Terraform Modules Restructured

Terraform
Modules
Restructured
✓ Reusable
✓ Composable
✓ Automated Tests
✓ Confidence to Make Changes
✓ Implementation Abstraction

These slides are heavily influenced by the slides and talks from
Yevgeniy Brikman from terragrunt.io
It goes hand in hand with the following talk:
https://blog.gruntwork.io/5-lessons-learned-from-writing-over-300-000-lines-of-infrastructure-code-36ba7fadeac1
Credits

Ami Mahloof
Senior Cloud Architect at DoIT International.
LinkedIn Profile
Medium blog posts
Who Am I?
At DoiT International, we tackle complex problems of scale which are
sometimes unique to internet-scale customers while using our expertise
in resolving problems, coding, algorithms, complexity analysis, and large-
scale system design.

1. Introduction to Terraform
2. Module Anatomy
3. Modules Structure
4. Testing
5. Terraform Modules Best Practices
6. Migrating Existing Infrastructure Into New Code Structure
Outline

An Introduction To Terraform
⌾ Terraform can manage existing and new infrastructure
⌾ Talk to multiple cloud/infrastructure providers
⌾ Ensure creation and consistency
⌾ Single DSL (Domain Specific Language) to express API agnostic calls
⌾ Preview changes, destroy when needed
⌾ Single source of truth infrastructure state
⌾ Even order a pizza from Domino’s
Terraform is a tool for building, changing, and versioning infrastructure
safely and efficiently

AnIntroductiontoTerraform
Just like in code a function has inputs (arguments) and outputs
(attributes)
The following sudo code creates an EC2 instance from the given args
function create_ec2(name, type) {
ec2 = aws.create_instance(name, type)
print ec2.instance_ip
}
HashiCorp Configuration Language (HCL) Syntax

Mapping A Code To HCL Syntax
}
Inputs (arguments)
outputs (attributes)

}
create_ec2("test", "t2.micro")
resource "aws_ec2_instance" "create_ec2" {
name = "test"
type = "t2.micro"
}
output "instance_ip" {
value = aws_ec2_instance.create_ec2.ipv4_address
}
name label

}
name = "test"
type = "t2.micro"
}
}
provider resource API

}
name = "test"
type = "t2.micro"
}
}
resource arguments

}
name = "test"
type = "t2.micro"
}
}
output values

resource blocks are for create API calls:
resource "aws_ec2_instance" "create_ec2" {...
data blocks are for get API calls:
data "aws_ec2_instance" "instance_data" {
name = "test"
}
output "az" {
value = data.aws_ec2_instance.instance_data.availbility_zone
}

⌾ JSON representation of known
infrastrurcture state provisioned by
terraform
⌾ Stored in file or externally
⌾ Locking (useful for team working on
the same project or tasks)
⌾ Source of truth for infrastructure
Terraform State File
{
"version": 4,
"terraform_version": "0.12.6",
"serial": 18,
"lineage": "b3bfb3fc-8417-cc89-d87f-6ab0008e2056",
"outputs": {
"group-id": {
"value": "6521262259226325019",
"type": "string"
}
},
"resources": [
{
"mode": "data",
"type": "template_file",
"name": "filter_pattern",
"provider": "provider.template",
"instances": ...
terraform.tfstate

Monolithic Terraform
⌾ One/several huge files - fear of making changes, no reusability, a mistake
anywhere can break everything
⌾ Hard to find/debug - variables and sections are harder to find when one needs
to make a change
⌾ Guess work - going back and forth between variables and resources just to
understand what is required and what is the default
⌾ Slower development cycles - increased time and effort needed to start
working with it
Issues with monolithic Terraform:

10,000 ft View Approach
Since Terraform will combine all the files into a plan, we can use that to
create smaller files with better visibility using the following simple module
anatomy.
Being able to quickly find what you’re looking for during development and
debugging an issue is crucial when working with Terraform.

Module Anatomy
The module anatomy is a scaffold that provides better visibility and guidelines
for developing, and working with Terraform modules.
Since Terraform compiles all resources in all files into an execution plan, we can
use that to create better visibility and readability.
There are no hard-coded values, as each hard-coded value becomes a default
variable, and every attribute is a variable.

