Chatr - Azure Web PubSub Sample App

This is a demonstration & sample application designed to be a simple multi-user web based chat system.
It provides persistent group chats, user to user private chats, a user list, idle (away from keyboard) detection and several other features.

It is built on several Azure technologies, including: Web PubSub, Static Web Apps and Table Storage

👁‍🗨 Note. This was created as a personal project, created to aid learning while building something interesting. The code comes with all the caveats you might expect from such a project.

Goals:

Learn about using websockets
Write a 'fun' thing
Try out the new Azure Web PubSub service
Use the authentication features of Azure Static Web Apps
Deploy everything using Azure Bicep

Use cases & key features:

Sign-in with Microsoft, Twitter or GitHub accounts
Realtime chat with users
Shared group chats, only the creator can remove the chat
Detects where users are idle and away from keyboard (default is one minute)
Private 'user to user' chats, with notifications and popups

Screenshot

Architecture

Client / Frontend

This is the main web frontend as used by end users via the browser.

The source for this is found in client/ and consists of a static standalone pure ES6 JS application, no bundling or Node.js is required. It is written using Vue.js as a supporting framework, and Bulma as a CSS framework.

Some notes:

ES6 modules are used so the various JS files can use import/export without the need to bundle.
Vue.js is used as a browser side library loaded from CDN as a ESM module, this is an elegant & lightweight approach supported by modern browsers, rather than the usual vue-cli style app which requires Node and webpack etc.
client/js/app.js shows how to create a Vue.js app with child components using this approach. The majority of client logic is here.
client/js/components/chat.js is a Vue.js component used to host each chat tab in the application
The special .auth/ endpoint provided by Static Web Apps is used to sign users in and fetch their user details, such as userId.

Server

This is the backend, handling websocket events to and from Azure Web PubSub, and providing REST API for some operations.

The source for this is found in api/ and consists of a Node.js Azure Function App. It connects to Azure Table Storage to persist group chat and user data (Table Storage was picked as it's simple & cheap). This is not hosted in a standalone Azure Function App but instead deployed into the Static Web App as part of it's serverless API support

There are four HTTP functions all served from the default /api/ path

eventHandler - Webhook receiver for "upstream" events sent from Azure Web PubSub service, contains the majority of application logic. Not called directly by the client, only Azure WebPub Sub.
getToken - Called by the client to get an access token and URL to connect via WebSockets to the Azure Web PubSub service. Must be called with userId in the URL query, e.g. GET /api/getToken?userId={user}
getUsers - Returns a list of signed in users, note the route for this function is /api/users
getChats - Returns a list of active group chats, note the route for this function is /api/chats

State is handled with state.js which is an ES6 module exporting functions supporting state CRUD for users and chats. This module carries out all the interaction with Azure Tables, and provides a relatively transparent interface, so a different storage backend could be swapped in.

WebSocket & API Message Flows

There is two way message flow between clients and the server via Azure Web PubSub and event handlers

The json.webpubsub.azure.v1 subprotocol is used rather than basic WebSockets, this provides a number of features: users can be added to groups, clients can send custom events (using type: event), and also send messages direct to other clients without going via the server (using type: sendToGroup)

Notes:

Chat IDs are simply randomly generated GUIDs, these correspond to "groups" in the subprotocol.
Private chats are a special case, they are not persisted in state, and they do not trigger chatCreated events. Also the user doesn't issue a joinChat event to join them, that is handled by the server as a kind of "push" to the clients.
User IDs are simply strings which are considered to be unique, this could be improved, e.g. with prefixing.

Client Messaging

Events & chat are sent using the json.webpubsub.azure.v1 subprotocol

Chat messages sent from the client use sendToGroup and a custom JSON payload with three fields message, fromUserId & fromUserName, these messages are relayed client to client by Azure Web PubSub, the server is never notified of them:

{
  type: 'sendToGroup',
  group: <chatId>,
  dataType: 'json',
  data: {
    message: <message text>,
    fromUserId: <userId>,
    fromUserName: <userName>,
  },
}

Events destined for the backend server are sent as WebSocket messages from the client via the same subprotocol with the event type, and an application specific sub-type, e.g.

{
  type: 'event',
  event: 'joinChat',
  dataType: 'text',
  data: <chatId>,
}

The types of events are:

createChat - Request the server you want to create a group chat
createPrivateChat - Request the server you want to create a private chat
joinChat - To join a chat, the server will add user to the group for that chatId
leaveChat - To leave a group chat
deleteChat - Called from a chat owner to delete a chat
userEnterIdle - Let the server know user is now idle
userExitIdle - Let the server know user is no longer idle

The backend API eventHandler function has cases for each of these user events, along with handlers for connection & disconnection system events.

