The next GoReleaser version will have a more flexible Docker configuration format. In this post we’ll explore it a bit.

When I first added Docker support on GoReleaser, I did a couple of assumptions - and most of them were wrong.

At that time, the config file looked like this:

  binary: foo
	goos: linux
	goarch: amd64
	goarm: ''
	image: bar/foo
  dockerfile: Dockerfile

It didn’t allow multiple tags, registries or anything like that… at least not without building the image several times.

Of course, some users needed those things, and the it went through a few interactions:

And now, I think we finally arrived at something closer to the what will be on 1.0:

    goos: linux
    goarch: amd64
    goarm: ''
    - mybinary
    - otherbinary
    - "myuser/myimage:latest"
    - "myuser/myimage:{{ .Tag }}"
    skip_push: false
    dockerfile: Dockerfile
    - "--label=org.label-schema.schema-version=1.0"
    - "--label=org.label-schema.version={{.Version}}"
    - "{{.ProjectName}}"
    - "--build-arg=FOO={{.ENV.Bar}}"
    - config.yml

This allows several things that were not possible before:

  • more than one binary in the same image
  • several images names, registries and tags on a single build
  • extra files
  • build arguments
  • templating on image names
  • control wether to push the images or not

Lot’s of things you can do with a rather simple config format!

The only feature that is still missing is multiple platforms support, which should be added before 1.0.

A thing that I’m tried hard to do from the start is to keep it possible to build docker images without running goreleaser and without changing any configs. This was true on the first format is still true as today!

For example, imagine you have a simple project, more or less like this:

├── Dockerfile
├── config
│   └── foo.conf
├── .goreleaser.yml
└── main.go

Your Dockerfile may look like:

FROM scratch
ADD foo /usr/bin/foo
ADD config/foo.conf /etc/foo.conf

So, as a developer, you would go build your project and then simply docker build:

go build -o foo
docker build -t caarlos0/goreleaser-docker-test .

And that would work, as expected.

You could then automate that using GoReleaser by creating a .goreleaser.yaml config file that looks like this:

project_name: foo
- env:
    - foo
    - "caarlos0/goreleaser-docker-test:latest"
    - "caarlos0/goreleaser-docker-test:{{ .Tag }}"
    - config/foo.conf

And then, when you want to release, you can just run goreleaser and it will do everything for you, using the same Dockerfile.

This allows for quick interactions while changing dockerfiles or to test other things. Simple, yet, powerful!

Note that it would also work with multiple binaries, given that you do the required changes on the config files.

So, that’s it. Few words just to point out how things are evolving. Looking forward to make releasing easier, better and more reliable for everyone!