Reusable, composable, battle-tested
TERRAFORM
MODULES
gruntwork.io
Your project is code complete
and it’s time to deploy it!
I know, I’ll use AWS!
You login to the AWS console
(Spend hours reading docs)
OK, I have a server running!
What else do I need?
Well, you probably want more than
one server for high availability
And a load balancer to distribute
traffic across those servers
And EBS Volumes and RDS databases
to store all of your data
You’ll need an S3 bucket for files
CloudWatch for monitoring
Don’t forget the VPC, subnets, route
tables, NAT gateways
Route 53 for DNS
ACM for SSL/TLS certs and KMS to
encrypt / decrypt secrets
stage prod
And you need all of that in separate
environments for stage and prod
stage prod
Plus DevOps tooling to manage it all
stage prod
And a CI server to test it all
stage prod
Plus alerts and on-call rotation to
notify you when it all breaks
stage prod
And also…
And you have to maintain it all.
Forever.
AWS: 1,000 new releases in 2016
Terraform: release every ~2 weeks
Security vulnerabilities: daily
There’s a better way to deploy
and manage infrastructure:
TERRAFORM
MODULES
TERRAFORM
MODULES
Reusable, composable, battle-tested
infrastructure code
In this talk, I’ll show you how
Terraform Modules work
stage prod
And how they will allow you to do all
of this…
> terraform init <…>
> terraform apply
In just a few simple commands
I’m
Yevgeniy
Brikman
ybrikman.com
Author
Co-founder of
Gruntwork
gruntwork.io
Your entire AWS
infrastructure.
Defined as code.
In about a day.
gruntwork.io
Outline
1. What’s a Module
2. How to use a Module
3. How Modules work
4. The future of Modules
Outline
1. What’s a Module
2. How to use a Module
3. How Modules work
4. The future of Modules
The two primary types of
infrastructure providers:
CPU Memory Disk Drive Network Server DB
Infrastructure as a Service (IaaS): e.g.,
AWS, Azure, Google Cloud
CPU Memory Disk Drive Network Server DB
They offer many primitives and it’s up
to you to put them together
Rails MySQL GitHub
CPU Memory Disk Drive Network Server DB
Platform as a Service (PaaS): e.g.,
Heroku, Docker Cloud, Engine Yard
Rails MySQL GitHub
CPU Memory Disk Drive Network Server DB
They hide all the lower-level details so
you can focus on your apps
> cd my-nodejs-app
> heroku create my-nodejs-app
> git push heroku master
Getting started with a PaaS is
easy!
Heroku limitations
1. Can only use supported runtimes & versions (e.g., python-3.6.2 or python-2.7.13)
2. Can only use supported system software & libraries
3. Can only run web services (data stores and other services available only via paid add-ons)
4. Apps can’t access the shell
5. Devs can’t access servers via SSH
6. Local disk is read-only
7. Load balancing is HTTP/HTTPS only
8. Requests are limited to 30 seconds
9. Limited to one AWS region
10. App must boot in 60 seconds or less
11. Apps can be at most 100MB
12. Build must take less than 15 min
13. Logs are limited to 1500 lines unless you use supported (paid) add-ons
14. Manual scaling only
15. Pricing gets very steep as you scale up
However, customizing, debugging,
16.
17.
Support only available on PST time zone
Limited control over security settings
and scaling is not.
CPU Memory Disk Drive Network Server DB
For most software companies, IaaS is
the only way to grow
stage prod
But that requires dealing with this…
We are developers.
We know how to fix this.
Use code!
Terraform is a tool for defining and
managing infrastructure as code
provider "aws" {
region = "us-east-1"
}
resource "aws_instance" "example" {
ami = "ami-408c7f28"
instance_type = "t2.micro"
}
Example: Terraform code to
deploy a server in AWS
> terraform apply
aws_instance.example: Creating...
ami: "" => "ami-408c7f28"
instance_type: "" => "t2.micro"
key_name: "" => "<computed>"
private_ip: "" => "<computed>"
public_ip: "" => "<computed>”
aws_instance.example: Creation complete
Apply complete! Resources: 1 added, 0 changed, 0 destroyed.
