lucaslorentz/caddy-docker-proxy

Caddy-Docker-Proxy

Introduction

This plugin enables Caddy to be used as a reverse proxy for Docker containers via labels.

How does it work?

The plugin scans Docker metadata, looking for labels indicating that the service or container should be served by Caddy.

Then, it generates an in-memory Caddyfile with site entries and proxies pointing to each Docker service by their DNS name or container IP.

Every time a Docker object changes, the plugin updates the Caddyfile and triggers Caddy to gracefully reload, with zero-downtime.

Table of contents

• Basic usage example, using docker compose
• Labels to Caddyfile conversion
  • Tokens and arguments
  • Ordering and isolation
  • Sites, snippets and global options
  • Go templates
• Template functions
  • upstreams
• Examples
• Docker configs
• Proxying services vs containers
  • Services
  • Containers
• Execution modes
  • Server
  • Controller
  • Standalone (default)
• Options
• Docker images
  • Choosing the version numbers
  • Chosing between default or alpine images
  • CI images
  • ARM architecture images
  • Windows images
  • Custom images
• Connecting to Docker Host
• Volumes
• Trying it
  • With Docker Compose (compose.yaml)
  • With run commands
• Building it

Basic usage example (Docker Compose)

$ docker network create caddy --ipv6

[!NOTE] The --ipv6 flag instructs Docker to assign IPv6 addresses for all containers connected to this network. Without this flag, Caddy (as well as any upstream services) will see Docker's gateway IP address instead of the actual client IP addresses for IPv6 clients.

caddy/compose.yaml

services:
  caddy:
    image: lucaslorentz/caddy-docker-proxy:ci-alpine
    ports:
      - 80:80
      - 443:443/tcp
      - 443:443/udp
    environment:
      - CADDY_INGRESS_NETWORKS=caddy
    networks:
      - caddy
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
      - caddy_data:/data
    restart: unless-stopped

networks:
  caddy:
    external: true

volumes:
  caddy_data: {}

$ docker compose up -d

whoami/compose.yaml

services:
  whoami:
    image: traefik/whoami
    networks:
      - caddy
    labels:
      caddy: whoami.example.com
      caddy.reverse_proxy: "{{upstreams 80}}"

networks:
  caddy:
    external: true

$ docker compose up -d

Now, visit https://whoami.example.com. The site will be served automatically over HTTPS with a certificate issued by Let's Encrypt or ZeroSSL.

Labels to Caddyfile conversion

Please first read the Caddyfile Concepts documentation to understand the structure of a Caddyfile.

Any label prefixed with caddy will be converted into a Caddyfile config, following these rules:

Tokens and arguments

Keys are the directive name, and values are whitespace separated arguments:

caddy.directive: arg1 arg2
↓
{
	directive arg1 arg2
}

If you need whitespace or line-breaks inside one of the arguments, use double-quotes or backticks around it:

caddy.respond: / "Hello World" 200
↓
{
	respond / "Hello World" 200
}

caddy.respond: / `Hello\nWorld` 200
↓
{
	respond / `Hello
World` 200
}

caddy.respond: |
	/ `Hello
	World` 200
↓
{
	respond / `Hello
World` 200
}

Dots represent nesting, and grouping is done automatically:

caddy.directive: argA  
caddy.directive.subdirA: valueA  
caddy.directive.subdirB: valueB1 valueB2
↓
{
	directive argA {  
		subdirA valueA  
		subdirB valueB1 valueB2  
	}
}

Arguments for the parent directive are optional (e.g. no arguments to directive, setting subdirective subdirA directly):

caddy.directive.subdirA: valueA
↓
{
	directive {
		subdirA valueA
	}
}

Labels with empty values generate a directive without any arguments:

caddy.directive:
↓
{
	directive
}

Ordering and isolation

Be aware that directives are subject to be sorted according to the default directive order defined by Caddy, when the Caddyfile is parsed (after the Caddyfile is generated from labels).

