Skip to main content

Terraformpy is a library and command line tool to supercharge your Terraform configs using a full fledged Python environment!

Project description

pypi versions format license coverage


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.

Installing Terraformpy

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.

Using the CLI tool

The terraformpy command line tool operates as a shim for the underlying terraform tool. When invoked it will first find all * files in the current directory, loading them using the imp module, generate a file named, 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 * 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.

Writing files

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


    'aws_instance', 'example',

Things you can import from terraformpy:

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',
        dict(name='name', values=['\*amazon-ecs-optimized']),
        dict(name='owner-alias', values=['amazon'])

    'aws_instance', 'example',,

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: "${}".

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.


Configuring a backend happens in the Terraform object. See Configuring a Terraform Backend for more details.

Bellow we are using an S3 Backend:



Since Terraformpy gives you the full power of Python we encourage you to use “Resource Collections” (see the next section) when you’re building your own modular functionality and you don’t plan on sharing these modules outside of your current organization.

You can however leverage existing HCL modules using the Module object if you want to use pre-built, existing modules:



Resource Collections

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 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(,
                {'name': 'character_set_server', 'value': 'utf8'},
                {'name': 'character_set_client', 'value': 'utf8'}

        self.cluster = Resource(

        self.instances = Resource(
            'aws_rds_cluster_instance', '{0}_instances'.format(,

That definition can then be imported and used in your terraformpy configs.

from modules.rds import RDSCluster

cluster1 = RDSCluster(

# you can then refer to the resources themselves, for interpolation, through the attrs
# i.e.


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 to make them packages.

Next create the file configs/shared/

from terraformpy import Resource

    'aws_instance', 'example',,

Then create configs/stage/

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

    'aws_instance', 'example',,

Multiple providers

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'

Using file contents

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 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

    'aws_iam_role', 'role_name',

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.


Terraformy offers a “hooks” system that allows you to modify objects on the fly at compile time. This can be useful to apply transformations so that users of objects do not need to worry about some of the idiosyncratic schema details of Terraform’s JSON syntax, most notably “Attributes as Blocks”.

The best example of this is the aws_security_group object type, which requires that its ingress and egress blocks have all of their attributes present, even if they are null. To have users have to type out all of the different attributes and set them to None is cumbersome, so instead you can use the hook that we ship with this distribution:

from import install_aws_security_group_attributes_as_blocks_hook


Now, users only need to specify the attributes they care about in rules and the hook will take care of filling in all of the optional attributes that are mandatory to appear in the final compiled JSON.

For more information on how the hooks work see the file and the inline comments for the add_hook function on the different object types.

Notes and Gotchas

Security Group Rules and self

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',

Developer notes

Running tests

We use tox to run tests. While developing locally you can run:


Formatting with black

We use black to format code. To apply formatting run:

tox -e black -- .

Release Steps

  1. Make a branch

  2. Make your changes

  3. Bump the version in the VERSION file and add an entry to the file

  4. Open a PR, tag @NerdWalletOSS/dynamorm in your PR description

  5. Once approved and merged to master the new version will be pushed to pypi

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

terraformpy-1.3.5.tar.gz (25.2 kB view hashes)

Uploaded source

Built Distribution

terraformpy-1.3.5-py2.py3-none-any.whl (22.6 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page