tfwrapper is a python wrapper for OpenTofu and legacy Terraform which aims to simplify their usage and enforce best practices.

Note: the term Terraform is used in this documentation when talking about generic concepts like providers, modules, stacks and the HCL based domain specific language.

• claranet-tfwrapper
  • Table Of Contents
  • Features
  • Setup Dependencies
  • Runtime Dependencies
  • Recommended setup
  • Installation
  • Setup command-line completion
  • Upgrade from tfwrapper v7 or older
    • Required files
  • Configuration
    • tfwrapper configuration
    • Stacks configurations
    • States centralization configuration
    • How to migrate from one backend to another for state centralization
  • Stacks file structure
  • Usage
    • Stack bootstrap
    • Working on stacks
    • Passing options
  • Environment
    • S3 state backend credentials
    • Azure Service Principal credentials
    • Azure authentication isolation
    • GCP configuration
    • GKE configurations
    • Stack configurations and credentials
    • Stack path
• Development
  • Tests
  • Debug command-line completion
  • Python code formatting
  • Checks
  • README TOC
  • Using OpenTofu development builds
  • git pre-commit hooks
  • Review and merge open Dependabot PRs
  • Tagging and publishing new releases to PyPI

• OpenTofu and Terraform behaviour overriding
• State centralization enforcement
• Standardized file structure
• Stack initialization from templates
• AWS credentials caching
• Azure credentials loading (both Service Principal or User)
• GCP and GKE user ADC support
• Plugins caching
• Tab completion

• python3 >= 3.8.1 <4.0
• python3-pip
• python3-venv
• pipx (recommended)

• terraform >= 0.10 (>= 0.15 for fully working Azure backend with isolation due to hashicorp/terraform#25416)
• azure-cli when using context based Azure authentication

• OpenTofu 1.6+ (recommended) or Terraform 1.0+ (warning: versions above 1.6 are not open-source, and may cause legal issues depending on the context you are using it).
• An AWS S3 bucket and DynamoDB table for state centralization in AWS.
• An Azure Blob Storage container for state centralization in Azure.

tfwrapper should installed using pipx (recommended) or pip:

pipx install claranet-tfwrapper

Add the following to your shell's interactive configuration file, e.g. .bashrc for bash:

eval "$(register-python-argcomplete tfwrapper -e tfwrapper)"

You can then press the completion key (usually Tab ↹) twice to get your partially typed tfwrapper commands completed.

Note: the -e tfwrapper parameter adds an suffix to the defined _python_argcomplete function to avoid clashes with other packages (see kislyuk/argcomplete#310 (comment) for context).

If you used versions of the wrapper older than v8, there is not much to do when upgrading to v8 except a little cleanup. Indeed, the wrapper is no longer installed as a git submodule of your project like it used to be instructed and there is no longer any Makefile to activate it.

Just clean up each project by destroying the .wrapper submodule:

git rm -f Makefile
git submodule deinit .wrapper
rm -rf .git/modules/.wrapper
git rm -f .wrapper

Then check the staged changes and commit them.

tfwrapper expects multiple files and directories at the root of a project.

Stacks configurations are stored in the conf directory.

The templates directory is used to store the state backend configuration template and the Terraform stack templates used to initialize new stacks. Using a git submodule is recommended.

The following files are required:

• templates/{provider}/common/state.tf.jinja2: AWS S3 or Azure Blob Storage state backend configuration template.
• templates/{provider}/basic/main.tf: the default Terraform configuration for new stacks. The whole template/{provider}/basic directory is copied on stack initialization.

