This repository contains boilerplate code for building a new Pulumi provider which wraps an existing Terraform provider.
This repository is part of the guide for authoring and publishing a Pulumi Package.
Learn about the concepts behind Pulumi Packages.
The following instructions cover:
- providers maintained by Pulumi (denoted with a "Pulumi Official" checkmark on the Pulumi registry)
- providers published and maintained by the Pulumi community, referred to as "third-party" providers
We showcase a Pulumi-owned provider based on an upstream provider named terraform-provider-foo
. Substitute appropriate values below for your use case.
Note: If the name of the desired Pulumi provider differs from the name of the Terraform provider, you will need to carefully distinguish between the references - see https://github.com/pulumi/pulumi-azure for an example.
Ensure the following tools are installed and present in your $PATH
:
pulumictl
- Go 1.17 or 1.latest
- NodeJS 14.x. We recommend using nvm to manage NodeJS installations.
- Yarn
- TypeScript
- Python (called as
python3
). For recent versions of MacOS, the system-installed version is fine. - .NET
Pulumi offers this repository as a GitHub template repository for convenience. From this repository:
- Click "Use this template".
- Set the following options:
- Owner: pulumi (third-party: your GitHub organization/username)
- Repository name: pulumi-foo (third-party: preface your repo name with "pulumi" as standard practice)
- Description: Pulumi provider for Foo
- Repository type: Public
- Clone the generated repository.
From the templated repository:
-
Run the following command to update files to use the name of your provider (third-party: use your GitHub organization/username):
make prepare NAME=foo REPOSITORY=github.com/pulumi/pulumi-foo ORG=myorg
This will do the following:
- rename folders in
provider/cmd
topulumi-resource-foo
andpulumi-tfgen-foo
- replace dependencies in
provider/go.mod
to reflect your repository name - find and replace all instances of the boilerplate
openai
with theNAME
of your provider. - find and replace all instances of the boilerplate
Tonkean
with theORG
of your provider. - replace all instances of the
github.com/tonkean/pulumi-openai
repository with theREPOSITORY
location
Note for third-party providers:
- If you intend to publish on the Pulumi registry you will want to update the
DisplayName
,Publisher
, andHomepage
values inprovider/resources.go
to use your desired casing.
- rename folders in
-
Modify
README-PROVIDER.md
to include the following (we'll rename it toREADME.md
toward the end of this guide):- Any desired build status badges.
- An introductory paragraph describing the type of resources the provider manages, e.g. "The Foo provider for Pulumi manages resources for Foo.
- In the "Installing" section, correct package names for the various SDK libraries in the languages Pulumi supports.
- In the "Configuration" section, any configurable options for the provider. These may include, but are not limited to, environment variables or options that can be set via
pulumi config set
. - In the "Reference" section, provide a link to the to-be-published documentation.
- Feel free to refer to the Pulumi AWS provider README as an example.
Pulumi provider repositories have the following general structure:
examples/
contains sample code which may optionally be included as integration tests to be run as part of a CI/CD pipeline.provider/
contains the Go code used to create the provider as well as generate the SDKs in the various languages that Pulumi supports.provider/cmd/pulumi-tfgen-foo
generates the Pulumi resource schema (schema.json
), based on the Terraform provider's resources.provider/cmd/pulumi-resource-foo
generates the SDKs in all supported languages from the schema, placing them in thesdk/
folder.provider/pkg/resources.go
is the location where we will define the Terraform-to-Pulumi mappings for resources.
sdk/
contains the generated SDK code for each of the language platforms that Pulumi supports, with each supported platform in a separate subfolder.
-
In
provider/go.mod
, add a reference to the upstream Terraform provider in therequire
section, e.g.github.com/foo/terraform-provider-foo v0.4.0
-
In
provider/resources.go
, ensure the reference in theimport
section uses the correct Go module path, e.g.:github.com/foo/terraform-provider-foo/foo
-
Download the dependencies:
cd provider && go mod tidy && cd -
-
Create the schema by running the following command:
make tfgen
Note warnings about unmapped resources and data sources in the command's output. We map these in the next section, e.g.:
warning: resource foo_something not found in provider map; skipping warning: resource foo_something_else not found in provider map; skipping warning: data source foo_something not found in provider map; skipping warning: data source foo_something_else not found in provider map; skipping
In this section we will add the mappings that allow the interoperation between the Pulumi provider and the Terraform provider. Terraform resources map to an identically named concept in Pulumi. Terraform data sources map to plain old functions in your supported programming language of choice. Pulumi also allows provider functions and resources to be grouped into namespaces to improve the cohesion of a provider's code, thereby making it easier for developers to use. If your provider has a large number of resources, consider using namespaces to improve usability.