Development of a module is done through the examples folder which holds a
main.tf file with:
⌾ Hard-coded values for all variables
⌾ Lock down a specific version for a Terraform provider
⌾ State location
⌾ Terraform version
This will serve as a usage example when you finish development on the module
ModuleAnatomy

terraform {
backend "s3" {
region = "eu-west-3"
bucket = "some-s3-bucket"
key = "dev/eu-west-3/infrastructure"
}
required_version = ">=12.6"
}
# This is where you setup the provider to use with the module
provider "aws" {
version = "~> 2.0"
region = "us-east-1"
}
module "route53_record_name_cname_exmaple" {
source = "../"
domain_name = "tf.domain.com"
value = "1.2.3.4"
}
ModuleAnatomy examples/main.tf

⌾ examples - a folder containing examples for usage
⌾ test - Go Terratest folder
⌾ data.tf - Terraform data sources
⌾ main.tf - resources to be created
if it’s over 30 lines long, break it into files with names
applied for resources i.e., autoscaling.tf, ec2.tf, etc.
⌾ outputs.tf - outputs for the module
⌾ README.md - clear inputs/outputs and description for the
module as well as usage
⌾ default-variables - variables with default values
⌾ required-variables - variables with values that are required
ModuleAnatomy

Module Anatomy - Quick scaffolding with bash:
➜ export module_name="sample"
➜ mkdir -p $module_name/examples $module_name/test
➜ cd $module_name && touch
main.tf
versions.tf
default-variables.tf
required-variables.tf
outputs.tf
data.tf
ModuleAnatomy

3-TierModulesStructure
Create and extend your own library of primitives building blocks (resources)
The 3-Tier Module-Based Hierarchy Structure

Build services from these primitives building blocks

Deploy end-to-end environments from services (live deployments)

What are the 3-Tier modules structure major benefits:
⌾ Hide all lower level details to allow the end user to focus
on building the infrastructure
⌾ Each tier is tested providing a quicker
development/debugging cycle
⌾ Provides the confidence needed to make changes

The goal is to isolate each (live) environment (dev, staging, production),
then take each component in that environment and break it up into a
generic service module, and for each generic service module break it
into resource modules.
Restructuring Existing Infrastructure

Break your architecture code down by live environment
terraform-live-envs
L dev
L vpc
L mysql
L kubernetes
L staging
L vpc
L mysql
L kubernetes
L production
L vpc
L mysql
L kubernetes

Then by service (infrastructure type)
terraform-live-envs
L dev
L vpc
L mysql
L kubernetes
L staging
L vpc
L mysql
L kubernetes
L production
L vpc
L mysql
L kubernetes
terraform-services (generic modules)
L gke
L vpc
L sql
Implement infrastructure in modules

Build complex modules from smaller, simpler modules
terraform-live-envs
L dev
L vpc
L mysql
L kubernetes
L staging
L vpc
L mysql
L kubernetes
L production
L vpc
L mysql
L kubernetes
terraform-services
L gke
L vpc
L sql
terraform-resources
L vpc
L sql
L instance
L user

Tier-1 Terraform Resources Modules
This is the lowest tier
terraform-resources is a folder containing modules with a single resource to be
created
These resource modules are creating only one thing
These modules should have an output.tf file with outputs values providing
information on the resource created
This information can be used to create hard dependencies between modules
(required by the 2nd tier)

Tier-2 Terraform Services Modules
This is the middle tier
terraform-services is a folder containing modules combining resources modules
together from the terraform-resources folder
Each service module is a generic service that can create multiple versions based on
the variables passed in
Example; an SQL instance module can create postgreSQL or mySQL instance

Tier-3 Terraform-live-envs Modules
This is the top tier
The terraform-live-envs is a folder containing modules implementing the
infrastructure that is deployed
These modules are usually built from the services modules but can also have
resources modules mixed in
Every module attribute is a hard-coded value representing the value that is deployed

Tier-3 terraform-live-envs Modules
Each module should have one single file called main.tf that will contain:
⌾ Terraform state block
⌾ Modules with hard-coded values
⌾ Locals block (shared variables between modules in this file)
⌾ Outputs
This makes for a readable easy-to-use and maintainable deployment file

Terraform Remote State
By default, Terraform stores state locally in a file named terraform.tfstate
This does not scale because there’s no locking or central location to work with
Terraform in a team.
With remote state, Terraform writes the state data to a remote data store, which
can then be shared between all members of a team. Terraform supports storing
state in Terraform Cloud, HashiCorp Consul, Amazon S3, Alibaba Cloud OSS, and
more.

Base Remote State
Often you need to create a base infrastructure for other deployments to use/read
Example:
You might create a VPC in one region only once, but you can deploy multiple
services on that VPC.