For example with AWS:

mkdir -p templates/aws/common templates/aws/basic

# create state configuration template with AWS backend
cat << 'EOF' > templates/aws/common/state.tf.jinja2
{% if region is not none %}
{% set region = '/' + region + '/' %}
{% else %}
{% set region = '/' %}
{% endif %}

terraform {
  backend "s3" {
    bucket = "my-centralized-terraform-states-bucket"
    key    = "{{ client_name }}/{{ account }}/{{ environment }}{{ region }}{{ stack }}/terraform.state"
    region = "eu-west-1"

    dynamodb_table = "my-terraform-states-lock-table"
  }
}

resource "null_resource" "state-test" {}
EOF

# create a default stack templates with support for AWS assume role
cat << 'EOF' > templates/aws/basic/main.tf
provider "aws" {
  region     = var.aws_region
  access_key = var.aws_access_key
  secret_key = var.aws_secret_key
  token      = var.aws_token
}
EOF

For example with Azure:

mkdir -p templates/azure/common templates/azure/basic

# create state configuration template with Azure backend
cat << 'EOF' > templates/azure/common/state.tf.jinja2
{% if region is not none %}
{% set region = '/' + region + '/' %}
{% else %}
{% set region = '/' %}
{% endif %}

terraform {
  backend "azurerm" {
    subscription_id      = "00000000-0000-0000-0000-000000000000"
    resource_group_name  = "my-resource-group"
    storage_account_name = "my-centralized-terraform-states-account"
    container_name       = "terraform-states"

    key = "{{ client_name }}/{{ account }}/{{ environment }}{{ region }}{{ stack }}/terraform.state"
  }
}
EOF

# create a default stack templates with support for Azure credentials
cat << 'EOF' > templates/azure/basic/main.tf
provider "azurerm" {
  subscription_id = var.azure_subscription_id
  tenant_id       = var.azure_tenant_id
}
EOF

The .run directory is used for credentials caching and plan storage.

mkdir .run
cat << 'EOF' > .run/.gitignore
*
!.gitignore
EOF

Adding the following .gitignore at the root of your project is recommended:

cat << 'EOF' > .gitignore
.terraform
terraform.tfstate
terraform.tfstate.backup
terraform.tfvars
EOF

tfwrapper uses yaml files stored in the conf directory of the project.

tfwrapper uses some default behaviors that can be overridden or modified via a config.yml file in the conf directory.

---
always_trigger_init: False # Always trigger `terraform init` first when launching `plan` or `apply` commands
pipe_plan_command: "cat" # Default command used when you're invoking tfwrapper with `--pipe-plan`
use_local_azure_session_directory: False # Use the current user's Azure configuration in `~/.azure`. By default, the wrapper uses a local `azure-cli` session and configuration in the local `.run` directory.

Stacks configuration files use the following naming convention:

conf/${account}_${environment}_${region}_${stack}.yml

Here is an example for an AWS stack configuration:

---
state_configuration_name: "aws" # use "aws" backend state configuration
aws:
  general:
    account: &aws_account "xxxxxxxxxxx" # aws account for this stack
    region: &aws_region eu-west-1 # aws region for this stack
  credentials:
    profile: my-aws-profile # should be configured in .aws/config

terraform:
  legacy: false # Use legacy version of the tool instead of OpenTofu, defaults to false. Only used if version >= 1.6.0.
  version: "1.0" # OpenTofu version that tfwrapper will use for this stack. Automatically downloaded if it's not available locally. Defaults to 1.0
  vars: # variables passed to terraform
    aws_account: *aws_account
    aws_region: *aws_region
    client_name: my-client-name # arbitrary client name

Here is an example for a stack on Azure configuration using user mode and AWS S3 backend for state storage:

---
state_configuration_name: "aws-demo" # use "aws" backend state configuration
azure:
  general:
    mode: user # Uses personal credentials with MFA
    directory_id: &directory_id "00000000-0000-0000-0000-000000000000" # Azure Tenant/Directory UID
    subscription_id: &subscription_id "11111111-1111-1111-1111-111111111111" # Azure Subscription UID

terraform:
  legacy: false # Use legacy version of the tool instead of OpenTofu, defaults to false. Only used if version >= 1.6.0.
  version: "1.0" # OpenTofu version that tfwrapper will use for this stack. Automatically downloaded if it's not available locally. Defaults to 1.0
  vars:
    subscription_id: *subscription_id
    directory_id: *directory_id
    client_name: client-name #Replace it with the name of your client

It is using your account linked to a Microsoft Account. You must have access to the Azure Subscription if you want to use Terraform.

