10 October 2021

Tools You Should Know About: Terraform

In a nutshell

From the homepage:

Terraform is an open-source infrastructure as code software tool that provides a consistent CLI workflow to manage hundreds of cloud services. Terraform codifies cloud APIs into declarative configuration files.

The terms "infrastructure" and "cloud services" in that description are often thought of meaning the obvious big ones like AWS or Azure. Terraform is, in fact, much broader than that.

Why you should know about it

As a programmer in the twenty-first century, you interact with cloud services. If you don't yet know about Terraform, you manage these services through a mix of clicking on web UIs, curl commands and support tickets raised with your infrastructure team.

There's a better way. Terraform lets you turn pretty much any web service configuration into a text file. This is very powerful, because it means you can manage your web services with git.

git solves a number of problems people typically have with infrastructure:

  • Tracking the state of things over time.
  • Keeping an audit log of who made what change.
  • Building blocks for change requests through branching & merging.


A common language

As mentioned, many people who have heard of Terraform think of it as a common language to manage cloud resources on AWS, Azure & GCP. This is missing the point in two ways.

First, Terraform is most definitely not a way to paper over the differences between different cloud providers. If you're hoping that using Terraform will make it "easy" to switch cloud provider, or provide you with a way to write a single configuration that you can deploy against any provider, you're in for a world of pain. That way lies madness, and Terraform knows it. It's not even trying to go there.

While Terraform does offer a single language regardless of the provider, it's better to think of it as a common syntax. Just like, say, most of these provders let you interact with them through JSON blobs sent over HTTP. They're all using the JSON language, but that doesn't mean the same JSON blob will work across any two of them.

It's nice not to have to learn a new syntax whenever you have to interact with a new provider, but you'll still have to learn all of the concepts related to that provider, and the vast majority of Terraform code you write will still be extremely tied into that one specific provider.

It's a common syntax, not an abstraction layer.

Cloud infrastructure means AWS & co

While I'm sure AWS, GCP and Azure make up the majority of Terraform usage, the tool itself is very agnostic. Any situation where you can manage the state of resources through an API is a good use-case for Terraform. Here are a few examples outside the big three:

  • GitHub: You can use Terraform to manage GitHub resources. That may not seem very useful for a single user account, but in an entreprise context, you can quickly build up a pretty complex setup between organizations, teams, repo access rights, branch protection rules, etc.
  • Kubernetes: You can use Terraform to manage k8s resources. k8s comes with its own configuration language (YAML-based), but you'll typically need to provision the infrastructure under your k8s cluster as well as the resources running on it, and why should you use a different language for that?
  • Artifactory: At work, we use a service called Artifactory for, as the name might suggest, hosting artifacts. Again, this may seem like something you can easily manage through a web UI at first, but from experience I can say you quickly end up with a few dozen repositories and a few dozen users and it gets quite hard to keep track of the web of permissions there, especially when you have more than one admin.

There are many official and even more community-supported providers for Terraform. If you're using any kind of cloud service (a.k.a. remote resource), it may be worth checking if there's a provider for it. Or building your own, but I've never needed to do that myself.

Feature highlights

Text files

Terraform offers a big advantage over manual management through either API calls or web UIs: text files enable all sorts of process automation, as well as history auditing. This is the main selling point: you can collaborate, you can share, you can make change requests, and you can have a good view of the current state by looking at said files.

All using your familiar text editor and your familiar source code management tool.

Common language

Some cloud providers understand the power of text files and thus provide their own configuration language. That's great, but that means you have to learn a new syntax for each one. Also, a lot of cloud providers do not offer anything like this.

The language used by Terraform is by no means perfect; like any language, it has its own set of quirks and idiosyncracies. But investing in learning it gives you a lot of leverage given the range of services you can configure using it.

Drift detection / declarative approach

A perhaps more obvious way to turn a cloud configuration into a set of text files would be to just write down a set of curl commands in a Bash script to create the state you want to end with. Conceptually, that's a good first step towards having some form of audit log.

However, Terraform provides you with something more, which is quite a lot of work to build out yourself: once Terraform has created a resource, it keeps track of it. It will detect if it changes, and offer to change it back so it stays in sync with your provided description. If you remove a resource description from your configuration files, Terraform will also detect that and offer to delete the corresponding resource.

Dependency tracking

If you want to create resources that depend on each other, Terraform will detect that and create them in the correct order (provided there is one). For example, if you want to create a "classic" 3-tier application infrastructure, with three separate machines (database, middleware, frontend), and your configuration reflects the fact that the middleware needs the IP address of the database and the frontend needs the IP address of the middleware, Terraform will create the database first, wait until it gets an IP address assigned, then create the middleware, and so on with the frontend.

This type of dependency tracking works for resources across different providers, too. If you wanted to create an AWS instance to host a website, and then create a StatusCake configuration to monitor that machine, you could express that in a single Terraform configuration, and Terraform would be able to infer that the machine needs to be created first so Terraform can get its IP and put it in the StatusCake configuration.


Going through text files means that, assuming you keep those text files under source control, you get a log of who changed what when.

There's a small caveat there, though: while Terraform will detect changes in resources it has created, and delete said resources if they get removed from the configuration files, Terraform will not touch resources it has not created, and will, in the general case, not report them either.

While it is possible to set up your system such that Terraform is the only way to interact with your infrastructure configuration, Terraform itself does not try to prevent additional manual changes.


Like other entries in this series, the goal of this article is to let you know Terraform exists and give you a sense for when you'd want to use it. Once you have a use-case, you'll have to look elsewhere to actually learn how to use it.

Terraform is a terrific lightweight way to introduce devops habits into your organization. If you're on the ops side and wish developers were more involved, giving them a way to collaborate through text files managed by git is a great first step. If you're a developer and wish you had a better view of what's going on with your infrastructure, getting even view-only access to git repos hosting Terraform files is a nice way to get that.

Tags: tyska terraform