packer-cn/website/source/docs/builders/docker.html.markdown

4.4 KiB

layout
docs

Docker Builder

Type: docker

The Docker builder builds Docker images using Docker. The builder starts a Docker container, runs provisioners within this container, then exports the container for re-use.

Packer builds Docker containers without the use of Dockerfiles. By not using Dockerfiles, Packer is able to provision containers with portable scripts or configuration management systems that are not tied to Docker in any way. It also has a simpler mental model: you provision containers much the same way you provision a normal virtualized or dedicated server. For more information, read the section on Dockerfiles.

The Docker builder must run on a machine that has Docker installed. Therefore the builder only works on machines that support Docker (modern Linux machines). If you want to use Packer to build Docker containers on another platform, use Vagrant to start a Linux environment, then run Packer within that environment.

Basic Example

Below is a fully functioning example. It doesn't do anything useful, since no provisioners are defined, but it will effectively repackage an image.

{
  "type": "docker",
  "image": "ubuntu",
  "export_path": "image.tar"
}

Configuration Reference

Configuration options are organized below into two categories: required and optional. Within each category, the available options are alphabetized and described.

Required:

  • export_path (string) - The path where the final container will be exported as a tar file.

  • image (string) - The base image for the Docker container that will be started. This image will be pulled from the Docker registry if it doesn't already exist.

Optional:

  • pull (bool) - If true, the configured image will be pulled using docker pull prior to use. Otherwise, it is assumed the image already exists and can be used. This defaults to true if not set.

  • run_command (array of strings) - An array of arguments to pass to docker in order to run the container. By default this is set to ["run", "-d", "-i", "-t", "-v", "{{.Volumes}}", "{{.Image}}", "/bin/bash"]. As you can see, you have a couple template variables to customize, as well.

Using the Artifact

Once the tar artifact has been generated, you will likely want to import, tag, and push it to a container repository. Packer can do this for you automatically with the docker-import and docker-push post-processors.

The example below shows a full configuration that would import and push the created image:

{
    "post-processors": [
		[
			{
				"type": "docker-import",
				"repository": "mitchellh/packer",
				"tag": "0.7"
			},
			"docker-push"
		]
	]
}

If you want to do this manually, however, perhaps from a script, you can import the image using the process below:

docker import - registry.mydomain.com/mycontainer:latest < artifact.tar

You can then add additional tags and push the image as usual with docker tag and docker push, respectively.

Dockerfiles

This builder allows you to build Docker images without Dockerfiles.

With this builder, you can repeatably create Docker images without the use a Dockerfile. You don't need to know the syntax or semantics of Dockerfiles. Instead, you can just provide shell scripts, Chef recipes, Puppet manifests, etc. to provision your Docker container just like you would a regular virtualized or dedicated machine.

While Docker has many features, Packer views Docker simply as an LXC container runner. To that end, Packer is able to repeatably build these LXC containers using portable provisioning scripts.

Dockerfiles have some additional features that Packer doesn't support which are able to be worked around. Many of these features will be automated by Packer in the future:

  • Dockerfiles will snapshot the container at each step, allowing you to go back to any step in the history of building. Packer doesn't do this yet, but inter-step snapshotting is on the way.

  • Dockerfiles can contain information such as exposed ports, shared volumes, and other metadata. Packer builds a raw Docker container image that has none of this metadata. You can pass in much of this metadata at runtime with docker run.