To do that, break your deployment into 2-steps:
⌾ step-1-infrastructure
Creates and outputs the VPC information (vpc_id, subnets etc..)
⌾ step-2-some-service
Accepts the remote state location (defined in step-1) as an input that will be
used to read the output information for step-1, and creates the service on that
VPC
BaseRemoteState

BaseRemoteState
terraform {
backend "s3" {
bucket = "my-terraform-bucket"
key = "dev/eu-west-3/infrastructure"
}
}
module "vpc" {
vpc_name = "vpc-dev"
vpc_availability_zones = ["eu-west-3a", "eu-west-3b"]
vpc_private_subnets = ["10.10.1.0/24", "10.10.2.0/24"]
vpc_public_subnets = ["10.10.11.0/24", "10.10.12.0/24"]
}
output "infra" {
value = module.vpc
}
Step 1 - Infrastructure

BaseRemoteState
terraform {
backend "s3" {
bucket = "my-terraform-bucket"
# separate the state file location from the infrastructure state location
key = "dev/eu-west-3/deployment/code-pipeline/nodejs-app"
}
}
module "code_pipeline" {
source = "../codebuild-pipeline"
# these are taken from step-1 terraform backend block
terraform_state_store_region = "eu-west-3"
terraform_state_store_bucket = "my-terraform-bucket"
infrastructure_terraform_state_store_key = "dev/eu-west-3/infrastructure"
}
...
Step 2 - Services on Infrastructure

BaseRemoteState
deployment/step2-codepipeline/main.tf
module "code_pipeline" {
source = "../codebuild-pipeline"
...
}
Codebuild-pipeline module
# Read the information from the remote state file of step 1 infrastructure
data "terraform_remote_state" "infra" {
backend = "s3"
config = {
region = var.terraform_state_store_region
bucket = var.terraform_state_store_bucket
key = var.infrastructure_terraform_state_store_key
}
}
# Assign data to locals to read the data only once
locals {
vpc_id = data.terraform_remote_state.infra.outputs.infra.vpc.vpc_id
subnets = data.terraform_remote_state.infra.outputs.infra.vpc.subnets
}
# Use locals in the modules to get to the infrastructure data
module "pipeline" {
source = "../../modules/pipeline"
vpc_id = local.vpc_id
...
}

Refactoring existing terraform code
⌾ Create a new bucket for the new terraform state to be stored at.
⌾ Rewrite your new code into the 3-tiers modules (as illustrated above and
detailed in the slides).
⌾ Import each of the resources into your live-envs terraform code.
⌾ terraform will show you the execution plan for the import operation:
○ values that exist in the deployed version but not in your code
will be marked with a (-) minus sign for removal.
○ values that do not exist in the deployed version but do exists in your code
will be marked with a (+) plus sign for addition.
The goal is to get a no change plan.

Modules Git Repo
Create a separate git repository for each of the tiers, and an additional to hold the
shared Go code for testing the modules:
⌾ terraform-resources
⌾ terraform-services
⌾ terraform-live-envs
⌾ terratest-common
Codeversioning

Managing Module Versions
For the development process, it is recommended to use a relative path when
working with the source attribute of a module
source = "../gcp/sql_instance"
You should change the source attribute value to a git repo when the module is ready
for release
When a module is released, it should be tagged and added to the source attribute
value using the ref argument
source = "git@github.com:unicorn/terraform-resources//gcp/sql_instance?ref=...v1.0.0"
Codeversioning

Modules in Subdirectories
Since we are using modules in a repo, the module itself is in a subdirectory relative
to the root of the repo.
A special double-forward-slash syntax is interpreted by Terraform to indicate that
the remaining path after that point is a subdirectory.
source = "git@github.com:unicorn/terraform-resources//gcp/sql_instance?ref=v1.0.0"
The ref argument can be either a tag or a branch name
Codeversioning

Modules Tagging Convention
Here is a recommended tagging convention for a module in the same repo:
<module-name>-v<semantic_versioning>
The module name should follow the directory structure you have in place.
Example:
Feel free to come up with your own tagging convention.
Codeversioning
gcp-sql-instance-v1.0.0

Terraform Modules Best Practices

Lock Down Terraform Version
Lock down the Terraform version that was used to create the module.
Place the following content in a file called versions.tf in the module:
terraform {
required_version = ">= 0.12"
}
TerraformModulesBestPractices

Using Provider in Module
provider "aws" {
region = var.region
version = "~> 2.24"
}
module "this_module" {
source "../"
name = "unicorn"
}
Terraform provider is inherited in modules.
This means that a provider will be inherited by the
modules your main module is calling.
Use an inline provider block inside your examples folder.
Only use the examples folder to test/develop your
module.