Directives from labels are ordered alphabetically by default:

caddy.bbb: value
caddy.aaa: value
↓
{
	aaa value 
	bbb value
}

Suffix _<number> isolates directives that otherwise would be grouped:

caddy.route_0.a: value
caddy.route_1.b: value
↓
{
	route {
		a value
	}
	route {
		b value
	}
}

Prefix <number>_ isolates directives but also defines a custom ordering for directives (mainly relevant within route blocks), and directives without order prefix will go last:

caddy.1_bbb: value
caddy.2_aaa: value
caddy.3_aaa: value
↓
{
	bbb value
	aaa value
	aaa value
}

Sites, snippets and global options

A label caddy creates a site block:

caddy: example.com
caddy.respond: "Hello World" 200
↓
example.com {
	respond "Hello World" 200
}

Or a snippet:

caddy: (encode)
caddy.encode: zstd gzip
↓
(encode) {
	encode zstd gzip
}

It's also possible to isolate Caddy configurations using suffix _<number>:

caddy_0: (snippet)
caddy_0.tls: internal
caddy_1: site-a.com
caddy_1.import: snippet
caddy_2: site-b.com
caddy_2.import: snippet
↓
(snippet) {
	tls internal
}
site_a {
	import snippet
}
site_b {
	import snippet
}

Global options can be defined by not setting any value for caddy. They can be set in any container/service, including caddy-docker-proxy itself. Here is an example

caddy.email: [email protected]
↓
{
	email [email protected]
}

Named matchers can be created using @ inside labels:

caddy: localhost
[email protected]: /sourcepath /sourcepath/*
caddy.reverse_proxy: @match localhost:6001
↓
localhost {
	@match {
		path /sourcepath /sourcepath/*
	}
	reverse_proxy @match localhost:6001
}

Go templates

Golang templates can be used inside label values to increase flexibility. From templates, you have access to current Docker resource information. But, keep in mind that the structure that describes a Docker container is different from a service.

While you can access a service name like this:

caddy.respond: /info "{{.Spec.Name}}"
↓
respond /info "myservice"

The equivalent to access a container name would be:

caddy.respond: /info "{{index .Names 0}}"
↓
respond /info "mycontainer"

Sometimes it's not possile to have labels with empty values, like when using some UI to manage Docker. If that's the case, you can also use our support for go lang templates to generate empty labels.

caddy.directive: {{""}}
↓
directive

Template functions

The following functions are available for use inside templates:

upstreams

Returns all addresses for the current Docker resource separated by whitespace.

For services, that would be the service DNS name when proxy-service-tasks is false, or all running tasks IPs when proxy-service-tasks is true.

For containers, that would be the container IPs.

Only containers/services that are connected to Caddy ingress networks are used.

:warning: caddy docker proxy does a best effort to automatically detect what are the ingress networks. But that logic fails on some scenarios: #207. To have a more resilient solution, you can manually configure Caddy ingress network using CLI option ingress-networks, environment variable CADDY_INGRESS_NETWORKS. You can also specify the ingress network per container/service by adding to it a label caddy_ingress_network with the network name.

Usage: upstreams [http|https] [port]

Examples:

caddy.reverse_proxy: {{upstreams}}
↓
reverse_proxy 192.168.0.1 192.168.0.2

caddy.reverse_proxy: {{upstreams https}}
↓
reverse_proxy https://192.168.0.1 https://192.168.0.2

caddy.reverse_proxy: {{upstreams 8080}}
↓
reverse_proxy 192.168.0.1:8080 192.168.0.2:8080

caddy.reverse_proxy: {{upstreams http 8080}}
↓
reverse_proxy http://192.168.0.1:8080 http://192.168.0.2:8080

:warning: Be carefull with quotes around upstreams. Quotes should only be added when using yaml.

caddy.reverse_proxy: "{{upstreams}}"
↓
reverse_proxy "192.168.0.1 192.168.0.2"

Examples

Proxying all requests to a domain to the container

caddy: example.com
caddy.reverse_proxy: {{upstreams}}

Proxying all requests to a domain to a subpath in the container

caddy: example.com
caddy.rewrite: * /target{path}
caddy.reverse_proxy: {{upstreams}}

Proxying requests matching a path

caddy: example.com
caddy.handle: /source/*
caddy.handle.0_reverse_proxy: {{upstreams}}

Proxying requests matching a path, while stripping that path prefix

caddy: example.com
caddy.handle_path: /source/*
caddy.handle_path.0_reverse_proxy: {{upstreams}}

Proxying requests matching a path, rewriting to different path prefix

caddy: example.com
caddy.handle_path: /source/*
caddy.handle_path.0_rewrite: * /target{uri}
caddy.handle_path.1_reverse_proxy: {{upstreams}}

