November 25, 2024

Pyp Vaporisimo

more advanced technology

Traefik, Docker and dnsmasq to simplify container networking

Good tech adventures start with some frustration, a need, or a requirement. This is the story of how I simplified the management and access of my local web applications with the help of Traefik and dnsmasq. The reasoning applies just as well for a production server using Docker.

My dev environment is composed of a growing number of web applications self-hosted on my laptop. Such applications include several websites, tools, editors, registries, … They use databases, REST APIs, or more complex backends. Take the example of Supabase, the Docker Compose file includes the Studio, the Kong API gateway, the authentication service, the REST service, the real-time service, the storage service, the meta service, and the PostgreSQL database.

The result is a growing number of containers started on my laptop, accessible at localhost on various ports. Some of them use the default ports and cannot run in parallel to avoid conflicts. For example, the 3000 and 8000 ports are common to a lot of containers present on my machine. To circumvent the issue, some containers use custom ports which I often happen to forget.

The solution is to create local domain names which are easy to remember and use a web proxy to route the requests to the correct container. Traefik helps in the routing and the discovery of those services and dnsmasq provides a custom top-level domain (pseudo-TLD) to access them.

Another usage of Traefik is a production server using multiple Docker Compose files for various websites and web applications. The containers communicate inside an internal network and are exposed through a proxy service, in our case implemented with Caddy.

Problem description

Out of many, let’s take 3 web applications running locally. All of them are managed with Docker Compose:

  • Adaltas website, 1 container, Gatsby-based static website
  • Alliage website, 10 containers, Next.js frontend, Node.js backend, and Supabase
  • Penpot, 6 containers, Penpot frontend, backend services plus Inbucket for email testing (personal addition)

By default, those containers expose the following ports on localhost:

  • Adaltas
    • 8000 Gatsby server in dev mode
    • 9000 Gatsby service to serve a build website
  • Alliage
    • 3000 Next.js website both dev and build mode
    • 3001 Node.js custom API
    • 3000 Supabase Studio
    • 5555 Supabase Meta
    • 8000 Kong HTTP
    • 8443 Kong HTTPS
    • 5432 PostgreSQL
    • 2500 Inbucket SMTP server
    • 9000 Inbucket Web interface
    • 1100 Inbucket POP3 server
  • Penpot
    • 2500 Inbucket SMTP server
    • 9000 Inbucket Web interface
    • 1100 Inbucket POP3 server
    • 9001 Penpot frontend

Note, depending on your environment and desires, some ports might be restricted while other ports might be accessible.

As you can see, many ports collide with each other. It is not just the 2 instances of Inbucket running in parallel. For example, port 8000 is used both by Gatsby and Kong. It is a common default port for several applications. The same goes for ports 3000, 8080, 8443, …

One solution is to assign distinctive ports for each service. However, this approach is not scalable. Soon enough, I forget to which port each service is assigned.

Expected behavior

A better solution is the usage of a reverse proxy with hostnames easy to remember. Here is what we expect:

  • Adaltas
    • www.adaltas.local Gatsby server in dev mode
    • build.adaltas.local Gatsby service to serve a build website
  • Alliage
    • www.alliage.local Next.js website both dev and build mode
    • api.alliage.local Node.js custom API
    • studio.alliage.local Supabase Studio
    • meta.alliage.local Supabase Meta
    • kong.alliage.local Kong HTTP
    • kong.alliage.local Kong HTTPS
    • sql.alliage.local PostgreSQL
    • smtp.alliage.local Inbucket SMTP server
    • mail.alliage.local Inbucket Web interface
    • pop3.alliage.local Inbucket POP3 server
  • Penpot
    • www.penpot.local Penpot frontend
    • smtp.penpot.local Inbucket SMTP server
    • mail.penpot.local Inbucket Web interface
    • pop3.penpot.local Inbucket POP3 server

In a traditional setting, the reverse proxy is configured with one or multiple configuration files with all the routing information. However, a central configuration is not so convenient. It is preferable to have each service declares which hostname they resolve.

Automatic routing registration

All my web services are managed with Docker Compose. Ideally, I expect information to be present inside the Docker Compose file. Traefik is cloud-native in the sense that it configures itself using cloud-native workflows. The application provides some instructions present in its docker-compose.yml file and the containers are automatically exposed.

The way Traefik works with Docker, it plugs into the Docker socket, detects new services, and creates the routes for you.

Starting Traefik

To start Traefik inside Docker is straightforward (never say easy). The docker-compose.yml file is:

version: '3'
services:
  reverse-proxy:
    
    image: traefik:v2.9
    
    command: --api.insecure=true --providers.docker
    ports:
      
      - "80:80"
      
      - "8080:8080"
    volumes:
      
      - /var/run/docker.sock:/var/run/docker.sock

Registering new services

Let’s consider an additional service. The Adaltas website is a single container based on Gatsby. In development mode, it starts a web server on port 8000. I expect it to be accessible with the hostname www.adaltas.local on port 80.