The following instructions all pertain to provider/resources.go
, in the section of the code where we construct a tfbridge.ProviderInfo
object:
-
Add resource mappings: For each resource in the provider, add an entry in the
Resources
property of thetfbridge.ProviderInfo
, e.g.:// Most providers will have all resources (and data sources) in the main module. // Note the mapping from snake_case HCL naming conventions to UpperCamelCase Pulumi SDK naming conventions. // The name of the provider is omitted from the mapped name due to the presence of namespaces in all supported Pulumi languages. "foo_something": {Tok: tfbridge.MakeResource(mainPkg, mainMod, "Something")}, "foo_something_else": {Tok: tfbridge.MakeResource(mainPkg, mainMod, "SomethingElse")},
-
Add CSharpName (if necessary): Dotnet does not allow for fields named the same as the enclosing type, which sometimes results in errors during the dotnet SDK build. If you see something like
error CS0542: 'ApiKey': member names cannot be the same as their enclosing type [/Users/guin/go/src/github.com/pulumi/pulumi-artifactory/sdk/dotnet/Pulumi.Artifactory.csproj]
you'll want to give your Resource a CSharpName, which can have any value that makes sense:
"foo_something_dotnet": { Tok: tfbridge.MakeResource(mainPkg, mainMod, "SomethingDotnet"), Fields: map[string]*tfbridge.SchemaInfo{ "something_dotnet": { CSharpName: "SpecialName", }, }, },
-
Add data source mappings: For each data source in the provider, add an entry in the
DataSources
property of thetfbridge.ProviderInfo
, e.g.:// Note the 'get' prefix for data sources "foo_something": {Tok: tfbridge.MakeDataSource(mainPkg, mainMod, "getSomething")}, "foo_something_else": {Tok: tfbridge.MakeDataSource(mainPkg, mainMod, "getSomethingElse")},
-
Add documentation mapping (sometimes needed): If the upstream provider's repo is not a part of the
terraform-providers
GitHub organization, specify theGitHubOrg
property oftfbridge.ProviderInfo
to ensure that documentation is picked up by the codegen process, and that attribution for the upstream provider is correct, e.g.:GitHubOrg: "foo",
-
Add provider configuration overrides (not typically needed): Pulumi's Terraform bridge automatically detects configuration options for the upstream provider. However, in rare cases these settings may need to be overridden, e.g. if we want to change an environment variable default from
API_KEY
toFOO_API_KEY
. Examples of common uses cases:"additional_required_parameter": {}, "additional_optional_string_parameter": { Default: &tfbridge.DefaultInfo{ Value: "default_value", }, "additional_optional_boolean_parameter": { Default: &tfbridge.DefaultInfo{ Value: true, }, // Renamed environment variables can be accounted for like so: "apikey": { Default: &tfbridge.DefaultInfo{ EnvVars: []string{"FOO_API_KEY"}, },
-
Build the provider binary and ensure there are no warnings about unmapped resources and no warnings about unmapped data sources:
make provider
You may see warnings about documentation and examples, including "unexpected code snippets". These can be safely ignored for now. Pulumi will add additional documentation on mapping docs in a future revision of this guide.
-
Build the SDKs in the various languages Pulumi supports:
make build_sdks
-
Ensure the Golang SDK is a proper go module:
cd sdk && go mod tidy && cd -
This will pull in the correct dependencies in
sdk/go.mod
as well as setting the dependency tree insdk/go.sum
. -
Finally, ensure the provider code conforms to Go standards:
make lint_provider
Fix any issues found by the linter.
Note: If you make revisions to code in resources.go
, you must re-run the make tfgen
target to regenerate the schema.
The make tfgen
target will take the file schema.json
and serialize it to a byte array so that it can be included in the build output.
(This is a holdover from Go 1.16, which does not have the ability to directly embed text files. We are working on removing the need for this step.)
In this section, we will create a Pulumi program in TypeScript that utilizes the provider we created to ensure everything is working properly.
-
Create an account with the provider's service and generate any necessary credentials, e.g. API keys.
- Email: bot@pulumi.com
- Password: (Create a random password in 1Password with the maximum length and complexity allowed by the provider.)
- Ensure all secrets (passwords, generated API keys) are stored in Pulumi's 1Password vault.
-
Copy the
pulumi-resource-foo
binary generated bymake provider
and place it in your$PATH
($GOPATH/bin
is a convenient choice), e.g.:cp bin/pulumi-resource-foo $GOPATH/bin
-
Tell Yarn to use your local copy of the SDK:
make install_nodejs_sdk
-
Create a new Pulumi program in the
examples/
directory, e.g.:mkdir examples/my-example/ts # Change "my-example" to something more meaningful. cd examples/my-example/ts pulumi new typescript # (Go through the prompts with the default values) npm install yarn link @pulumi/foo
-
Create a minimal program for the provider, i.e. one that creates the smallest-footprint resource. Place this code in
index.ts
. -
Configure any necessary environment variables for authentication, e.g
$FOO_USERNAME
,$FOO_TOKEN
, in your local environment. -
Ensure the program runs successfully via
pulumi up
. -
Once the program completes successfully, verify the resource was created in the provider's UI.
-
Destroy any resources created by the program via
pulumi destroy
.