Proxying all websocket requests, and all requests to /api*, to the container

caddy: example.com
[email protected]_header: Connection *Upgrade*
[email protected]_header: Upgrade websocket
caddy.0_reverse_proxy: @ws {{upstreams}}
caddy.1_reverse_proxy: /api* {{upstreams}}

Proxying multiple domains, with certificates for each

caddy: example.com, example.org, www.example.com, www.example.org
caddy.reverse_proxy: {{upstreams}}

Redirecting

caddy: example.com
caddy.redir_0: /favicon.ico  /alternative/icon.ico 302
caddy.redir_1: /photo.png    /updated-photo.png    302

More community-maintained examples are available in the Wiki.

Docker configs

Note: This is for Docker Swarm only. Alternatively, use CADDY_DOCKER_CADDYFILE_PATH or -caddyfile-path

You can also add raw text to your Caddyfile using Docker configs. Just add Caddy label prefix to your configs and the whole config content will be inserted at the beginning of the generated Caddyfile, outside any server blocks.

Here is an example

Proxying services vs containers

Caddy docker proxy is able to proxy to swarm services or raw containers. Both features are always enabled, and what will differentiate the proxy target is where you define your labels.

Services

To proxy swarm services, labels should be defined at service level. With your compose.yaml, labels should be inside deploy, like:

services:
  foo:
    deploy: # <-- labels should be _inside_ `deploy`
      labels:
        caddy: service.example.com
        caddy.reverse_proxy: {{upstreams}}

Caddy will use service DNS name as target or all service tasks IPs, depending on configuration proxy-service-tasks.

Containers

To proxy containers, labels should be defined at container level. With your compose.yaml, labels should be outside deploy, like:

services:
  foo:
    labels:
      caddy: service.example.com
      caddy.reverse_proxy: {{upstreams}}

Execution modes

Each caddy docker proxy instance can be executed in one of the following modes.

Server

Acts as a proxy to your Docker resources. The server starts without any configuration, and will not serve anything until it is configured by a "controller".

In order to make a server discoverable and configurable by controllers, you need to mark it with label caddy_controlled_server and define the controller network via CLI option controller-network or environment variable CADDY_CONTROLLER_NETWORK.

Server instances doesn't need access to Docker host socket and you can run it in manager or worker nodes.

Configuration example

Controller

Controller monitors your Docker cluster, generates Caddy configuration, and pushes it to all servers it finds in your Docker cluster.

When controller instances are connected to more than one network, it is also necessary to define the controller network via CLI option controller-network or environment variable CADDY_CONTROLLER_NETWORK.

Controller instances require access to Docker host socket.

A single controller instance can configure all server instances in your cluster.

:warning: Controller mode requires server nodes to serve traffic.

Configuration example

Standalone (default)

This mode executes a controller and a server in the same instance and doesn't require additional configuration.

Configuration example

Options

This plugin extends Caddy's CLI with the caddy docker-proxy command. Each option below can be configured either as a CLI flag or via its corresponding environment variable — use whichever fits your deployment.

Run caddy help docker-proxy to see the raw flag output.

Check the examples folder to see how to set them in a Docker Compose file.

Viewing the generated Caddyfile

The generated Caddyfile is automatically saved to Caddyfile.autosave inside Caddy's config directory whenever it changes. The full path is logged at startup and depends on the environment (typically /config/caddy/Caddyfile.autosave in Docker). You can inspect it to troubleshoot label-to-Caddyfile conversion issues:

docker exec <caddy-container> cat /config/caddy/Caddyfile.autosave

Docker images

Docker images are available at Docker hub: https://hub.docker.com/r/lucaslorentz/caddy-docker-proxy/

Choosing the version numbers

The safest approach is to use a full version numbers like 0.1.3. That way you lock to a specific build version that works well for you.

But you can also use partial version numbers like 0.1. That means you will receive the most recent 0.1.x image. You will automatically receive updates without breaking changes.

Chosing between default or alpine images

Our default images are very small and safe because they only contain Caddy executable. But they're also quite hard to troubleshoot because they don't have shell or any other Linux utilities like curl or dig.

The alpine images variant are based on the Linux Alpine image, a very small Linux distribution with shell and basic utilities tools. Use -alpine images if you want to trade security and small size for a better troubleshooting experience.