Here is an example for a stack on Azure configuration using Service Principal mode:

---
azure:
  general:
    mode: service_principal # Uses an Azure tenant Service Principal account
    directory_id: &directory_id "00000000-0000-0000-0000-000000000000" # Azure Tenant/Directory UID
    subscription_id: &subscription_id "11111111-1111-1111-1111-111111111111" # Azure Subscription UID

    credentials:
      profile: customer-profile # To stay coherent, create an AzureRM profile with the same name as the account-alias. Please checkout `azurerm_config.yml.sample` file for configuration structure.

terraform:
  legacy: false # Use legacy version of the tool instead of OpenTofu, defaults to false. Only used if version >= 1.6.0.
  version: "1.0" # OpenTofu version that tfwrapper will use for this stack. Automatically downloaded if it's not available locally. Defaults to 1.0
  vars:
    subscription_id: *subscription_id
    directory_id: *directory_id
    client_name: client-name # Replace it with the name of your client

The wrapper uses the Service Principal's credentials to connect the Azure subscription. The given Service Principal must have access to the subscription. The wrapper loads client_id, client_secret and tenant_id properties from your config.yml file located in ~/.azurerm/config.yml.

~/.azurerm/config.yml file structure example:

---
claranet-sandbox:
  client_id: aaaaaaaa-bbbb-cccc-dddd-zzzzzzzzzzzz
  client_secret: AAbbbCCCzzz==
  tenant_id: 00000000-0000-0000-0000-000000000000

customer-profile:
  client_id: aaaaaaaa-bbbb-cccc-dddd-zzzzzzzzzzzz
  client_secret: AAbbbCCCzzz==
  tenant_id: 00000000-0000-0000-0000-000000000000

Here is an example for a GCP/GKE stack with user ADC and multiple GKE instances:

---
gcp:
  general:
    mode: adc-user
    project: &gcp_project project-name
  gke:
    - name: kubernetes-1
      zone: europe-west1-c
    - name: kubernetes-2
      region: europe-west1

terraform:
  legacy: false # Use legacy version of the tool instead of OpenTofu, defaults to false. Only used if version >= 1.6.0.
  version: "1.0" # OpenTofu version that tfwrapper will use for this stack. Automatically downloaded if it's not available locally. Defaults to 1.0
  vars:
    gcp_region: europe-west1
    gcp_zone: europe-west1-c
    gcp_project: *gcp_project
    client_name: client-name

You can declare multiple providers configurations, context is set up accordingly.

⚠ This feature is only supported for Azure stacks for now and only works with Azure authentication isolation

---
azure:
  general:
    mode: service_principal # Uses an Azure tenant Service Principal account
    directory_id: &directory_id "00000000-0000-0000-0000-000000000000" # Azure Tenant/Directory UID
    subscription_id: &subscription_id "11111111-1111-1111-1111-111111111111" # Azure Subscription UID

    credentials:
      profile: customer-profile # To stay coherent, create an AzureRM profile with the same name as the account-alias. Please checkout `azurerm_config.yml.sample` file for configuration structure.

  alternative:
    mode: service_principal # Uses an Azure tenant Service Principal account
    directory_id: "00000000-0000-0000-0000-000000000000" # Azure Tenant/Directory UID
    subscription_id: "22222222-2222-2222-2222-222222222222" # Azure Subscription UID

    credentials:
      profile: claranet-sandbox # To stay coherent, create an AzureRM profile with the same name as the account-alias. Please checkout `azurerm_config.yml.sample` file for configuration structure.

terraform:
  version: "1.0" # OpenTofu version that tfwrapper will use for this stack. Automatically downloaded if it's not available locally. Defaults to 1.0
  legacy: false # Use legacy version of the tool instead of OpenTofu, defaults to false. Only used if version >= 1.6.0.
  vars:
    subscription_id: *subscription_id
    directory_id: *directory_id
    client_name: client-name # Replace it with the name of your client

This configuration is useful when having various service principals with a dedicated rights scope for each.

