/magic-modules

Add Google Cloud Platform support to Terraform

Primary LanguageHTMLApache License 2.0Apache-2.0

Magic Modules Logo

Magic Modules

Magic Modules is a code generator and CI system that's used to develop the Terraform providers for Google Platform, google (or TPG) and google-beta (or TPGB).

Magic Modules allows contributors to make changes against a single codebase and develop both provider versions simultaneously. After sending a pull request against this repository, the modular-magician robot user will manage (most of) the heavy lifting from generating a complete output, running presubmit tests, and updating the providers following your change.



Getting Started with Magic Modules

Cloning Terraform providers

If you're generating the Terraform providers (google and google-beta), you'll need to check out the repo(s) you're generating in your GOPATH. For example:

git clone https://github.com/hashicorp/terraform-provider-google.git $GOPATH/src/github.com/hashicorp/terraform-provider-google
git clone https://github.com/hashicorp/terraform-provider-google-beta.git $GOPATH/src/github.com/hashicorp/terraform-provider-google-beta

Or run the following to check them all out:

./tools/bootstrap

Magic Modules won't work with old versions of the Terraform provider repos. If you're encountering issues with vendoring and paths, make sure both MM and the Terraform provider are running on up to date copies of main.

Setting up a container-based environment

NOTE this approach is in beta and still collecting feedback. Please file an issue if you encounter challenges, and try pulling the latest container (see command below) first to see if any recent changes may fix you.

You do not need to run these instructions if you are setting up your environment manually.

For ease of contribution, we provide containers with the required dependencies for building magic-modules, as well as the option to build them yourself.

You can work with containers with either Podman or Docker.

scripts/make-in-container.sh includes all the instructions to build and run containers by hand. Refer to the script for individual steps, but for most users you only have to run the script directly, as a replacement for make.

Here is an example of how to build Terraform (after cloning the provider):

./scripts/make-in-container.sh \
  terraform VERSION=ga \
  OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google"

Generally, you can replace any reference to make in this guide with scripts/make-in-container.sh.

Preparing your environment manually

NOTE: you don't need to run these instructions if you are using a container-based environment.

You can also build magic-modules within your local development environment.

To get started, you'll need:

  • Go
    • If you're using a Mac with Homebrew installed, you can follow these instructions to set up Go: YouTube video.
    • If you're using Cloud Shell, Go is already installed.
  • Ruby 2.6.0
    • You can use rbenv to manage your Ruby version(s).
    • To install rbenv:
      • Homebrew: run brew install rbenv ruby-build
      • Debian, Ubuntu, and their derivatives: run sudo apt install rbenv
    • Then run rbenv install 2.6.0.
      • For M1 Mac users, run RUBY_CFLAGS="-Wno-error=implicit-function-declaration" rbenv install 2.6.0
  • Bundler
    • This can be installed with gem install bundler
  • Gems for magic-modules
    • This can be installed with cd mmv1 && bundler install
  • Goimports
    • go install golang.org/x/tools/cmd/goimports / go install golang.org/x/tools/cmd/goimports@latest
  • Terraform
  • If you are getting "Too many open files" ulimit needs to be raised.
    • Mac OSX: ulimit -n 1000

Now, you can verify you're ready with:

./tools/doctor
Expected output:
Check for rbenv in path...
   found!
Checking ruby version...
2.6.0 (set by [PATH]/magic-modules/mmv1/.ruby-version)
Check for bundler in path...
   found!
Check for go in path...
   found!
Check for goimports in path...
   found!
Check for git in path...
   found!

Generating the Terraform Providers

Before making any changes, you can compile the Terraform provider you're working on by running the following command. If Magic Modules has been installed correctly, you'll get no errors.

The following commands should be run from the root directory of the repository. OUTPUT_PATH should be set to the location of your provider repository, which is recommended to be inside your GOPATH.

cd magic-modules

make terraform VERSION=ga OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google"
make terraform VERSION=beta OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google-beta"

# Only generate a specific product (plus all common files)
make terraform VERSION=ga OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google" PRODUCT=pubsub

# Only generate only a specific resources for a product
make terraform VERSION=ga OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google" PRODUCT=pubsub RESOURCE=Topic

The PRODUCT variable values correspond to folder names in mmv1/products for generated resources. The RESOURCE variable value needs to match the name of the resource inside the api.yaml file for that product.

Handwritten files in mmv1/third_party are always compiled. If you are only working on common files or third_party code, you can pass a non-existent PRODUCT to reduce the generation time.

# Only generate common files, including all third_party code
make terraform VERSION=ga OUTPUT_PATH="$GOPATH/src/github.com/hashicorp/terraform-provider-google" PRODUCT=foo

It's worth noting that Magic Modules will only generate new files when run locally. The "Magician"- the Magic Modules CI system- handles deletion of old files when creating PRs.


