pongs received in time. This helps to identify broken connections quicker.

You must opt-in to use this feature by setting the pongs option to true:

export default createCable({
  protocol: 'actioncable-v1-ext-json',
  protocolOptions: {
    pongs: true
  }
})

Refreshing authentication tokens

If you use a token-based authentication with expirable tokens (e.g., like AnyCable PRO JWT identification), you need a mechanism to refresh tokens for a long-lived clients (to let them reconnect in case of a connection failure).

AnyCable client can help you to make this process as simple as possible: just provide a function, which could retrieve a new token and update the connection url. AnyCable will take care of everything else (tracking expiration and reconnecting). Here is an example:

// cable.js
import { createCable } from '@anycable/web'

export default createCable({
  tokenRefresher: async transport => {
    let response = await fetch('/token.json')
    let data = await response.json()

    // Update URL for the underlying transport
    transport.setURL('ws://example.com/cable?token=' + data['token'])
  }
})

For browser usage, we provide a built-in helper method, which allows you to extract a new connection URL from an HTML page (requested via fetch):

// cable.js
import { createCable, fetchTokenFromHTML } from '@anycable/web'

// By default, the current page is loaded in the background,
// and the action-cable-url (or cable-url) meta tag is used to update
// the connection url
export default createCable({tokenRefresher: fetchTokenFromHTML()})

// You can also specify an alternative URL
export default createCable({
  tokenRefresher: fetchTokenFromHTML({ url: '/custom-token-refresh-endpoint' })
})

NOTE: the tokenRefresher only activates when a server sends a disconnection message with reason token_expired (i.e., {"type":"disconnect","reason":"token_expired","reconnect":false}).

NODE: the fetchTokenFromHTML performs an HTTP request with a specific header attached (X-ANYCABLE-OPERATION=token-refresh), which you could use to minimize the amount of HTML to return in response.

Hotwire (Turbo Streams) support

To use AnyCable client with Turbo Streams, we provide a tiny plugin—@anycable/turbo-stream. It allows you to configure a Cable instance yourself to use with Turbo Stream source elements:

import { start } from "@anycable/turbo-stream"
import cable from "cable"
// Explicitly activate stream source elements
start(cable)

Read more in the package's Readme.

Testing

For testing your channel you can use test cable implementation from @anycable/core/testing.

By using test cable implementation you can test channel's output actions. All actions store in cable outgoing property. Also test implementation helps to test channel disconnect event.

For example we have the following channel implementation.

import { Channel } from "@anycable/core";

class ChatChannel extends Channel {
  static identifier = "ChatChannel";

  async speak(message) {
    return this.perform("speak", { message });
  }

  leave() {
    // some custom logic
    return this.disconnect();
  }
}

We can test it like this (using Jest):

import { Channel } from './channel.js'
import { TestCable } from '@anycable/core/testing'

describe('ChatChannel', () => {
  let channel: Channel
  let cable: TestCable

  beforeEach(() => {
    cable = new TestCable()
    channel = new Channel()
    cable.subscribe(channel)
  })

  it('perform an speak action', async () => {
    await channel.speak('hello')
    await channel.speak('bye')

    expect(cable.outgoing).toEqual([
      { action: 'speak', payload: { message: 'hello' } },
      { action: 'speak', payload: { message: 'bye' } }
    ])
  })

  it('disconnects when leave', async () => {
    channel.leave()

    expect(channel.state).toEqual('closed')
  })
})

Babel/Browserlist configuration

This library uses ECMAScript 6 features (such as native classes), and thus, is not compatible with ES5 (for example, IE11 is not supported out-of-the-box).

You should either configure Babel to transform the lib's source code or do not compile into ES5 (that could be done by specifying the following Browserlist query: ["defaults", "not IE 11"]).

If you're using babel-loader, esbuild-loader or similar, you can use the include option to add @anycable/* libraries to the processed files. For example:

{
  include: [
    path.resolve("src"),
    path.resolve('node_modules/@anycable'),
  ]
}

CommonJS compatibility

We use ESM and don't any have any plans to support CommonJS ourselves. You can try commonizer for that.

Usage: Node.js

Currently, we do not provide a dedicated Node.js package. You can use @anycable/core with Node.js:

// WebSocket implementation compatible with the web WebSocket API is required
import WebSocket from 'ws'
import { createCable } from '@anycable/core'

// NOTE: Passing url is required
let cable = createCable(url, {
  websocketImplementation: WebSocket
})

// You can also pass additional connections options,
// supported by ws via the websocketOptions
let cableWithHeader = createCable(url, {
  websocketImplementation: WebSocket,
  websocketOptions: { headers: { 'x-token': 'secret' }}
})

IMPORTANT: We use ES modules, hence setting NODE_OPTIONS='--experimental-vm-modules' is currently required.