Prefer Hard Dependencies Over depends_on
⌾ depends_on doesn’t work with modules (currently on 0.12.6)
⌾ depends_on doesn’t work with data sources
⌾ There are some cases where depends_on would fail if the resource is it
depends_on is conditionally created
⌾ It’s better to be consistent across all the code that needs dependencies

Instead of using depends_on (i.e., resources), create a hard dependency in
Terraform between resources:

Or between modules:

No Hard-coded Values
Each module should have the following files:
⌾required-variables.tf
⌾default-variables.tf
All of the resource attributes should be variables. If an existing module is
hard-coded, you should move it into the default-variables.tf file.
You don’t have to use all attributes as documented in Terraform docs,
you can add them as you go.

No tfvar Files
tfvar files are key=value lines of variables passed into a module.
The main problem with using this feature is that you can’t tell which variable
belongs to which module.
This makes code usability hard to maintain and understand quickly.

Plugins Cache
Instead of having to download the same provider plugin to each module over and
over again, you should set your plugin cache folder via an environment variable
like so:
export TF_PLUGIN_CACHE_DIR="$HOME/.terraform.d/plugin-cache"
This will ensure that the provider plugin is linked to this folder and speed up
running Terraform init on new modules.

Terraform State Management
⌾ Create a storage bucket (S3/GCS) per environment
Do not use the same bucket for multiple envs
⌾ Enable versioning on the bucket - this will serve as a backup if state is
corrupted or can be used to compare concurrent executions
⌾ Use prefix with the same folder structure you set in terraform-live-envs folder
⌾ Use a separate prefix for infrastructure
i.e., vpc-network should be put into infrastructure/us-west2/blog-network

Terraform Null Attribute
Use a null value for an attribute you want to
remove from the resource.
Example; aws_s3_bucket can either be a
standalone bucket or a website.
If you need to make a single resource for
that, you would then make a default variable
with the value null, which effectively
removes the attribute from the resource
before creating that resource.

Terraform and Lists
Terraform will create a membership
resource per user, but behind the
scene, the count is also saved as an
index in the state file.
If you remove someone from the
middle of the list, the rest of the index
will shift up, causing the rest to be
added/recreated
In github this means delete user with
its forks!

Cascade Variables and Outputs
Always cascade (copy over) the default and required variables along with the
outputs to the next module tier, so the variable applied goes through all the
modules.

Terraform Testing Using Terratest
Terratest is a Go library by terragrunt.io
It automates the creation of the IaC (Infrastructure as code), and then tests that
the actual result is what you are expecting to get.
Once the tests are completed, Terratest will tear down and cleanup the resources
it has created using Terraform destroy command.
Tips:
⌾ You can use Terreatest with Docker, Packer and even helm charts!
⌾ Use vscode with its Go extension for a quick coding with Go
⌾ Learn Go interactively https://tour.golang.org

Typical Test Structure
// TestVPCCreatedWithDefaults - test VPC is created without overriding any of the default variables
func TestVPCCreatedWithDefaults(t *testing.T) {
terraformOptions := &terraform.Options{
// The path to where our Terraform code is located
TerraformDir: "../step1-infrastructure",
// Variables to pass to our Terraform code using -var options
Vars: map[string]interface{}{
"region": "us-east-1",
},
}
}
TerraformTestingwithTerratest
terraformOptions is a Golang struct defining the location of the code, as well as
terraform variables for the execution.

...
}
// At the end of the test, run `terraform destroy` to clean up any resources that were created
defer terraform.Destroy(t, terraformOptions)
...
}
defer will run at the end of the test and call terraform destroy to clean up the
resources created by this test.