Following the Traefik’s getting started with Docker, the integration is made with the property traefik.http.routers.router_name.rule present in the labels field of the docker service. It defines the hostname under which our website is accessible on port 80. It is set to www.adaltas.localhost because the .localhost TLD resolves locally by default. Since I prefer to use the .local domain, we set the domain to www.adaltas.local later using dnsmasq. The traffic is then routed to the container IP on port 8000. The container port is obtained by Traefik from the Docker Compose’s ports field.

version: '3'
services:
  www:
    container_name: adaltas-www
    ...
    labels:
    - "traefik.http.routers.adaltas-www.rule=Host(`www.adaltas.localhost`)"
    ports:
    - "8000:8000"

This works when both the Traefik and the Adaltas services are defined in the same Docker compose file. Firing docker-compose up and you can:

  • http://localhost:8080: Access the Traefik web UI
  • http://localhost:8080/api/rawdata: Access the Traefik’s API rawdata
  • http://www.adaltas.localhost: Access the Adaltas website in development mode
  • http://localhost:8080: Same as http://www.adaltas.localhost

There are 3 limitations we need to deal with:

  • Internal networking
    It only works because all the services are declared inside the same Docker Compose file. With separated Docker Compose files, an internal network must be used to communicate between the Traefic container and the targetted containers.
  • Domain name
    I wish to use a pseudo top-level domain (TLD), for example, www.adaltas.local instead of www.adaltas.localhost. The .local TLD does not yet resolve locally, a local DNS server must be configured.
  • Port label
    The port of Adaltas is defined inside the Docker Compose file. Thus, it is exposed on the host machine and it collides with other services. Port forwarding must be disabled and Traefik must be instructed about the port with another mechanism than the ports field.

Internal networking

When defined across separated files, the container cannot communicate. Each Docker Compose file generates a dedicated network. The targeted service is visible inside the Traefik UI. However, the request fails to be routed.

The containers must share a common network to communicate. When the Traefik container is started, a traefik_default network is created, see docker network list. Instead of creating a new network, let’s reuse it. Enrich the Docker Compose file of the targetted container, the Adaltas website in our case, with the network field:

version: '3'
services:
  www:
    container_name: adaltas-www
    
networks:  default:    name: traefik_default

After starting the 2 Docker Compose setups with docker-compose up, the Traefik and the Website containers start communicating.

Domain name

It is time to tackle the FQDN of our services. The current TLD in use, .localhost, is perfectly fine. It works by default and it is officially reserved for this usage. However, I wish to use my own top-level domains (pseudo-TLD name), we’ll use .local in this example.

Disclaimer, using a pseudo-TLD name is not recommended. The .local TLD is used by multicast DNS / zero-configuration networking. In practice, I haven’t encountered any issues. To mitigate the risk of conflicts, RFC 2606 reserves the following TLD names: .test, .example, .invalid, .localhost.

A local DNS server is used to resolve the *.local addresses. I had some experience with Bind in the past. A simpler and more lightweight option is the usage of dnsmasq. The instructions below cover the installation on MacOS and Ubuntu Desktop. In both cases, dnsmaq is installed and configured to not interfere with the current DNS settings.

MacOS instructions:


brew install dnsmasq

mkdir -pv $(brew --prefix)/etc/
echo 'address=/.local/127.0.0.1' >> $(brew --prefix)/etc/dnsmasq.conf

sudo brew services start dnsmasq

sudo mkdir -v /etc/resolver
sudo bash -c 'echo "nameserver 127.0.0.1" > /etc/resolver/test'

scutil --dns

Linux instructions with NetworkManager (eg Ubuntu Desktop):


systemctl disable systemd-resolved
systemctl stop systemd-resolved
unlink /etc/resolv.conf

cat <<CONF | sudo tee /etc/NetworkManager/conf.d/00-use-dnsmasq.conf
[main]
dns=dnsmasq
CONF

cat <<CONF | sudo tee /etc/NetworkManager/dnsmasq.d/00-dns-public.conf
server=8.8.8.8
CONF
cat <<CONF | sudo tee /etc/NetworkManager/dnsmasq.d/00-address-local.conf
address=/.local/127.0.0.1
CONF
systemctl restart NetworkManager

Use dig to validate that any FQDN using our pseudo-TLD resolves to the local
machine:

Port label

With the introduction of a reverse proxy like Traefik, exposing the container port on the host machine is no longer necessary, thus, eliminating the risk of collision between the exposed port and the ones of other services.

One label is already present to define the hostname of the website service. Traefik comes with a lot of complementary labels. The traefik.http.services.service_name.loadbalancer.server.port property tells Traefik to use a specific port to connect to a container.

The final Docker Compose file looks like this:

version: '3'
services:
  www:
    container_name: adaltas-www
    image: node:18
    volumes:
      - .:/app
    user: node
    working_dir: /app
    command: bash -c "yarn install && yarn run develop"
    labels:
    - "traefik.http.routers.adaltas-www.rule=Host(`www.adaltas.local`)"
    - "traefik.http.services.adaltas-www.loadbalancer.server.port=8000"
networks:
 default:
   name: traefik_default

Conclusion

With Traefik, I like the idea of my container services registering automatically in a cloud-native philosophy. It provided me with confort and simplicity. Also, dnsmasq has proved to be well-documented and quick to adjust to my various requirements.