Server Messaging

Messages sent from the server have a custom Chatr app specific payload as follows:

{
  chatEvent: <eventType>,
  data: <JSON object type dependant>
}

Where eventType is one of:

chatCreated - Let all users know a new group chat has been created
chatDeleted - Let all users know a group chat has been removed
userOnline - Let all users know a user has come online
userOffline - Let all users know a user has left
joinPrivateChat - Sent to both the initiator and recipient of a private chat
userIsIdle - Sent to all users when a user enters idle state
userNotIdle - Sent to all users when a user exits idle state

The client code in client/js/app.js handles these messages as they are received by the client, and reacts accordingly.

Some Notes on Design and Service Choice

The plan of this project was to use Azure Web PubSub and Azure Static Web Apps, and to host the server side component as a set of serverless functions in the Static Web Apps API support (which is in fact Azure Functions under the hood). Azure Static Web Apps was selected because it has amazing support for codeless and config-less user sign-in and auth, which I wanted to leverage.

Some comments on this approach:

API support in Static Web Apps is quite limited and can't support the new bindings and triggers for Web PubSub. HOWEVER You don't need to use these bindings 😂. You can create a standard HTTP function to act as a webhook event handler instead of using the webPubSubConnection binding. For sending messages back to Web PubSub, the server SDK can simply be used within the function code rather than using the webPubSub output binding.
Table Storage was picked for persisting state as it has a good JS SDK (the new SDK in @azure/data-table was used), it's extremely lightweight and cheap and was good enough for this project, see deails below

State & Entity Design

State in Azure Tables consists of two tables (collections) named chats and users

Chats Table

As each chat contains nested objects inside the members field, each chat is stored as a JSON string in a field called data. The PartitionKey is not used and hardcoded to a string "chatr". The RowKey and the id field inside the data object are the same.

PartitionKey: "chatr"
RowKey: The chatId (random GUID created client side)
data: JSON stringified chat entity

Example of a chat data entity

{
  "id": "eab4b030-1a3d-499a-bd89-191578395910",
  "name": "This is a group chat",
  "members": {
    "0987654321": {
      "userId": "0987654321",
      "userName": "Another Guy"
    },
    "1234567890": {
      "userId": "1234567890",
      "userName": "Ben"
    }
  },
  "owner": "1234567890"
}

Users Table

Users are stored as entities with the fields (columns) described below. As there are no nested fields, there is no need to encode as a JSON string. Again the PartitionKey is not used and hardcoded to a string "chatr".

PartitionKey: "chatr"
RowKey: The userId field returned from Static Web Apps auth endpoint
userName: The username (could be email address or handle) of the user
userProvider: Which auth provided the user signed in with twitter, aad or github
idle: Boolean, indicating if the user us currently idle

Running and Deploying the App

Working Locally

See makefile

$ make
help                 💬 This help message
lint                 🔎 Lint & format, will not fix but sets exit code on error
lint-fix             📜 Lint & format, will try to fix errors and modify code
run                  🏃 Run server locally using Static Web Apps CLI
clean                🧹 Clean up project
deploy               🚀 Deploy everything to Azure using Bicep
tunnel               🚇 Start loophole tunnel to expose localhost

Deploying to Azure

Deployment is slightly complex due to the number of components and the configuration between them. The makefile target deploy should deploy everything for you in a single step using Bicep templates found in the deploy/ folder

See readme in deploy folder for details and instructions

Running Locally

This is possible but requires a little effort as the Azure Web PubSub service needs to be able call the HTTP endpoint on your location machine, so a tunnel has employed.

When running locally the Static Web Apps CLI is used and this provides a fake user authentication endpoint for us.

A summary of the steps is:

Deploy an Azure Storage account, get name and access key.
Deploy an Azure Web Pub Sub instance, get connection string from the 'Keys' page.
Copy api/local.settings.sample.json to api/local.settings.json and edit the required settings values.
Start a localhost tunnel service such as ngrok or loophole. The tunnel should expose port 7071 over HTTP.
I use loophole as it allows me to set a custom host & DNS name, e.g.
- loophole http 7071 --hostname chatr
In Azure Web Pub Sub settings.
- Add a hub named chat
- In the URL template put https://{{hostname-of-tunnel-service}}/api/eventHandler
- In system events tick connected and disconnected
Run make run
Open http://localhost:4280/index.html

Known Issues

Won't run in Firefox as top level await is not yet supported

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

benc-uk / chatr

Programming Languages

Labels

Projects that are alternatives of or similar to chatr