...
}
// At the end of the test, run `terraform destroy` to clean up any resources that were created
defer terraform.Destroy(t, terraformOptions)
// Run `terraform init` and `terraform apply` and fail the test if there are any errors
terraform.InitAndApply(t, terraformOptions)
...
}
Run terraform init followed by terraform apply

...
}
…
vpc_id := terraform.Output(t, terraformOptions, "vpc_id"),
validateVPC(t, vpcID)
}
func ValidateVPC(t *testing.T, vpcID string) {...}
Validate it works as expected

Terratestbuilt-infunctions
Terratest has many built-in functions to check your infrastructure - but it’s relatively easy to
extend and write your own.
Terratest
Terratest Built-in Functions

Terratest
Validate Function Example

Run Go test.
You now have a unit test you can run after every commit!
Note:
Go tests timeout after 10 minutes, so make sure you set a
greater timeout to allow infrastructure to be created
Running the Test
Terratest

Reproducible Test Results:
Running tests does not take remote state configuration, thus the local module will
end up having a local state file, which can lead to a stale test results.
Don’t put the hard-coded path to the module in the test like so:
// The path to where our Terraform code is located
TerraformDir: "../step1-infrastructure",
…
Use a built-in function in Terratest that will copy the module to a temporary folder
and return the path to that folder:
Terratest Testing Techniques
Terratest

import (
test_structure "github.com/gruntwork-io/terratest/modules/test-structure"
)
func TestRoute53CNAMERecord(t *testing.T) {
rootFolder := "../../.."
modulePathRelativeToRootFolder := "terraform-aws-route53-resource/record-set"
terraformTempDir := test_structure.CopyTerraformFolderToTemp(
t,
rootFolder,
modulePathRelativeToRootFolder,
)
TerraformDir: tempTestFolder,
…
}
Terratest Testing Techniques
Terratest

In Go, a test will only report PASS or FAILED on the function name:
Multiple Tests Within a Test (subtests)
Terratest
Often you will need a few tests to check for the same function and reflect that in the
test outputs. This concept is called subtests.

Use the following to run multiple validations for the same test:
Multiple Tests Within a Test (subtests)
Terratest

Just about every method foo in Terratest comes in two versions: foo and fooE
(e.g., terraform.Apply and terraform.ApplyE).
⌾ foo: The base method takes a t *testing.T as an argument. If the method
hits any errors, it calls t.Fatal to fail the test.
⌾ fooE: Methods that end with the capital letter E always return an error as the
last argument and never call t.Fatal themselves. This allows you to decide
how to handle errors.
Terratest Error Handling
Terratest

You will use the base method name most of the time, as it allows you to keep your
code more concise by avoiding if err != nil checks all over the place:
terraform.Init(t, terraformOptions)
terraform.Apply(t, terraformOptions)
url := terraform.Output(t, terraformOptions, "url")
In the code above, if Init, Apply, or Output hits an error, the method will call t.Fatal
and fail the test immediately, which is typically the behavior you want. However, if
you are expecting an error and don't want it to cause a test failure, use the method
name that ends with a capital E:
if _, err := terraform.InitE(t, terraformOptions); err != nil {
// Do something with err
}
Terratest Error Handling
Terratest

The Test Pyramid
As you go up the pyramid, tests get more expensive, brittle and slower

The Test Pyramid
How the test pyramid works with infrastructure code:

The Test Pyramid
Lots of unit tests:
test individual sub-modules (keep them small!)
and static analysis (TFLint, Terraform validate)
Individual modules {

The Test Pyramid
Fewer integration tests:
test multiple sub-modules together
Individual modules
Multiple modules {

The Test Pyramid
A handful of high value E2E tests:
test entire environments (stage, prod)
Individual modules
multiple modules
Entire stack {

The Test Pyramid
Note the test times!
This is another reason to use small modules
60-240 minutes
(terraform-live-envs modules)
5-60 minutes
(terraform-services modules)
1-20 minutes
(terraform-resources modules)
1-60 seconds
(tflint/terraform validate)

To minimize the downsides of testing infrastructure as a code on a real platform,
follow the guidelines below:
1. Unit tests, integration tests, end-to-end tests
2. Testing environment
3. Namespacing
4. Cleanup
5. Timeouts and logging
6. Debugging interleaved test output
7. Avoid test caching
8. Error handling
9. Iterating locally using Docker
10. Iterating locally using test stages
Terratest Best Practices
Terratest

● Use VSCode for Terraform and Go integration
(you can opt to use JetBrains intelliJ too too or Vim)
● VSCode extensions:
Once installed you need to open the command palette (View -> Command
Palette) and type: “Terraform: Enable Language Server” which will prompt you
for installing the latest package for the terraform 0.12 support.
Pro Tips
Terratest

Terraform Modules Restructured

More Related Content

What's hot

Similar to Terraform Modules Restructured

More from DoiT International

Recently uploaded

Terraform Modules Restructured