Run terraform apply to deploy
the server
Terraform supports modules
They are like reusable blueprints
for your infrastructure
resource "aws_autoscaling_group" "example" {
name = "${var.name}-service"
min_size = "${var.num_instances}"
max_size = "${var.num_instances}"
}
resource "aws_launch_configuration" "example" {
image_id = "${var.image_id}"
instance_type = "${var.instance_type}"
root_block_device {
volume_type = "gp2"
volume_size = 200
Example: create a module to
}
}
deploy a microservice
module "service_foo" {
source = "/modules/microservice"
image_id = "ami-123asd1"
num_instances = 3
}
Now you can use the module to
deploy one microservice
module "service_foo" {
source = "/modules/microservice"
image_id = "ami-123asd1"
num_instances = 3
}
module "service_bar" {
source = "/modules/microservice"
image_id = "ami-f2bb05ln"
num_instances = 6
}
module "service_baz" {
source = "/modules/microservice"
image_id = "ami-ny6v24xa"
Or multiple microservices
num_instances = 3
}
CPU Memory Disk Drive Network Server DB
Modules allow you to use your
favorite IaaS provider…
Rails MySQL GitHub
CPU Memory Disk Drive Network Server DB
With easy-to-use, high-level
abstractions, like a PaaS
Rails MySQL GitHub
CPU Memory Disk Drive Network Server DB
But since you have all the code, you
still have full control!
Best of all: code can be shared
and re-used!
The Terraform Module Registry
A collection of reusable, verified,
supported Modules
Example: Vault Module for AWS
Example: Consul for Azure
Example: Nomad for GCP
Outline
1. What’s a Module
2. How to use a Module
3. How Modules work
4. The future of Modules
Imagine you wanted to deploy Vault
The old way:
1. Open up the Vault docs
2. Deploy a few servers
3. Install Vault
4. Install supervisord
5. Configure mlock
6. Generate self-signed TLS cert
7. Create Vault config file
8. Create an S3 bucket as storage backend
9. Figure out IAM policies for the S3 bucket
10. Tinker with security group rules
11. Figure out IP addresses to bind to and advertise
12. Fight with Vault for hours because it won’t accept your TLS cert
13. Regenerate cert with RSA encryption
14. Update OS certificate store to accept self-signed certs
15. Realize you need to deploy a Consul cluster for high availability
16. Open up the Consul docs…
The new way:
> terraform init hashicorp/vault/aws
> terraform apply
And now you have this deployed
As simple as a PaaS!
> tree
.
├── README.md
├── main.tf
├── outputs.tf
├── packer
├── user-data
└── variables.tf
But you also have all the code.
Feel free to edit it!
module "vault_cluster" {
source = "hashicorp/vault/aws"
cluster_name = "example-vault-cluster"
cluster_size = 3
vpc_id = "${data.aws_vpc.default.id}"
subnet_ids = "${data.aws_subnets.default.ids}"
}
module "consul_cluster" {
source = "hashicorp/consul/aws"
cluster_name = "example-consul-cluster"
cluster_size = 3
vpc_id = "${data.aws_vpc.default.id}"
subnet_ids = "${data.aws_subnets.default.ids}"
Example: modify main.tf
}
> terraform apply
Run apply when you’re done!
Outline
1. What’s a Module
2. How to use a Module
3. How Modules work
4. The future of Modules
def add(x, y):
return x + y
Most programming languages
support functions
def add(x, y):
return x + y
The function has a name
def add(x, y):
return x + y
It can take in inputs
def add(x, y):
return x + y
And it can return outputs
def add(x, y):
return x + y
add(3, 5)
add(10, 35)
add(-45, 6)
Key idea: code reuse
def add(x, y):
return x + y
assert add(3, 5) == 8
assert add(10, 35) == 45
assert add(-45, 6) == -39
Key idea: testing
def add(x, y):
return x + y
def sub(x, y):
return x - y
sub(add(5, 3), add(4, 7))
Key idea: composition
def run_classifier(data_set):
X_pca = PCA(n_components=2).fit_transform(X_train)
clusters = clf.fit_predict(X_train)
fig, ax = plt.subplots(1, 2, figsize=(8, 4))
fig.subplots_adjust(top=0.85)
predicted = svc_model.predict(X_test)
images_and_predictions = list(zip(images_test, predicted))
ax[0].scatter(X_pca[:, 0], X_pca[:, 1], c=clusters)
ax[0].set_title('Predicted Training Labels')
ax[1].scatter(X_pca[:, 0], X_pca[:, 1], c=y_train)
ax[1].set_title('Actual Training Labels')
Key idea: abstraction
def run_classifier(data_set):
X_pca = PCA(n_components=2).fit_transform(X_train)
clusters = clf.fit_predict(X_train)
fig, ax = plt.subplots(1, 2, figsize=(8, 4))
fig.subplots_adjust(top=0.85)
predicted = svc_model.predict(X_test)
images_and_predictions = list(zip(images_test, predicted))
ax[0].scatter(X_pca[:, 0], X_pca[:, 1], c=clusters)
ax[0].set_title('Predicted Training Labels')
ax[1].scatter(X_pca[:, 0], X_pca[:, 1], c=y_train)
You want to hide a large
ax[1].set_title('Actual Training Labels')
“volume”…
def run_classifier(data_set):
X_pca = PCA(n_components=2).fit_transform(X_train)
clusters = clf.fit_predict(X_train)
fig, ax = plt.subplots(1, 2, figsize=(8, 4))
fig.subplots_adjust(top=0.85)
predicted = svc_model.predict(X_test)
images_and_predictions = list(zip(images_test, predicted))
ax[0].scatter(X_pca[:, 0], X_pca[:, 1], c=clusters)
ax[0].set_title('Predicted Training Labels')
ax[1].scatter(X_pca[:, 0], X_pca[:, 1], c=y_train)
Behind a small “surface area”
ax[1].set_title('Actual Training Labels')
Modules are Terraform’s
equivalent of functions
A simple module:
> tree minimal-module
.
