Managing multiple stages with Terraform

This content is more than 4 years old and the cloud moves fast so some information may be slightly out of date.



Managing multiple environments in Terraform

Introduction

I recently started learning Terraform. For those who haven’t encountered it: Terraform is in essence a framework to describe Infrastructure as code by Hashicorp. When I began doing that, I was struggling with the staging-concept of Terraform. I did my research and came upon numerous 1 articles and blogs that described ways to manage (multiple) environments or stages in Terraform2.

Since I wasn’t really happy with the other solutions and there didn’t seem to be a canonical way to handle multiple environments, I decided to try and figure out my own solution.

(You can find the code for all of this on Github)

Background

What is an environment or stage?

In this context I’m refering to an instance of an application and/or a group of infrastructure components.

Why do I need multiple environments or stages?

When Deploying Applications and/or Infrastructure you typically have multiple instances of that Application and/or Infrastructure.

There are multiple reasons why you might want to do this, some include:

  • Separate instances for production, qa, testing and development
  • Separate instances for each customer
  • Blue/Green deployments

What are the specific challenges in Terraform?

Terraform maintains it’s perception or representation of an environment’s state in a .tfstate-File - by default that’s called terraform.tfstate. This file is very important, because without it, Terraform has no idea, which resources it has deployed and how they look like3.

Because the good folks at Hashicorp know this, they invented a way to deal with this problem: Remote Backends for Remote State. I won’t go into much detail about this, but in essence this keeps a copy of your terraform.tfstate in a remote location and may enable collaboration. As AWS afficionados we obviously prefer the S3-Backend :).

Different environments might use different kinds of backends to store there state, so we need to have a way to separate backends for separate environments.

However, that’s only half of the story:

Your environments probably use different configurations. This can include separate Webservice-Endpoints for productive and non-productive stages, different server-sizing and much more. Terraform has - as many other tools do as well - the concept of Variables and Variable-Files (.tfvars) to deal with this. The latter can be passed as input to some Terraform commands. This works fine, but is a little annoying:

terraform apply -var-file=test.tfvars

This is the second challenge we need to deal with: handle environment-specific variables.


A word on Workspaces: If you have already heard about them, you might think that workspaces are an ideal solution for this, but according to the documentation “Workspaces are technically equivalent to renaming your state file” - and it goes on:

In particular, organizations commonly want to create a strong separation between multiple deployments of the same infrastructure serving different development stages (e.g. staging vs. production) or different internal teams. In this case, the backend used for each deployment often belongs to that deployment, with different credentials and access controls. Named workspaces are not a suitable isolation mechanism for this scenario.


For my solution this boils down to the design goals discussed in the next section.

Design Goals

The solution should be:

  • easy to set up
  • easy to use
  • able to support environment-specific backends
  • able to support environment-specific variable files
  • usable on Linux (if it works on Mac, that would be a nice benefit as well)

Solution

My solution makes a few assumptions about its environment:

  • it runs on Linux
  • the user is lazy (you shouldn’t extrapolate from yourself to others, but I can’t help myself…)
  • the project is structured like this:
├── environments
│   ├── production
│   │   ├── backend.config
│   │   └── variables.tfvars
│   ├── qa
│   │   ├── backend.config
│   │   └── variables.tfvars
│   └── test
│       ├── backend.config
│       └── variables.tfvars
├── main.tf
└── switch_environment.sh

You might already be able to spot the pattern here - every directory under the environments directory is an environment or stage we can deploy into. Each of them has their own set of variables (variables.tfvars) and backend configuration (backend.config). You can use those files to configure your environment-specific values.

The start of the show is switch_environment.sh - let’s look at this in more detail.

Implementation

The solution consists of a simple shell-script, which does two things:

  1. Run terraform init with the appropriate backend configuration
  2. Export a function called tf which is something like a smart alias for terraform - more on that later.

The script makes sure, the relevant backend configuration exists before running terraform init with it. It also verifies we have a variables.tfvars before exporting the tf function.

#!/usr/bin/env bash

# How to: . switch_environment.sh ENVIRONMENT_NAME

STAGE=$1

if [[ ! -d "environments/${STAGE}" ]]; then
    echo "The environment '${STAGE}' doesn't exist under environments/ - please check the spelling!"
    echo "These environments are available:"
    ls environments/
    return 1
fi

if [[ -f environments/${STAGE}/backend.config ]]; then
    # Configure the Backend
    echo "Running: terraform init -backend-config=environments/${STAGE}/backend.config ."
    terraform init -backend-config=environments/${STAGE}/backend.config .
else
    echo "The backend configuration is missing at environments/${STAGE}/backend.config!"
    return 2
fi

if [[ -f "environments/${STAGE}/variables.tfvars" ]]; then
    # Configure a function that runs terraform with the variables attached
    # --> "tf apply" will run "terraform apply -var-file=path/to/variables.tfvars"
    echo "The alias 'tf' runs terraform with the correct variable file when appropriate"
    tf () {

