authelia/docs/deployment-production.md

2.5 KiB

Deployment for Production

Authelia can be deployed on bare metal or on Kubernetes with two different kind of artifacts: the distributable version (binary and public_html) or a Docker image.

NOTE: If not done already, we highly recommend you first follow the Getting Started documentation.

On Bare Metal

Authelia has been designed to be a proxy companion handling the authentication and authorization requests for your entire infrastructure.

As Authelia will be key to your architecture, it requires several components to make it highly-available. Deploying it in production means having an LDAP server for storing the information about the users, a Redis cache to store the user sessions in a distributed manner, a SQL server like MariaDB to persist user configurations and one or more nginx reverse proxies configured to be used with Authelia. With such a setup Authelia can easily be scaled to multiple instances to evenly handle the traffic.

NOTE: If you don't have all those components, don't worry, there is a way to deploy Authelia with only nginx. This is described in Deployment for Devs.

Here are the available steps to deploy Authelia given the configuration file is /path/to/your/configuration.yml. Note that you can create your own configuration file from config.template.yml located at the root of the repo.

Deploy with the distributable version

# Build it if not done already
$ authelia-scripts build
$ PUBLIC_DIR=./dist/public_html authelia --config /path/to/your/configuration.yml

Deploy With Docker

$ docker run -v /path/to/your/configuration.yml:/etc/authelia/configuration.yml -e TZ=Europe/Paris authelia/authelia

On top of Kubernetes

Authelia can also be installed on top of Kubernetes using nginx ingress controller.

Please refer to the following documentation for more information.

FAQ

Why is this not automated?

Ansible would be a very good candidate to automate the installation of such an infrastructure on bare metal. We would be more than happy to review any PR on that matter.

Regarding Kubernetes, the right way to go would be to write a Helm recipe. Again, we would be glad to review any PR implementing this.