6.3 KiB

Raw Blame History

layout
docs

Developing Plugins

This page will document how you can develop your own Packer plugins. Prior to reading this, it is assumed that you're comfortable with Packer and also know the basics of how Plugins work, from a user standpoint.

Packer plugins must be written in Go, so it is also assumed that you're familiar with the language. This page will not be a Go language tutorial. Thankfully, if you are familiar with Go, the Go toolchain makes it extremely easy to develop Packer plugins.

Warning! This is an advanced topic. If you're new to Packer, we recommend getting a bit more comfortable before you dive into writing plugins.

Plugin System Architecture

Packer has a fairly unique plugin architecture. Instead of loading plugins directly into a running application, Packer runs each plugin as a separate application. Inter-process communication and RPC is then used to communicate between the many running Packer processes. Packer core itself is responsible for orchestrating the processes and handles cleanup.

The beauty of this is that your plugin can have any dependencies it wants. Dependencies don't need to line up with what Packer core or any other plugin uses, because they're completely isolated into the process space of the plugin itself.

And, thanks to Go's interfaces, it doesn't even look like inter-process communication is occuring. You just use the interfaces like normal, but in fact they're being executed in a remote process. Pretty cool.

Plugin Development Basics

Developing a plugin is quite simple. All the various kinds of plugins have a corresponding interface. The plugin simply needs to implement this interface and expose it using the Packer plugin package (covered here shortly), and that's it!

There are two packages that really matter that every plugin must use. Other than the following two packages, you're encouraged to use whatever packages you want. Because plugins are their own processes, there is no danger of colliding dependencies.

github.com/mitchellh/packer - Contains all the interfaces that you have to implement for any given plugin.
github.com/mitchellh/packer/plugin - Contains the code to serve the plugin. This handles all the inter-process communication stuff.

There are two steps involved in creating a plugin:

Implement the desired interface. For example, if you're building a builder plugin, implement the packer.Builder interface.
Serve the interface by calling the appropriate plugin serving method in your main method. In the case of a builder, this is plugin.ServeBuilder.

A basic example is shown below. In this example, assume the Builder struct implements the packer.Builder interface:

import (
  "github.com/mitchellh/packer/plugin"
)

// Assume this implements packer.Builder
type Builder struct{}

func main() {
	plugin.ServeBuilder(new(Builder))
}

That's it! plugin.ServeBuilder handles all the nitty gritty of communicating with Packer core and serving your builder over RPC. It can't get much easier than that.

Next, just build your plugin like a normal Go application, using go build or however you please. The resulting binary is the plugin that can be installed using standard installation procedures.

The specifics of how to implement each type of interface are covered in the relevant subsections available in the navigation to the left.

Lock your dependencies. Unfortunately, Go's dependency management story is fairly sad. There are various unofficial methods out there for locking dependencies, and using one of them is highly recomended since the Packer codebase will continue to improve, potentially breaking APIs along the way until there is a stable release. By locking your dependencies, your plugins will continue to work with the version of Packer you lock to.

Logging and Debugging

Plugins can use the standard Go log package to log. Anything logged using this will be available in the Packer log files automatically. The Packer log is visible on stderr when the PACKER_LOG environmental is set.

Packer will prefix any logs from plugins with the path to that plugin to make it identifiable where the logs come from. Some example logs are shown below:

2013/06/10 21:44:43 ui: Available commands are:
2013/06/10 21:44:43 Loading command: build
2013/06/10 21:44:43 packer-command-build: 2013/06/10 21:44:43 Plugin minimum port: 10000
2013/06/10 21:44:43 packer-command-build: 2013/06/10 21:44:43 Plugin maximum port: 25000
2013/06/10 21:44:43 packer-command-build: 2013/06/10 21:44:43 Plugin address: :10000

As you can see, the log messages from the "build" command plugin are prefixed with "packer-command-build". Log output is extremely helpful in debugging issues and you're encouraged to be as verbose as you need to be in order for the logs to be helpful.

Plugin Development Tips

Here are some tips for developing plugins, often answering common questions or concerns.

Naming Conventions

It is standard practice to name the resulting plugin application in the format of packer-TYPE-NAME. For example, if you're building a new builder for CustomCloud, it would be standard practice to name the resulting plugin packer-builder-custom-cloud. This naming convention helps users identify the purpose of a plugin.

Testing Plugins

While developing plugins, you can configure your Packer configuration to point directly to the compiled plugin in order to test it. For example, building the CustomCloud plugin, I may configure packer like so:

{
  "builders": {
    "custom-cloud": "/an/absolute/path/to/packer-builder-custom-cloud"
  }
}

This would configure Packer to have the "custom-cloud" plugin, and execute the binary that I am building during development. This is extremely useful during development.

Distributing Plugins

It is recommended you use a tool like goxc in order to cross-compile your plugin for every platform that Packer supports, since Go applications are platform-specific. goxc will allow you to build for every platform from your own computer.

6.3 KiB Raw Blame History