The wrapper will generate the following Terraform variables that can be used in your stack:

• <config_name>_azure_subscription_id with Azure subscription ID. From the example, variable is: alternative_subscription_id = "22222222-2222-2222-2222-222222222222"
• <config_name>_azure_tenant_id with Azure tenant ID. From the example, variable is: alternative_tenant_id = "00000000-0000-0000-0000-000000000000"
• <config_name>_azure_client_id with Service Principal client id. From the example, variable is: alternative_client_id = "aaaaaaaa-bbbb-cccc-dddd-zzzzzzzzzzzz"
• <config_name>_azure_client_secret with Service Principal client secret. From the example, variable is: alternative_client_secret = "AAbbbCCCzzz=="

Also, an isolation context is set to the local .run/aure_<config_name> directory for each configuration.

The conf/state.yml configuration file defines the configuration used to connect to state backends.

These backends can be of AWS S3 and/or AzureRM types.

The resources for these backends are not created by tfwrapper, and thus must exist beforehand:

• AWS: an S3 bucket (and optionally but highly recommended a DynamoDB table for locking). It is also recommended to enable versioning on the S3 bucket.
• Azure: a Blob storage account

You can use other backends (e.g. Google GCS or Hashicorp Consul) not specifically supported by the wrapper, if you manage authentication yourself and omit the conf/state.yml file or make it empty:

---

Example configuration with both AWS and Azure backends defined:

---
aws:
  - name: "aws-demo"
    general:
      account: "xxxxxxxxxxx"
      region: eu-west-1
    credentials:
      profile: my-state-aws-profile # should be configured in .aws/config
azure:
  # This backend use storage keys for authentication
  - name: "azure-backend"
    general:
      subscription_id: "xxxxxxx" # the Azure account to use for state storage
      resource_group_name: "tfstates-xxxxx-rg" # The Azure resource group with state storage
      storage_account_name: "tfstatesxxxxx"
  - name: "azure-alternative"
    general:
      subscription_id: "xxxxxxx" # the Azure account to use for state storage
      resource_group_name: "tfstates-xxxxx-rg" # The Azure resource group with state storage
      storage_account_name: "tfstatesxxxxx"
  # This backend use Azure AD authentication
  - name: "azure-ad-auth"
    general:
      subscription_id: "xxxxxxx" # the Azure account to use for state storage
      resource_group_name: "tfstates-xxxxx-rg" # The Azure resource group with state storage
      storage_account_name: "tfstatesxxxxx"
      azuread_auth: true

backend_parameters: # Parameters or options which can be used by `state.j2.tf` template file
  state_snaphot: "false" # Example of Azure storage backend option

Note: the first backend will be the default one for stacks not defining state_backend_type.

If for example you have both an AWS and Azure state backend configured in your conf/state.yml file, you can migrate your stack state from one backend to another.

Here is a quick howto:

1. Make sure your stack is clean:

$ cd account/path/env/your_stack
$ tfwrapper init
$ tfwrapper plan
# should return no changes

2. Change your backend in the stack configuration yaml file:

---
#state_configuration_name: 'aws-demo' # previous backend
state_configuration_name: "azure-alternative" # new backend to use

3. Back in your stack directory, you can perform the change:

$ cd account/path/env/your_stack
$ rm -v state.tf # removing old state backend configuration
$ tfwrapper bootstrap # regen a new state backend configuration based on the stack yaml config file
$ tfwrapper init # Terraform will detect the new backend and propose to migrate it
$ tfwrapper plan
# should return the same changes diff as before

Terraform stacks are organized based on their:

• account: an account alias which may refer to provider accounts or subscriptions, e.g. project-a-prod, customer-b-dev.
• environment: production, preproduction, dev, etc. With global as a special case eliminating the region part.
• region: eu-west-1, westeurope, etc.
• stack: defaults to default. web, admin, tools, etc.