CI images

Images with the ci tag suffix means they were automatically generated by automated builds. CI images reflect the current state of master branch and their stability is not guaranteed. You may use CI images if you want to help testing the latest features before they're officially released.

ARM architecture images

Currently we provide linux x86_64 images by default.

You can also find images for other architectures like arm32v6 images that can be used on Raspberry Pi.

Windows images

We recently introduced experimental windows containers images with the tag suffix nanoserver-ltsc2022.

Be aware that this needs to be tested further.

This is an example of how to mount the windows Docker pipe using CLI:

$ docker run --rm -it -v //./pipe/docker_engine://./pipe/docker_engine lucaslorentz/caddy-docker-proxy:ci-nanoserver-ltsc2022

Custom images

If you need additional Caddy plugins, or need to use a specific version of Caddy, then you may use the builder variant of the official Caddy Docker image to make your own Dockerfile.

The main difference from the instructions on the official image is that you must override CMD to have the container run using the caddy docker-proxy command provided by this plugin.

ARG CADDY_VERSION=2.6.1
FROM caddy:${CADDY_VERSION}-builder AS builder

RUN xcaddy build \
    --with github.com/lucaslorentz/caddy-docker-proxy/v2 \
    --with <additional-plugins>

FROM caddy:${CADDY_VERSION}-alpine

COPY --from=builder /usr/bin/caddy /usr/bin/caddy

CMD ["caddy", "docker-proxy"]

Connecting to Docker Host

The default connection to Docker host varies per platform:

• At Unix: unix:///var/run/docker.sock
• At Windows: npipe:////./pipe/docker_engine

You can modify Docker connection using the following environment variables:

• DOCKER_HOST: to set the URL to the Docker server.
• DOCKER_API_VERSION: to set the version of the API to reach, leave empty for latest.
• DOCKER_CERT_PATH: to load the TLS certificates from.
• DOCKER_TLS_VERIFY: to enable or disable TLS verification; off by default.

Volumes

On a production Docker swarm cluster, it's very important to store Caddy folder on persistent storage. Otherwise Caddy will re-issue certificates every time it is restarted, exceeding Let's Encrypt's quota.

To do that, map a persistent Docker volume to /data folder.

For resilient production deployments, use multiple Caddy replicas and map /data folder to a volume that supports multiple mounts, like Network File Sharing Docker volumes plugins.

Multiple Caddy instances automatically orchestrate certificate issuing between themselves when sharing /data folder.

Trying it

With Docker Compose (compose.yaml)

Clone this repository.

Deploy the compose file to swarm cluster:

$ docker stack deploy -c examples/standalone.yaml caddy-docker-demo

Wait a bit for services to startup...

Now you can access each service/container using different URLs

$ curl -k --resolve whoami0.example.com:443:127.0.0.1 https://whoami0.example.com
$ curl -k --resolve whoami1.example.com:443:127.0.0.1 https://whoami1.example.com
$ curl -k --resolve whoami2.example.com:443:127.0.0.1 https://whoami2.example.com
$ curl -k --resolve whoami3.example.com:443:127.0.0.1 https://whoami3.example.com
$ curl -k --resolve config.example.com:443:127.0.0.1 https://config.example.com
$ curl -k --resolve echo0.example.com:443:127.0.0.1 https://echo0.example.com/sourcepath/something

After testing, delete the demo stack:

$ docker stack rm caddy-docker-demo

With run commands

$ docker run --name caddy -d -p 443:443 -v /var/run/docker.sock:/var/run/docker.sock lucaslorentz/caddy-docker-proxy:ci-alpine

$ docker run --name whoami0 -d -l caddy=whoami0.example.com -l "caddy.reverse_proxy={{upstreams 80}}" -l caddy.tls=internal traefik/whoami

$ docker run --name whoami1 -d -l caddy=whoami1.example.com -l "caddy.reverse_proxy={{upstreams 80}}" -l caddy.tls=internal traefik/whoami

$ curl -k --resolve whoami0.example.com:443:127.0.0.1 https://whoami0.example.com
$ curl -k --resolve whoami1.example.com:443:127.0.0.1 https://whoami1.example.com

$ docker rm -f caddy whoami0 whoami1

Building it

You can build Caddy using xcaddy or caddy docker builder.

Use module name github.com/lucaslorentz/caddy-docker-proxy/v2 to add this plugin to your build.