See also examples/benchmark_channel.js.

Usage: React Native

Currently, we do not provide a dedicated React Native package. You can use @anycable/core just like with Node.js:

import { createCable } from '@anycable/core'

// NOTE: Passing url is required
let cable = createCable(url)

// You can also pass additional connections options,
// such as headers, via the websocketOptions
let cableWithHeader = createCable(url, {
  websocketOptions: { headers: { 'x-token': 'secret' }}
})

Duplicating or reusing channel instances?

It is safe to call cable.subscribe(channel) multiple times—only a single subscription (from the protocol point of view) is made, i.e., this action is idempotent. At the same time, it's safe to have multiple channel instances with the same identifiers client-side—only a single real subscription would be made.

Let's consider an example. Suppose you have two components relying on the same channel:

// component-one.js
import cable from 'cable'
import { NotificationsChannel } from 'channels/notifications_channel'

// Build an instance of a NotificationChannel class.
const channel = new NotificationChannel()

// Subscribe to the server channel via the client.
cable.subscribe(channel)

channel.on('message', msg => console.log("component one received message", `${msg.name}: ${msg.text}`))

// component-two.js
import cable from 'cable'
import { NotificationsChannel } from 'channels/notifications_channel'

// Build an instance of a NotificationChannel class.
const channel = new NotificationChannel()

// Subscribe to the server channel via the client.
cable.subscribe(channel)

channel.on('message', msg => console.log("component two received message", `${msg.name}: ${msg.text}`))

The code above would work as expected: both channel instances would receive updates from the server. Calling channel.disconnect() would detach this particular channel from the cable, but wouldn't perform the actual unsubscribe command (from the server perspective) unless that's the last channel with this identifier.

Alternatively, you may consider extracting a channel instance to a separate module and reuse it:

// channels/notifications_channel.js
import { Channel } from '@anycable/core'
import 'cable' from 'cable'

export class NotificationsChannel extends Channel {
  // ...
}

let instance

export function createChannel() {
  if (!instance) {
    instance = new NotificationChannel()
    cable.subscribe(channel)
  }

  return instance
}

// component-one.js
import cable from 'cable'
import { createChannel } from 'channels/notifications_channel'

const channel = createChannel()

channel.on('message', msg => console.log("component one received message", `${msg.name}: ${msg.text}`))

// component-two.js
import cable from 'cable'
import { createChannel } from 'channels/notifications_channel'

const channel = createChannel()

channel.on('message', msg => console.log("component two received message", `${msg.name}: ${msg.text}`))

Which way to choose is up to the developer. From the library point of view, both are viable and supported.

Fine-tuning for higher loads 📈

Customizing PING interval

The default PING interval is 3 seconds. When server handles tons of connections, sending pings that often might result in a noticeable overhead.

This value is defined by the WebSocket server. If you use AnyCable, you can customize it via the --ping_interval parameter. Also, since v1.4.3, AnyCable-Go allows you to configure a ping interval for an individual connection by adding a query param to the connection URL (?pi=10).

We recommend increasing the ping interval for high-load applications.

You MUST also update the client-side configuration to use the same value (so the client won't decide to reconnect due to missing pings):

export default createCable({
  pingInterval: 10000 // 10 seconds
})

Customizing subscription confirmation timeout

By default, we expect a subscription confirmation (or rejection) to arrive within 5 seconds. If it doesn't happen, AnyCable client re-issues the subscription request and waits for another 5 seconds. If the second attempt fails, the subscription is considered to be rejected due to timeout.

Under load (usually, during connection avalanches, i.e., when most clients are re-connecting), the server might not be able to respond within 5 seconds (and even 10 seconds). In that case, you can increase the retry interval:

export default createCable({
  protocolOptions: {
    subscribeRetryInterval: 10000 // 10 seconds
  }
})

Linearizing subscription requests

Another way to reduce the load on the server and avoid subscription timeouts is to linearize subscription requests. By default, AnyCable client sends subscription requests concurrently. However, you have many active subscriptions, this might result in a huge number of requests sent to the server at the same time during re-connections.

To smooth the load, you can disable concurrent subscription requests (so, every next subscription request would be sent only after the previous one is confirmed or rejected):

export default createCable({
  concurrentSubscribes: false
});

Further reading