The following file structure is then enforced:

# project root
└── account
│   └── environment
│       └── region
│           └── stack
└── account
    └── _global
        └── stack

A real-life example:

# project root
├── aws-account-1
│   ├── _global
│   │   └── default
│   │       └── main.tf
│   └── production
│       ├── eu-central-1
│       │   └── web
│       │       └── main.tf
│       └── eu-west-1
│           ├── default
│           │   └── main.t
│           └── tools
│               └── main.tf
└── aws-account-2
    └── backup
        └── eu-west-1
            └── backup
                └── main.tf

After creating a conf/${account}_${environment}_${region}_${stack}.yml stack configuration file you can bootstrap it.

# you can bootstrap using the templates/{provider}/basic stack
tfwrapper -a ${account} -e ${environment} -r ${region} -s ${stack} bootstrap

# or another stack template, for example: templates/aws/foobar
tfwrapper -a ${account} -e ${environment} -r ${region} -s ${stack} bootstrap aws/foobar

# or from an existent stack, for example: customer/env/region/stack
tfwrapper -a ${account} -e ${environment} -r ${region} -s ${stack} bootstrap mycustomer/dev/eu-west/run

In the special case of a global stack, the configuration file should instead be named as conf/${account}_global_${stack}.yml.

You can work on stacks from their directory or from the root of the project.

# working from the root of the project
tfwrapper -a ${account} -e ${environment} -r ${region} -s ${stack} plan

# working from the root of a stack
cd ${account}/${environment}/${region}/${stack}
tfwrapper plan

You can also work on several stacks sequentially with the foreach subcommand from any directory under the root of the project. By default, foreach selects all stacks under the current directory, so if called from the root of the project without any filter, it will select all stacks and execute the specified command in them, one after another:

# working from the root of the project
tfwrapper foreach -- tfwrapper init

Any combination of the -a, -e, -r and -s arguments can be used to select specific stacks, e.g. all stacks for an account across all environments but in a specific region:

# working from the root of the project
tfwrapper -a ${account} -r ${region} foreach -- tfwrapper plan

The same can be achieved with:

# working from an account directory
cd ${account}
tfwrapper -r ${region} foreach -- tfwrapper plan

Complex commands can be executed in a sub-shell with the -S/--shell argument, e.g.:

# working from an environment directory
cd ${account}/${environment}
tfwrapper foreach -S 'pwd && tfwrapper init >/dev/null 2>&1 && tfwrapper plan 2>/dev/null -- -no-color | grep "^Plan: "'

You can pass anything you want to terraform using --.

tfwrapper plan -- -target resource1 -target resource2

tfwrapper sets the following environment variables.

The default AWS credentials of the environment are set to point to the S3 state backend. Those credentials are acquired from the profile defined in conf/state.yml

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• AWS_SESSION_TOKEN

Those AzureRM credentials are loaded only if you are using the Service Principal mode. They are acquired from the profile defined in ~/.azurerm/config.yml

• ARM_CLIENT_ID
• ARM_CLIENT_SECRET
• ARM_TENANT_ID

AZURE_CONFIG_DIR environment variable is set to the local .run/azure directory if global configuration value use_local_azure_session_directory is set to true, which is the default, which is the default.

If you have multiple configurations in your stacks, you also have <CONFIG_NAME>_AZURE_CONFIG_DIR which is set to the local .run/azure_<config_name> directory.

Those GCP related variables are available from the environment when using the example configuration:

• TF_VAR_gcp_region
• TF_VAR_gcp_zone
• TF_VAR_gcp_project

Each GKE instance has its own kubeconfig, the path to each configuration is available from the environment:

• TF_VAR_gke_kubeconfig_${gke_cluster_name}

kubeconfig is automatically fetched by the wrapper (using gcloud) and stored inside the .run directory of your project. It is refreshed automatically at every run to ensure you point to correct Kubernetes endpoint. You can disable this behaviour by setting refresh_kubeconfig: never in your cluster settings.

