tfregistry

Terraform module registry with GCS bucket as a storage backend

Configure the server

We chose to configure the server only with envars, easy to set in a pod

PORT : port to listen to, default 8080
LISTEN : accepted IP range, default 0.0.0.0
BACKEND : storage backend to use, gcs or fake, default gcs
OVERWRITE : accepts to overwrite existing modules with same version, default 0
VERBOSE : debug logs, default 0
MODULE_PATH : default ""
GOOGLE_BUCKET : name of the GCS bucket to use. Mandatory is backend is gcp

curl localhost:8080/test/mymodule/gcp/versions

To create the module v0.0.2 from local tar.gz

curl -X POST --data-binary "@myfile.tar.gz" localhost:8080/test/mymodule/gcp/0.0.2 -H "module-source: https://whatever.com/wherever.git"

curl localhost:8080/test/mymodule/gcp

If you need to change the API, you have to install oapi-codegen to generate code

go get github.com/deepmap/oapi-codegen/cmd/oapi-codegen
make generate

This generates the file pkg/modules/modules.gen.go

We build and push the image using ko from Google

go install github.com/google/ko
make push

You can change the repository by overriding the variable IMAGE_REPOSITORY

make KO_DOCKER_REPO=wherever.com/whatever build

A test bucklet has been created in ml-calleocho-st project. Retrieve the JSON file in https://vault.magicleap.io/ui/vault/secrets/ci/show/sre/tfregistry/gcp_authent

Then

export GOOGLE_APPLICATION_CREDENTIALS=path/to/file.json
export GOOGLE_BUCKET=ml-test-modules-registry
export MODULE_PATH=/
make server

In another window :

curl -v localhost:8080/test/mymodule/gcp/versions

A fake server is also available to test without having to connect to GCP. It is used for unit testing