Testing

Once you've made changes to resource definition, you can run Magic Modules to generate changes to your provider; see "Generating the Terraform Providers" above if you need a refresher. Once it's generated, you should run the tests to ensure your change work for the providers.

Tests generally assume the following environment variables must be set in order to run tests:

GOOGLE_PROJECT
GOOGLE_CREDENTIALS|GOOGLE_CLOUD_KEYFILE_JSON|GCLOUD_KEYFILE_JSON|GOOGLE_USE_DEFAULT_CREDENTIALS
GOOGLE_REGION
GOOGLE_ZONE

Note that the credentials you provide must be granted wide permissions on the specified project. These tests provision real resources, and require permission in order to do so. Most developers on the team grant their test service account roles/editor or roles/owner on their project. Additionally, to ensure that your tests are performed in a region and zone with wide support for GCP features, GOOGLE_REGION should be set to us-central1 and GOOGLE_ZONE to us-central1-a.

Additional variable may be required for other tests, and should get flagged when running them by Go skipping the test and flagging in the output it was skipped, with a skip message explaining why. The most typical extra values required are those required for project creation:

GOOGLE_ORG
GOOGLE_BILLING_ACCOUNT

You can run tests against the provider you generated in the OUTPUT_PATH location. When running tests, specify which to run using TESTARGS, such as:

# for ga provider
cd $GOPATH/src/github.com/hashicorp/terraform-provider-google
TF_LOG=TRACE make testacc TEST=./google TESTARGS='-run=TestAccContainerNodePool_basic' > output.log

# for beta provider
cd $GOPATH/src/github.com/hashicorp/terraform-provider-google-beta
TF_LOG=TRACE make testacc TEST=./google-beta TESTARGS='-run=TestAccContainerNodePool_basic' > output.log

The TESTARGS variable is regexp-like, so multiple tests can be run in parallel by specifying a common substring of those tests (for example, TestAccContainerNodePool to run all node pool tests). There are 2000+ tests, and running all of them takes over 9 hours and requires a lot of GCP quota.

Note: TF_LOG=TRACE is optional; it enables verbose logging during tests, including all API request/response cycles. > output.log redirects the test output to a file for analysis, which is useful because TRACE logging can be extremely verbose.

Tests will use whatever version of the terraform binary is found on your path. To test with multiple versions of terraform core, you must run the tests multiple times with different versions. You can use tfenv to manage your system terraform versions.

Using released terraform binary with local provider binary

After the provider code is generated into the location at OUTPUT_PATH, you need to build the code in your generated provider to compile the provider binary:

cd $GOPATH/src/github.com/hashicorp/terraform-provider-google
make build

Setup:

mkdir -p ~/.terraform.d/plugins/registry.terraform.io/hashicorp/google/5.0.0/darwin_arm64
mkdir -p ~/.terraform.d/plugins/registry.terraform.io/hashicorp/google-beta/5.0.0/darwin_arm64
ln -s $GOPATH/bin/terraform-provider-google ~/.terraform.d/plugins/registry.terraform.io/hashicorp/google/5.0.0/darwin_arm64/terraform-provider-google_v5.0.0
ln -s $GOPATH/bin/terraform-provider-google-beta ~/.terraform.d/plugins/registry.terraform.io/hashicorp/google-beta/5.0.0/darwin_arm64/terraform-provider-google-beta_v5.0.0

NOTE: Substitute your machine architecture for darwin_arm64, i.e. darwin_amd64 or linux_amd64

Once this setup is complete, terraform will automatically use the binaries generated by the make build commands in the terraform-provider-google and terraform-provider-google-beta repositories instead of downloading the latest versions. To undo this, you can run:

rm -rf ~/.terraform.d/plugins/registry.terraform.io/hashicorp/

For more information, check out Hashicorp's documentation on the 0.13+ filesystem layout.

If multiple versions are available in a plugin directory (for example after terraform providers mirror is used), Terraform will pick the most up-to-date provider version within version constraints. As such, we recommend using a version that is several major versions ahead for your local copy of the provider, such as 5.0.0.


Submitting a PR

Before creating a commit, if you've modified any .rb files, make sure you run rake test! That will run rubocop to ensure that the code you've written will pass Travis.

Once you've created your commit(s), you can submit the code normally as a PR in the GitHub UI. The PR template includes some instructions to make sure we generate good PR messages for the tools' repo histories.

Once your PR is submitted, one of the Magic Modules maintainers will review it. They'll look over the code before running the "Magician", the Magic Modules CI system that generates PRs against each tool. Each review pass, your reviewer will run the Magician again to update the PRs against the tools.

If there are multiple tools affected, that first reviewer will be the "primary" reviewer, and for each other affected tool a maintainer for that specific tool will make a pass. The primary reviewer will make it clear which other maintainers need to review, and prompt them to review your code; you will communicate primarily with the first reviewer.

