Terraformpy is a library and command line tool to supercharge your Terraform configs using a full fledged Python environment!
Terraform is an amazing tool. Like, really amazing. When working with code that is managing third-party service definitions, and actually applying changes to those definitions by invoking APIs, a high-degree of confidence in the change process is a must-have, and that's where Terraform excels. The work flow it empowers allow teams to quickly make changes across a large (and ever growing) footprint in multiple providers/regions/technologies/etc.
But as your definitions grow the HCL syntax very quickly leaves a lot to be desired, and is it ever verbose... So many definitions of variables and outputs need to be repeated, over and over, as you compose more modules that use each other.
Since HCL is "fully JSON compatible" and Python is great at generating JSON data, we built Terraformpy to provide a more productive environment to build and maintain complex Terraform configs. It has been used daily in production at NerdWallet since 2016 and has proven very valuable in accelerating our adoption of Terraform across our engineering organization.
The recommended way to install and use Terraformpy is via Pipenv
An example would look like:
$ mkdir my-terraform-project
$ cd my-terraform-project
$ pipenv install terraformpy
You can then run Terraformpy using pipenv run
:
$ pipenv run terraformpy ...
Or you can use pipenv shell
to activate the virtualenv so you don't need to use pipenv run
. The rest of this document assumes that you've run pipenv shell
and can just run terraformpy
directly.
The terraformpy
command line tool operates as a shim for the underlying terraform
tool. When invoked it will first find all *.tf.py
files in the current directory, loading them using the imp module, generate a file named main.tf.json
, and then invoke underlying tool.
# just replace terraform in your regular workflow
terraformpy plan -out=tf.plan
# review changes...
# apply them!
# since we're going to operate on the generated plan here, we don't event need to use terraformpy anymore
terraform apply tf.plan
Each of the *.tf.py
files uses a declarative syntax, using objects imported from this library. You don't need to define a main function, you just create instances of classes (anonymous or otherwise) in the root of the module (you're building regular Python code here). Since you're in a full blown Python environment there is no limit on what you can do -- import things, connect to databases, etc.
The terraformpy
name space provides a number of classes that map directly to things you declare in normal .tf.
files. To write your definitions simply import these classes and begin creating instances of them. Below is the first example from the Terraform getting start guide.
from terraformpy import Provider, Resource
Provider(
'aws',
profile='default',
region='us-east-1'
)
Resource(
'aws_instance', 'example',
ami='ami-2757f631'
instance_type='t2.micro'
)
Things you can import from terraformpy
:
Provider
Variable
Data
Resource
Output
See the examples/
dir for fully functional examples.
So far, we've only used terraformpy anonymously, but the returned instances of the Data
and Resource
classes offer handy interpolation attributes. For example, a common task is using the Data
class to fetch remote data:
ami = Data(
'aws_ami', 'ecs_ami',
most_recent=True,
filter=[
dict(name='name', values=['\*amazon-ecs-optimized']),
dict(name='owner-alias', values=['amazon'])
]
)
Resource(
'aws_instance', 'example',
ami=ami.id,
instance_type='m4.xlarge'
)
Here we simply refer to the id attribute on the ami object when creating the aws_instance
. During the compile phase it would be converted to the correct syntax: "${data.aws_ami.ecs_ami.id}"
.
This works by having a custom __getattr__
function on our Data
and Resource
objects that will turn any attribute access for an attribute name that doesn't exist into the Terraform interpolation syntax.
Modules have been explicitly excluded from this implementation because they aim to solve the same problem -- building reusable blocks in your Terraform configs.
With all the features of Python at your disposal building reusable units is straightforward without using the native modules from Terraform, but do see Resource Collections (next) for some helper scaffolding!
A common pattern when building configs using Python is to want to abstract a number of different resources under the guise of a single object -- which is the same pattern native Terraform modules aim to solve. In terraformpy we provide a ResourceCollection
base class for building objects that represent multiple resources.
You can use Schematics to define the fields and perform validation.