        # List of commands that can accept the -var-file argument
        sub_commands_with_vars=(apply destroy plan)

        # List of commands that accept the backend argument
        sub_commands_with_backend=(init)

        # ${@:2} means that we append all of the arguments after tf init

        if [[ " ${sub_commands_with_vars[@]} " =~ " $1 " ]]; then
            # Only some of the subcommands can work with the -var-file argument
            echo "Running: terraform $1 -var-file=environments/${STAGE}/variables.tfvars ${@:2}"
            terraform $1 -var-file=environments/${STAGE}/variables.tfvars ${@:2}
        elif [[ " ${sub_commands_with_backend[@]} " =~ " $1 " ]]; then
            # Only some sub commands require the backend configuration
            echo "Running: terraform init -backend-config=environments/${STAGE}/backend.config ${@:2}"
            terraform init -backend-config=environments/${STAGE}/backend.config ${@:2}
        else
            echo "Running: terraform $@"
            terraform $@
        fi

    }
else
    echo "Couldn't find the variables file here: environments/${STAGE}/variables.tfvars "
    echo "Won't set up the tf function!"
    return 3
fi

How to

Switching to an existing environment

Run the following command substituting myenvironment with the name of the environment you’d like to switch to:

. switch_environment.sh myenvironment

Running Terraform commands /using tf

A feature of the environment switching script is, that it provides the tf shortcut for running terraform commands. This is not just an alias - it automatically adds the environments’ variable files to the commands that support them - here’s an example:

tf apply

is the short form for

terraform apply -var-file=environments/currentstage/variables.tfvars

This should work for pretty much any terraform command - it even appends the backend configuration when you run:

tf init .

Setting up a new environment

  1. Create a new directory with the name of your environment under environments/

  2. Add a backend.config under environments/new_environment/ with your configuration for the Terraform backend

  3. Add a variables.tfvars under environments/new_environment/ with your configuration for the environments’ variables

  4. Switch to the new environment using:

    . switch_environment.sh new_environment
    
  5. Read the next paragraph for a caveat about empty backends

Switching to an environment with an empty backend storage

If you just set up your shiny new configuration for an even shinier new backend and you run the switch command . switch_environment shiny_new_one for the first time, Terraform may prompt you something like this:

Backend configuration changed!

Terraform has detected that the configuration specified for the backend
has changed. Terraform will now check for existing state in the backends.


Do you want to copy existing state to the new backend?
  Pre-existing state was found while migrating the previous "s3" backend to the
  newly configured "s3" backend. No existing state was found in the newly
  configured "s3" backend. Do you want to copy this state to the new "s3"
  backend? Enter "yes" to copy and "no" to start with an empty state.

  Enter a value: no

This basically means: “Hey, we noticed you changed to an empty environment, would you like us to copy the current state to that?” - usually the answer should be no, since we created a new backend precisely to start with an empty slate.

How do I use it in my pipeline

If you a Pipeline to deploy the Terraform configuration in your account, you might be wondering if you could use this script for that.

From a technical perspective there is nothing stopping you, but I’d advise against it - mostly because of personal preference. When collaborating with others, I prefer transparent solutions.

What I mean by that is that I don’t like scripts that hide some commands and do some magic, when there is an obvious way to solve the problem. In build pipelines I want to see what exactly is being executed, as this makes debugging easier, this means I’d prefer to pass the correct tfvars-File as an argument, since I have to type it only once anyways.

Wrapping up

Today I shared with you a script that might be useful if you work with multiple environments within your Terraform stack. The script is most likely not perfect - but it’s currently good enough for my purpose. If you have ideas on how to improve it or some other kind of feedback - I’d love to hear from you. Just message me on twitter: @Maurice_Brg or write me an email.

You can find the code for all of this on Github


  1. Just a small sample:

     ↩︎
  2. By the way - I’m going to use the terms environment and stage interchangeably. ↩︎

  3. If you lose that file, your application will continue to work, but you could end up with a bunch of expensive resources distributed across the world and no idea how they fit together. ↩︎

Similar Posts You Might Enjoy

tRick: simple network 2 - Geschwindigkeit

Vergleich Infrastructure as Code (IaC) Frameworks - tRick Alle Posts Abstraktion und Lines of Code Geschwindigkeit Diversity (polyglott), Tooling, Fazit Benchmark Ausführungsgeschindigkeit Ausführungsgeschwindigkeit Direkt aus dem tRick Repository wird mehrfach (n=10) der Zyklus Build -> Check -> Deploy -> Remove ausgeführt. Damit sollen Cache Effekte statistisch gemittelt werden. Dazu nehme ich das Tool hyperfine zur Hilfe. Es führt Kommandos automatisch mehrfach aus und mittelt die Ergebnisse. Meine Annahme ist es, dass Terraform vorne liegt, da das Programm selber statisch kompiliert in go geschrieben ist. Außerdem geht die Ausführung direkt auf die API. - by Gernot Glawe

tRick: simple network 3 - Diversity (polyglott), Tooling, Fazit

Vergleich Infrastructure as Code (IaC) Frameworks - tRick Teil 3 Alle Posts Abstraktion und Lines of Code Geschwindigkeit Diversity (polyglott), Tooling, Fazit - by Gernot Glawe

Building Lambda with terraform

Note: An updated version of this post is available here Building Lambda Functions with Terraform Introduction Many of us use Terraform to manage our infrastructure as code. As AWS users, Lambda functions tend to be an important part of our infrastructure and its automation. Deploying - and especially building - Lambda functions with Terraform unfortunately isn’t as straightforward as I’d like. (To be fair: it’s very much debatable whether you should use Terraform for this purpose, but I’d like to do that - and if I didn’t, you wouldn’t get to read this article, so let’s continue) - by Maurice Borgmeier