Akka HTTP / Scala.js / Websocket Chat App

A simple chat app that uses akka-http backend and a scala.js frontend to implement a simple websocket based chat application.

To run:

sbt

> project backend
> re-start

Navigate to http://localhost:8080/.

You can build a fully self-contained jar using assembly in the backend project.

Configuration

You can set app.interface and app.port in application.conf to configure where the server should listen to.

This also works on the command line using JVM properties, e.g. using re-start:

> re-start --- -Dapp.interface=0.0.0.0 -Dapp.port=8080

will start the server listening on all interfaces.

CLI

The cli project contains a command line client for the chat to demonstrate the Websocket client and how to deal with console input in a streaming way.

It runs best directly from a terminal.

Start the server as explained above. Then, to build a fat jar use

sbt

> project cli
> assembly

Run

java -jar cli/target/scala-2.11/cli-assembly-0.1-SNAPSHOT.jar

or

./chat

Here's another screencast that shows live tab completion in action:

Known issues

Handling of backpressure

The chat (actor) itself doesn't yet implement any meaningful backpressure logic.

• On the incoming side you probably want to backpressure (rate-limit) each client itself and the total rate of messages maybe as well
• On the ougoing side you don't want one slow chat participant to slow down the complete chat. Right now the outgoing side uses a Source.actorRef with an overflow strategy of fail: if a client doesn't keep up with receiving messages and the network send buffer on the chat server for that client fills up the client will be failed. (This is somewhat similar to what Twitter does in its streaming APIs). A better strategy may be to drop messages (and leave a note for the user) until the client catches up with the action.

Usage of stream combinators

Ideally, akka-stream would support dynamic merge/broadcast operations, so that you never need to break out of stream logic. Right now, collecting and broadcasting messages is done by the chat actor and for each user a manual stream pipeline needs to be setup.

The "frontend"

There isn't more than absolutely necessary there right now.