At Stockflare, we use this docker container to enable Websocket connections through AWS Elastic Load Balancers.

The container uses confd for a one-time configuration of Nginx when the container starts, using environment variables. The Nginx configuration file is written to work in conjunction with an Elastic Load Balancer, itself using ProxyProtocol.

The container requires the presence of three environment variables. These variables will be used to configure Nginx when the container starts.

Note: The resulting configuration is checked for validity.

Lets say that we have the following container running on a host:

docker run -p 45490:45490 -n app -d my_websocket_application

To enable this container to proxy traffic to that application, we would run the following command:

docker run --link app:app \
           -p 45400:45400 \
           -e LISTEN_ON=45400 \
           -e DESTINATION_HOSTNAME=app \
           -e DESTINATION_PORT=45490 \
           -d stockflare/nginx-elb-websocket

This container would then start listening to traffic on port 45400, forwarding it to your application container on port 45490.

If you were running these containers on EC2 hosts that were part of a load balanced auto scaling group, the load balancer would be configured with the aforementioned ProxyProtocol on port 45400, complete with listeners forwarding TCP type traffic to port 45400.

The container also provides a naive health check for ELBs located at /ping.

Out-of-the-box, ELBs do not support Websockets and therefore require an additional proxy for which the connections must be negotiated through. There exist many blog posts on the internet, documenting the process for what we've done here. We wanted to create a more generic and complete approach to this. Below are some of the resources that we used:

• Web Apps: WebSockets with AWS Elastic Load Balancing
• Load-balancing Websockets on EC2
• 600k concurrent websocket connections on AWS using Node.js
• Load balancing WebSockets with ELB and nginx on EC2