A basic chat application built with Hapi.js Socket.io and Redis Pub/Sub
Try it: https://hapi-chat.herokuapp.com/
Node.js Chat Apps are practically the "Hello World" of real-time apps. If you Google for "node.js chat example" you will see thousands of results! But ... 90% of the examples use Express.js, 95% use MongoDB to store data/messages and 100% don't have any tests. So, this example is for the the people who identify with:
"We use hapi.js because we want our code to be performant and reliable"
As with all our examples we have a suite of tests.
Real-Time Chat is an integral part of any communications system.
Building a (basic) chat system is easy with Socket.io.
This example app shows you how to use Socket.io with Hapi.js and Redis for a Horizontally Scalable chat capable of hundrededs of thousands of concurrent clients.
We are using the following components to build our chat app:
- Hapi.js (node.js web framework) - If you haven't used Hapi.js before, checkout our introductory tutorial: https://github.com/nelsonic/learn-hapi
- Socket.io (WebSockets with fallback for older clients) - If you're new to Soecket.io see: http://socket.io/get-started/chat/
- Redis (high performance message storage and publish/subscribe) - If you or anyone on your team are completely new to Redis, check out: https://github.com/dwyl/learn-redis
Socket.io only handles distributing messages, if people disconnect from the chat they will miss any subsequent messages and when anyone connects there will see no history ... so we need a place to store messages for retrieval.
Top 3 reasons why Redis is the clear choice for storing chat messages.
- Speed - Redis is much faster than MongoDB, CouchDB or PostgreSQL
- Simple - pushing messages onto a list (set) is the simplest possible way to store a chat history. Given that we can store up to 512Mb per chat and stream chat history to new clients (low http overhead) its an incredibly simple setup!
- Scalable Publish/Subscribe ("pattern") means you can scale out (add more node.js/socket.io servers when you need to serve more clients) Redis can already handle an order of magnitude more than other NoSQL Databases, so your most likely "bottleneck" is node (nuts, hey!?)
The Publish Subscribe "Pattern" is (still) the simple_st_ way of scaling software applications. if you are new to this idea, see: https://en.wikipedia.org/wiki/Publish%E2%80%93subscribe_pattern
Given the simplicity of the UI, the chat app is mobile-first by default.
If anyone has time to Pull Request a few CSS media queries to make the UI even better on mobile devices, we would massively appreciate the contribution!
We use cookies to store the person's name on the client. If the person clears cookies (or uses private browsing / incognito mode) they will be asked for their name each time they open the chat window. (this is pretty standard)
At present we are caching all the messages in Redis. But a less RAM-hungry way to scale the app would be to store only the 50-100 most recent chat messages in Redis (RAM) and the remaining history in a cheaper on-disk storage e.g. ElasticSearch (which would also enable searchability)
Data Model we have used is incredibly simple. It translates to an array of objects:
var chat = [
'{"m":"Hi everyone!","t":1436263590869,"n":"Steve"}',
'{"m":"Hi Steve! Welcome to Hapi Chat!","t":1436263599489,"n":"Foxy"}',
'{"m":"Hapi Chat lets you chat with your friends!","t":1436263613141,"n":"Oprah"}',
'{"m":"Cool! How does it scale?","t":1436263620853,"n":"Steve"}',
'{"m":"Funny you should ask! It scales nicely because it uses Hapi.js and Redis!","t":1436263639989,"n":"Chroma"}',
'{"m":"Sweet! ","t":1436263645610,"n":"Steve"}',
'{"m":"Hello","t":1436264664835,"n":"Timmy"}',
'{"m":"Hi!","t":1436267152379,"n":"Timmy"}',
'{"m":"lkjlkjlk","t":1436270948402,"n":"dd"}',
'{"m":"Big fan of the little notifications at the top when a person joins","t":1436273109909,"n":"iteles"}'
]We use:
- m for the key of the message.
- n for the name of the person who wrote the message
- t for the timestamp the message was received by the node server (to avoid time-zone issues);
- Matt Harrison has basic example, but no tests (bad habits ...): http://matt-harrison.com/using-hapi-js-with-socket-io
- Scalability: https://en.wikipedia.org/wiki/Scalability#Horizontal_and_vertical_scaling
- Difference between scaling horizontally and vertically for databases: http://stackoverflow.com/questions/11707879/difference-between-scaling-horizontally-and-vertically-for-databases
- Using Pub/Sub for Asynchronous Communication: http://www.rediscookbook.org/pubsub_for_asynchronous_communication.html
- How to use redis PUBLISH/SUBSCRIBE with nodejs to notify clients when data values change? http://stackoverflow.com/questions/4441798/how-to-use-redis-publish-subscribe-with-nodejs-to-notify-clients-when-data-value (_don't you love it when someone else has aready asked/answered your questions...?)
- node_redis pub/sub example: https://github.com/mranney/node_redis/blob/master/examples/pub_sub.js
- Redis PubSub example using express (no tests!): https://github.com/rajaraodv/redispubsub