Optionally, you may create additional examples for SDKs in other languages supported by Pulumi:
-
Python:
mkdir examples/my-example/py cd examples/my-example/py pulumi new python # (Go through the prompts with the default values) source venv/bin/activate # use the virtual Python env that Pulumi sets up for you pip install pulumi_foo
-
Follow the steps above to verify the program runs successfully.
We can run integration tests on our examples using the *_test.go
files in the examples/
folder.
-
Add code to
examples_nodejs_test.go
to call the example you created, e.g.:// Swap out MyExample and "my-example" below with the name of your integration test. func TestAccMyExampleTs(t *testing.T) { test := getJSBaseOptions(t). With(integration.ProgramTestOptions{ Dir: filepath.Join(getCwd(t), "my-example", "ts"), }) integration.ProgramTest(t, &test) }
-
Add a similar function for each example that you want to run in an integration test. For examples written in other languages, create similar files for
examples_${LANGUAGE}_test.go
. -
You can run these tests locally via Make:
make test
You can also run each test file separately via test tags:
cd examples && go test -v -tags=nodejs
- Follow the instructions laid out in the deployment templates.
In this section, we'll add the necessary configuration to work with GitHub Actions for Pulumi's standard CI/CD workflows for providers.
-
Generate GitHub workflows per the instructions in the ci-mgmt repository and copy to
.github/
in this repository. -
Ensure that any required secrets are present as repository-level secrets in GitHub. These will be used by the integration tests during the CI/CD process.
-
Repository settings: Toggle
Allow auto-merge
on in your provider repo to automate GitHub Actions workflow updates.
-
Ensure all required configurations (API keys, etc.) are documented in README-PROVIDER.md.
-
Replace this file with the README for the provider and push your changes:
mv README-PROVIDER.md README.md
-
If publishing the npm package fails during the "Publish SDKs" Action, perform the following steps:
- Go to NPM Packages and sign in as pulumi-bot.
- Click on the bot's profile pic and navigate to "Packages".
- On the left, under "Organizations, click on the Pulumi organization.
- On the last page of the listed packages, you should see the new package.
- Under "Settings", set the Package Status to "public".
Now you are ready to use the provider, cut releases, and have some well-deserved π¨!
There are 2 ways the provider can be built locally:
make provider
will use the current operating system and architecture to create a binary that can be used on your PATH.
To build the provider for another set of operating systems / architectures, the project uses goreleaser. Goreleaser, a CLI tool, that allows a user to build a matrix of binaries.
Create a .goreleaser.yml
file in the root of your project:
archives:
- id: archive
name_template: "{{ .Binary }}-{{ .Tag }}-{{ .Os }}-{{ .Arch }}"
before:
hooks:
- make tfgen
builds:
- binary: pulumi-resource-openai
dir: provider
goarch:
- amd64
- arm64
goos:
- darwin
- windows
- linux
ignore: []
ldflags:
- -X github.com/tonkean/pulumi-openai/provider/pkg/version.Version={{.Tag}}
main: ./cmd/pulumi-resource-openai/
sort: asc
use: git
release:
disable: false
snapshot:
name_template: "{{ .Tag }}-SNAPSHOT"
To build the provider for the combination of architectures and operating systems, you can run the following CLI command:
goreleaser build --rm-dist --skip-validate
That will ensure that a list of binaries are available to use:
βΆ tree dist
dist
βββ CHANGELOG.md
βββ artifacts.json
βββ config.yaml
βββ metadata.json
βββ pulumi-openai_darwin_amd64_v1
β βββ pulumi-resource-openai
βββ pulumi-openai_darwin_arm64
β βββ pulumi-resource-openai
βββ pulumi-openai_linux_amd64_v1
β βββ pulumi-resource-openai
βββ pulumi-openai_linux_arm64
β βββ pulumi-resource-openai
βββ pulumi-openai_windows_amd64_v1
β βββ pulumi-resource-openai.exe
βββ pulumi-openai_windows_arm64
βββ pulumi-resource-openai.exe
Any of the provider binaries can be used to target the correct machine architecture
If you receive the following error: use of internal package github.com/example/terraform-provider-example/internal/provider not allowed
, you need to use
the shim model below, and replace the example item:
mkdir -p provider/shim
cat <<-EOF> provider/shim/go.mod
module github.com/example/terraform-provider-example/shim
go 1.18
require github.com/hashicorp/terraform-plugin-sdk/v2 v2.22.0
require github.com/example/terraform-provider-example v1.0.0
EOF
cat <<-EOF> provider/shim/shim.go
package shim
import (
"github.com/hashicorp/terraform-plugin-sdk/v2/helper/schema"
"github.com/example/terraform-provider-example/internal/provider"
)
// fix provider.Provider here to match whats in internal/provider
func Provider() *schema.Provider {
return provider.Provider()
}
EOF
cd provider/shim/ && go mod tidy && cd ../../
cat <<EOF>> provider/go.mod
replace github.com/example/terraform-provider-example/shim => ./shim
require github.com/example/terraform-provider-example/shim v0.0.0
EOF
cd provider && go mod tidy