├── main.tf
├── outputs.tf
├── variables.tf
└── README.md
It’s just Terraform code in a folder!
variable "name" {
description = "The name of the EC2 instance"
}
variable "image_id" {
description = "The ID of the AMI to run"
}
variable "port" {
description = "The port to listen on for HTTP requests"
}
The inputs are in variables.tf
output "url" {
value = "http://${aws_instance.example.ip}:${var.port}"
}
The outputs are in outputs.tf
resource "aws_autoscaling_group" "example" {
name = "${var.name}-service"
min_size = "${var.num_instances}"
max_size = "${var.num_instances}"
}
resource "aws_launch_configuration" "example" {
image_id = "${var.image_id}"
instance_type = "${var.instance_type}"
root_block_device {
volume_type = "gp2"
volume_size = 200
}
The resources are in main.tf
}
# Foo Module for AWS
This is a Terraform Module to deploy
[Foo](http://www.example.com) on AWS, including:
* foo
* bar
* baz
Documentation is in README.md
A more complicated module:
> tree complete-module
.
├── main.tf
├── outputs.tf
├── variables.tf
├── README.MD
├── modules
├── examples
└── test
We add three new folders:
modules, examples, test
> tree complete-module/modules
modules/
├── submodule-bar
│ ├── main.tf
│ ├── outputs.tf
│ └── variables.tf
└── submodule-foo
├── main.tf
├── outputs.tf
└── variables.tf
The modules folder contains
standalone “submodules”
> tree complete-module/modules
modules/
├── submodule-bar
│ ├── install-vault.sh
│ └── run-vault.sh
└── submodule-foo
└── main.go
Some of the submodules may not
even be Terraform code
For example, one submodule can be
used to install Vault in an AMI
Another to create self-signed TLS
certificates
Another to deploy the AMI across an
Auto Scaling Group (ASG)
Another to create an S3 bucket and
IAM policies as a storage backend
Another to configure the Security
Group settings
And one more to deploy a load
balancer (ELB)
You can use all the submodules
Or pick the ones you want and swap
in your own for the rest
> tree complete-module/examples
examples/
├── example-foo
│ ├── main.tf
│ ├── outputs.tf
│ └── variables.tf
└── example-bar
├── main.tf
├── outputs.tf
└── variables.tf
The examples folder shows how to
use the submodules
> tree complete-module
.
├── main.tf
├── outputs.tf
├── variables.tf
├── README.MD
├── modules
├── examples
└── test
Note: the code in the root is
usually a “canonical” example
It’s typically an opinionated way to
use all the submodules together
> tree complete-module/examples
examples/
├── example-foo
│ ├── main.tf
│ ├── outputs.tf
│ └── variables.tf
└── example-bar
├── main.tf
├── outputs.tf
└── variables.tf
The code in examples shows other
possible permutations
E.g., How to use just one or two of the
submodules together
Or how to combine with other
modules (e.g., Vault + Consul)
This is like function composition!