Even when multiple tools are affected, this will generally be a quick look by that maintainer with no changes needing to be made.

Once you've gotten approvals from the primary reviewer and the reviewers for any affected tools, the primary reviewer will merge your changes.


Contributing

General contributing steps

  1. Fork Magic Modules repository into your GitHub account if you haven't done before.

  2. Check the issue tracker to see whether your feature has already been requested.

    • if there's an issue and it's already has a dedicated assignee, it indicates that someone may have already started to work on a solution.
    • otherwise, you're welcome to work on the issue.
  3. Check whether the resource you would like to work on already exists in the providers (google / google-beta or check the website).

    • If it exists, check the header of the downstream file to identify the type of tools used to generate the resource. For some resources, the code file, the test file and the documentation file may not be generated via the same tools.
      • Generated resources like google_compute_address can be identified by looking in their Go source for an AUTO GENERATED CODE header as well as a Type. "Generated resources" typically refers to just the MMv1 type, and DCL type resources are considered "DCL-based". (Currently DCL-related contribution are not supported)
      • Handwritten resources like google_container_cluster can be identified if they have source code present under the mmv1/third_party/terraform/resources folder or by the absence of the AUTO GENERATED CODE header in their Go source.
    • If not, decide which tool you would like to use to implement the resource.
      • MMv1 is strongly preferred over handwriting the resource unless the resource can not be generated.
      • Currently, only handwritten datasources are supported.
  4. Make the actual code change.

    • The Contribution Guide below will guide you to the detailed instructions on how to make your change, based on the type of the change + the tool used to generate the code.
  5. Build the providers that includes your change. Check Generating the Terraform Providers section for details on how to generate the providers locally.

  6. Test the feature against the providers you generated in the last step locally. Check Testing Guidance for details on how to run provider test locally. (Testing the PR locally and pushing the commit to the PR only after the tests pass locally may significantly reduce review cycles)

  7. Push your changes to your magic-modules repo fork and send a pull request from that branch to the main branch on magic-modules. A reviewer will be assigned automatically to your PR. Check Submitting a PR section for details on how to submit a PR.

  8. Wait until the the modules magician to generate downstream diff (which should takes about 15 mins after creating the PR) to make sure all changes are generated correctly in downstream repos.

  9. Wait for the VCR test results.

    Get to know general workflow for VCR tests
    1. You submit your change.
    2. The recorded tests are ran against your changes by the modular-magician. Tests will fail if:
      1. Your PR has changed the HTTP request values sent by the provider
      2. Your PR does not change the HTTP request values, but fails on the values returned in an old recording
      3. The recordings are out of sync with the merge-base of your PR, and an unrelated contributor's change has caused a false positive
    3. The modular-magician will leave a message indicating the number of passing and failing VCR tests. If there is a failure, the modular-magician user will leave a message indicating the "Triggering VCR tests in RECORDING mode for the following tests that failed during VCR:" marking which tests failed.
      1. If a test does not appear related, it probably isn't!
    4. The modular-magician will kick off a second test run targeting only the failed tests, this time hitting the live GCP APIs. If there are tests that fail at this point, a message stating Tests failed during RECORDING mode: will be left indicating the tests.
      1. If a test that appears to be related to your change has failed here, it's likely your change has introduced an issue. You can view the debug logs for the test by clicking the "view" link beside the test case to attempt to debug what's going wrong.
      2. If a test appears to be completely unrelated has failed, it's possible that a GCP API has changed in a way that broke the provider or our environment capped on a quota.

    Where possible, take a look at the logs and see if you can figure out what needs to be fixed related to your change. The false positive rate on these tests is extremely high between changes in the API, Cloud Build bugs, and eventual consistency issues in test recordings so we don't expect contributors to wholly interpret the results- that's the responsibility of your reviewer.

  10. If your assigned reviewers does not reply/ review within a week, gently ping them on github.

  11. After your PR is merged, it will be released to customers in around a week or two.


Detailed contributing guide

Task Section
Resource handwritten / mmv1
Datasource handwritten / (only handwritten datasources are supported)
IAM resource handwritten / mmv1
Testing handwritten / mmv1
Documentation handwritten / mmv1
Beta feature handwritten / mmv1

Glossary

The maintainers of the repository will tend to use specific jargon to describe concepts related to Magic Modules; here's a quick reference of what some of those terms are.

Term Definition
tool One of the OSS DevOps projects Magic Modules generates GCP support in
provider Synonym for tool as referred to inside the codebase
downstream(s) A PR created by the Magician against a tool
upstream A PR created against Magic Modules or the Magic Modules repo
The Magician The Magic Modules CI system that drives the GitHub robot modular-magician

Other Resources