As an example, when provisioning an RDS cluster you may want to have a standard set of options that you ship with all your clusters. You can express that with a resource collection:
from schematics import types
from schematics.types import compound
from terraformpy import Resource, ResourceCollection
class RDSCluster(ResourceCollection):
# Defining attributes of your resource collection is like defining a Schematics Model, in fact the
# ResourceCollection class is just a specialized subclass of the Schematics Model class.
#
# Each attribute becomes a field on the collection, and can be provided as a keyword when constructing
# an instance of your collection.
#
# Validation works the same as in Schematics. You can attach validators to the fields themselves and
# also define "validate_field" functions.
name = types.StringType(required=True)
azs = compound.ListType(types.StringType, required=True)
instance_class = types.StringType(required=True, choices=('db.r3.large', ...))
# The create_resources function is invoked once the instance has been created and the kwargs provided have been
# processed against the inputs. All of the instance attributes have been converted to the values provided, so
# if you access self.name in create_resources you're accessing whatever value was provided to the instance
def create_resources(self):
self.param_group = Resource(
'aws_rds_cluster_parameter_group', '{0}_pg'.format(self.name),
family='aurora5.6',
parameter=[
{'name': 'character_set_server', 'value': 'utf8'},
{'name': 'character_set_client', 'value': 'utf8'}
]
)
self.cluster = Resource(
'aws_rds_cluster', self.name,
cluster_identifier=self.name,
availability_zones=self.azs,
database_name=self.name,
master_username='root',
master_password='password',
db_cluster_parameter_group_name=self.param_group.id
)
self.instances = Resource(
'aws_rds_cluster_instance', '{0}_instances'.format(self.name),
count=2,
identifier='{0}-${{count.index}}'.format(self.name),
cluster_identifier=self.cluster.id,
instance_class=self.instance_class
)
That definition can then be imported and used in your terraformpy configs.
from modules.rds import RDSCluster
cluster1 = RDSCluster(
name='cluster1',
azs=['us-west-2a','us-west-2b','us-west-2c'],
instance_class='db.r3.large'
)
# you can then refer to the resources themselves, for interpolation, through the attrs
# i.e. cluster1.cluster.id
Resource definitions that exist across many different environments often only vary slightly between each environment. To facilitate the ease of definition for these differences you can use variant grouping.
First create the folders: configs/stage/
, configs/prod/
, configs/shared/
. Inside each of them place a __init__.py
to make them packages.
Next create the file configs/shared/instances.py
:
from terraformpy import Resource
Resource(
'aws_instance', 'example',
ami=ami.id,
prod_variant=dict(
instance_type='m4.xlarge'
),
stage_variant=dict(
instance_type='t2.medium'
)
)
Then create configs/stage/main.tf.py
:
from terraformpy import Variant
with Variant('stage'):
import configs.shared.instances
Since the import of the instances file happens inside of the Variant context then the Resource will be created as if it had been defined like:
from terraformpy import Resource
Resource(
'aws_instance', 'example',
ami=ami.id,
instance_type='t2.medium'
)
Depending on your usage of Terraform you will likely end up needing to use multiple providers at some point in time. To use multiple providers in Terraform you define them using aliases and then reference those aliases in your resource definitions.
To make this pattern easier you can use the Terraformpy Provider
object as a context manager, and then any resources created within the context will automatically have that provider aliases referenced:
from terraformpy import Resource, Provider
with Provider("aws", region="us-west-2", alias="west2"):
sg = Resource('aws_security_group', 'sg', ingress=['foo'])
assert sg.provider == 'aws.west2'
Often times you will want to include the contents of a file that is located alongside your Python code, but when running terraform
along with the ${file('myfile.json')}
interpolation function pathing will be relative to where the compiled main.tf.json
file is and not where the Python code lives.
To help with this situation a function named relative_file
inside of the terraformpy.helpers
namespace is provided.
from terraformpy import Resource
from terraformpy.helpers import relative_file
Resource(
'aws_iam_role', 'role_name',
name='role-name',
assume_role_policy=relative_file('role_policy.json')
)
This would produce a definition that leverages the ${file(...)}
interpolation function with a path that reads the role_policy.json
file from the same directory as the Python code that defined the role.
When creating aws_security_group_rule
Resource
objects you cannot pass self=True
to the object since Python already passes a self
argument into the constructor. In this case you'll need to specify it directly in the _values
:
sg = Resource(
'aws_security_group_rule', 'my_rule',
_values=dict(self=True),
vpc_id=vpc.id,
...
)
- Create an issue, check out a branch, and make your code changes.
- Push to run CircleCI tests.
- Create Pull Request to Master including VERSION bump.
- Merge PR after Approval.
- Add tag like v1.0.0 that matches new version and push.