Today we will be diving into each section of our packer json file and their purpose.

This is the third post of a planned four part series:

If you haven’t gotten the gist by now, the permutations of what you can do with packer in terms of building images are nearly endless.  Thus far we have focused on a very narrow case of building a RHEL/CentOS 7 VMware compatible image, to then clone as needed.  However, with an understanding of the different tunables, you can easily adapt a working image build to do many interesting things.  We’ll again be using my sample as a reference to facilitate discussion.  You will also like to checkout the excellent documentation for packer.There are four sections of a packer template:

  • Builders
  • Provisioners
  • Post-Processors
  • Variables

The only truly required section is the builders section, but each section provides certain functionality.


Builders is the section of the packer json template that defines instructions which packer uses to create the actual virtual machine and install the operating system.  There are a number of builders, which allow packer to support a number of different platforms including VMware, OpenStack, Azure, AWS, etc.  Some builders, such as the AWS and VMware ones, contain multiple sub-types to allow for additional flexibility.

For example the VMware builder allows you to build a VM straight from an ISO or utilize an existing VM as a starting point. The key take away to remember about builders is that it is the definition of the virtual machine that you are creating.

Regardless of if the system being built is destined to be a vSphere template, an Amazon EC2 instance, or a vSphere virtual machine that you are provisioning directly; the builder section gives you full control over what your output will look like from a hardware perspective: network connectivity, CPU/Memory configuration, which answer file to use etc.

Speaking of the answer file, you are able to present this to the virtual machine by using a built-in to packer web server, or via a virtual floppy that is generated via a list of files or directories.


Provisoners is the section of the packer json template that allows you define actions that packer should take after installing the operating system is complete by the builder.  There are a number of provisioners which will ultimately allow you to dial in the configuration of your system or template exactly as you need to.  You can choose from Chef, Ansible, PowerShell, Puppet, the shell, or even create your own.  In our example we utilize the following provisioners:

file – This provisioner simply copies files from the machine executing packer to a specified location on the provisioned system, for use later on during the build by a different provisioner, or for longer term use down the road.

shell – This provisioner executes shell commands on the provisioned system.  It could be anything ranging from simple shell commands, to scripts placed there by something like the file provisioner.  There is a different variant of the shell provisoner for Windows.

A cool feature of all provioners is the ability to limit their execution to only specific builds – for instance if your packer json contains definitions for multiple templates/machines.  You can accomplish this by including the “only” directive in the provisioner.

You also have the ability to provide overrides for different builders.  Say for example if you are building a instance that needs a different parameter set for a command.  You can see an example of us using the provisioner override on lines 108 to 124 of our template.


Post-Processors aren’t directly covered in our example.  We actually have what could be called a custom post-processor in the deploy-image.ps1 script.  In the year plus since I originally developed this code, Hashicorp and the community have done a great job of fleshing out the options available to do post processing including taking the machine you built and putting it from say a Fusion or Workstation build environment and depositing it directly to a vSphere environment and running it, or converting to a template (and putting it into a vCenter).

Other options allow for importing it to Amazon, or execute things like our deploy-image.ps1 script directly from packer as opposed to doing it via the Jenkins pipeline like we did.


The last “section” I want to cover is not something that functionally does something to build your virtual machine, but as the name variables implies, it allows you to define a set of variables which can then be defined at run-time to provide a level of dynamic behavior.

Say for example you wanted to build a CentOS or Red Hat Image, but give yourself the flexibility to change certain aspects of the build – such as the base ISO file, or the Katello subscription key.  The variables section allows you to define it with a reasonable default value, which you can then override when you call the packer executable.

You can even reference environment variables of the executing system to manipulate the build – in our example we are reading an environment variable IMAGE_OUTPUT_DIR, which we then reference later when we instruct packer on where to build the image.

Wrapping it up

It’s hard to convey the things you can do with packer in a single series of posts, the possibilities are endless, and it can be a bit intimidating when you are digging in and trying to figure out the best way to leverage it.  Next time, I’ll wrap up this series by brain dumping some of the cool things I think you can accomplish with Packer to make your life as a builder of things easier, and give you more time for coffee and other fun activities.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.