Provision

In the previous page of this guide, you created your first image with Packer. The image you just built, however, was basically just a repackaging of a previously existing base AMI. The real utility of Packer comes from being able to install and configure software into the images as well. This stage is also known as the provision step. Packer fully supports automated provisioning in order to install software onto the machines prior to turning them into images.

In this section, we're going to complete our image by installing Redis on it. This way, the image we end up building actually contains Redis pre-installed. Although Redis is a small, simple example, this should give you an idea of what it may be like to install many more packages into the image.

Historically, pre-baked images have been frowned upon because changing them has been so tedious and slow. Because Packer is completely automated, including provisioning, images can be changed quickly and integrated with modern configuration management tools such as Chef or Puppet.

»Configuring Provisioners

Provisioners are configured as part of the template. We'll use the built-in shell provisioner that comes with Packer to install Redis. Modify the example.json template we made previously and add the following. We'll explain the various parts of the new configuration following the code block below.

{
  "variables": ["..."],
  "builders": ["..."],

  "provisioners": [
    {
      "type": "shell",
      "inline": [
        "sleep 30",
        "sudo apt-get update",
        "sudo apt-get install -y redis-server"
      ]
    }
  ]
}

To configure the provisioners, we add a new section provisioners to the template, alongside the builders configuration. The provisioners section is an array of provisioners to run. If multiple provisioners are specified, they are run in the order given.

By default, each provisioner is run for every builder defined. So if we had two builders defined in our template, such as both Amazon and DigitalOcean, then the shell script would run as part of both builds. There are ways to restrict provisioners to certain builds, but it is outside the scope of this getting started guide. It is covered in more detail in the complete documentation.

The one provisioner we defined has a type of shell. This provisioner ships with Packer and runs shell scripts on the running machine. In our case, we specify two inline commands to run in order to install Redis.

»Build

With the provisioner configured, give it a pass once again through packer validate to verify everything is okay, then build it using packer build example.json. The output should look similar to when you built your first image, except this time there will be a new step where the provisioning is run.

The output from the provisioner is too verbose to include in this guide, since it contains all the output from the shell scripts. But you should see Redis successfully install. After that, Packer once again turns the machine into an AMI.

If you were to launch this AMI, Redis would be pre-installed.

This is just a basic example. In a real world use case, you may be provisioning an image with the entire stack necessary to run your application. Or maybe just the web stack so that you can have an image for web servers pre-built. This saves tons of time later as you launch these images since everything is pre-installed. Additionally, since everything is pre-installed, you can test the images as they're built and know that when they go into production, they'll be functional.