> tree complete-module/test
test/
├── example_foo_test.go
└── example_bar_test.go
The test folder contains
automated tests
> tree complete-module/test
test/
├── example_foo_test.go
└── example_bar_test.go
The tests are typically “integration
tests”
func vaultTest(t *testing.T, options *terratest.Options) {
tlsCert := generateSelfSignedTlsCert(t)
defer cleanupSelfSignedTlsCert(t, tlsCert)
amiId := buildVaultAmi(t)
defer cleanupAmi(t, amiId)
terratest.Apply(options)
defer terratest.Destroy(options)
assertCanInitializeAndUnsealVault(t, options)
}
Example test case for Vault
func vaultTest(t *testing.T, options *terratest.Options) {
tlsCert := generateSelfSignedTlsCert(t)
defer cleanupSelfSignedTlsCert(t, tlsCert)
amiId := buildVaultAmi(t)
defer cleanupAmi(t, amiId)
terratest.Apply(options)
defer terratest.Destroy(options)
assertCanInitializeAndUnsealVault(t, options)
}
Create test-time resources
func vaultTest(t *testing.T, options *terratest.Options) {
tlsCert := generateSelfSignedTlsCert(t)
defer cleanupSelfSignedTlsCert(t, tlsCert)
amiId := buildVaultAmi(t)
defer cleanupAmi(t, amiId)
terratest.Apply(options)
defer terratest.Destroy(options)
assertCanInitializeAndUnsealVault(t, options)
}
Run terraform apply
func vaultTest(t *testing.T, options *terratest.Options) {
tlsCert := generateSelfSignedTlsCert(t)
defer cleanupSelfSignedTlsCert(t, tlsCert)
amiId := buildVaultAmi(t)
defer cleanupAmi(t, amiId)
terratest.Apply(options)
defer terratest.Destroy(options)
assertCanInitializeAndUnsealVault(t, options)
}
Run terraform destroy at the end
func vaultTest(t *testing.T, options *terratest.Options) {
tlsCert := generateSelfSignedTlsCert(t)
defer cleanupSelfSignedTlsCert(t, tlsCert)
amiId := buildVaultAmi(t)
defer cleanupAmi(t, amiId)
terratest.Apply(options)
defer terratest.Destroy(options)
assertCanInitializeAndUnsealVault(t, options)
}
Check the Vault cluster works!
Using a module:
module "service_foo" {
source = "./minimal-module"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
For simple Modules and learning,
deploy the root
module "submodule_foo" {
source = "./complete-module/modules/submodule-foo"
param_foo = "foo"
param_bar = 8080
}
module "submodule_bar" {
source = "./complete-module/modules/submodule-bar"
param_foo = "abcdef"
param_bar = 9091
}
For more complicated use-cases,
use the submodules
module "service_foo" {
source = "./minimal-module"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
Abstraction: simple Module API
for complicated infrastructure
module "service_foo" {
source = "./minimal-module"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
module "service_bar" {
source = "./minimal-module"
name = "Bar"
image_id = "ami-abcd1234"
port = 9091
}
Re-use: create a Module once,
deploy it many times
Versioning:
module "service_foo" {
source = "./foo"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
You can set source to point to
modules at local file paths
module "service_foo" {
source = "hashicorp/vault/aws"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
Alternatively, you can use
Terraform registry URLs
module "service_foo" {
source = "git::git@github.com:foo/bar.git"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
Or arbitrary Git URLs
module "service_foo" {
source = "git::git@github.com:foo/bar.git?ref=v1.0.0"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
You can even link to a specific Git
tag (recommended!)
module "service_foo" {
source = "git::git@github.com:foo/bar.git?ref=v1.0.0"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
Modules use semantic versioning
module "service_foo" {
source = "git::git@github.com:foo/bar.git?ref=v2.0.0"
name = "Foo"
image_id = "ami-123asd1"
port = 8080
}
So upgrading infrastructure is just
a version number bump
qa
stage
prod
Promote immutable, versioned
infrastructure across environments
Outline
1. What’s a Module
2. How to use a Module
3. How Modules work
4. The future of Modules
“You are not special.
Your infrastructure is
not a beautiful and
unique snowflake. You
have the same tech debt
as everyone else.”
— your sysadmin,
probably
stage prod
You need this
stage prod
And so does everyone else
Stop reinventing the wheel
Start building on top of
battle-tested code
Start building on top of
commercially-supported code
Start building on top of
code
Advantages of code
1. Reuse
2. Compose
3. Configure
4. Customize
5. Debug
6. Test
7. Version
8. Document
At Gruntwork, we’ve
been building
Modules for years
gruntwork.io
Many companies, all
running on the same
infrastructure code
gruntwork.io
stage prod
Modules allow us to turn this…
> terraform init <…>
> terraform apply
… into this
With some help from this
Questions?
modules@gruntwork.io