---
gcp:
  general:
    mode: adc-user
    project: &gcp_project project-name
  gke:
    - name: kubernetes-1
      zone: europe-west1-c
      refresh_kubeconfig: never

The terraform['vars'] dictionary from the stack configuration is accessible as Terraform variables.

The profile defined in the stack configuration is used to acquire credentials accessible from Terraform. There is two supported providers, the variables which will be loaded depends on the used provider.

• TF_VAR_client_name (if set in .yml stack configuration file)
• TF_VAR_aws_account
• TF_VAR_aws_region
• TF_VAR_aws_access_key
• TF_VAR_aws_secret_key
• TF_VAR_aws_token
• TF_VAR_azurerm_region
• TF_VAR_azure_region
• TF_VAR_azure_subscription_id
• TF_VAR_azure_tenant_id
• TF_VAR_azure_state_access_key (removed in v11.0.0)

The stack path is passed to Terraform. This is especially useful for resource naming and tagging.

• TF_VAR_account
• TF_VAR_environment
• TF_VAR_region
• TF_VAR_stack

All new code contributions should come with unit and/or integrations tests.

To run those tests locally, use tox:

poetry run tox -e py

Linters are also used to ensure code respects our standards.

To run those linters locally:

poetry run tox -e lint

You can get verbose debugging information for argcomplete by defining the following environment variable:

export _ARC_DEBUG=1

Our code is formatted with black.

Make sure to format all your code contributions with black ${filename}.

Hint: enable auto-format on save with black in your favorite IDE.

To run code and documentation style checks, run tox -e lint.

In addition to black --check, code is also checked with:

• flake8, a wrapper for pycodestyle and pyflakes.
• flake8-docstrings, a wrapper for pydocstyle.

This README's table of content is formatted with md_toc.

Keep in mind to update it with md_toc --in-place github README.md.

To build and use development versions of OpenTofu, put them in a ~/.terraform.d/versions/X.Y/X.Y.Z-dev/ folder:

# git clone https://github.com/opentofu/opentofu.git ~/go/src/github.com/opentofu/opentofu
# cd ~/go/src/github.com/opentofu/opentofu
# go build ./cmd/tofu/
# ./tofu version
OpenTofu v1.6.0-dev
on linux_amd64
# mkdir -p ~/.terraform.d/versions/1.6/1.6.0-dev
# mv ./opentofu ~/.terraform.d/versions/1.6/1.6.0-dev/

Some git pre-commit hooks are configured in .pre-commit-config.yaml for use with the pre-commit tool.

Using them helps avoiding to push changes that will fail the CI.

They can be installed locally with:

# pre-commit install

If updating hooks configuration, run checks against all files to make sure everything is fine:

# pre-commit run --all-files --show-diff-on-failure

Note: the pre-commit tool itself can be installed with pip or pipx.

Use the scripts/merge-dependabot-mrs.sh script from master branch to:

• list open Dependabot PRs that are mergeable,
• review, approve and merge them,
• pull changes from github and pushing them to origin.

Just invoke the script without any argument:

# ./scripts/merge-dependabot-mrs.sh

Check the help:

# ./scripts/merge-dependabot-mrs.sh --help

Use the scripts/release.sh script from master branch to:

• bump the version with poetry,
• update CHANGELOG.md,
• commit these changes,
• tag with last CHANGELOG.md item content as annotation,
• bump the version with poetry again to mark it for development,
• commit this change,
• push all commits and tags to all remote repositories.

This will trigger a Github Actions job to publish packages to PyPI.

To invoke the script, pass it the desired bump rule, e.g.:

# ./scripts/release.sh minor

For more options, check the help:

# ./scripts/release.sh --help