Make sure the provider is packaged correctly and can be deployed

[crossplane-contrib]

Make sure the provider is packaged correctly and can be deployed

[ulucinar]

We have some alternatives for configuring mirroring:

1. Define a HOME directory for the user who is running the provider process and put a ${HOME}/.terraformrc file there. Please note that we may also need to set the HOME env. variable properly if it's not available.
2. Place.terraformrc to an arbitrary location and run the provider process with the TF_CLI_CONFIG_FILE env. variable pointing to the location of the file.

We can initially start with a .terraformrc client configuration file similar to the following:

  provider_installation {
    filesystem_mirror {
      path    = "/provider-mirror"
      include = ["*/*"]
    }
    direct {
      exclude = ["*/*"]
    }
  }

For this configuration to work, we must have configured the /provider-mirror directory as a mirror directory containing the expected versions of Terraform provider plugin binaries.

And here's an example layout for the mirror folder:

/provider-mirror
└── registry.terraform.io
    ├── hashicorp
        ├── aws
            └── 3.53.0
                └── darwin_arm64
                    └── terraform-provider-aws_v3.53.0_x5

Please note that for this example configuration, tfcli should have been configured with something similar to the following. The configured provider's source and version must match a local mirror:

  tfcli.NewClient(ctx, 
    tfcli.WithProviderSource("hashicorp/aws"),
    tfcli.WithProviderVersion("~> 3.0"),
...)

For the generated provider, we can use an exact Terraform provider version.

In the above layout, terraform-provider-aws_v3.53.0_x5 is the executable Terraform plugin binary. It should have execute permission for the user running the provider process.

The Terraform provider plugin binary's download location can be queried from registry.terraform.io.

Let's also add TF_FORK=0 env. variable (and its value) to prevent the initial TF CLI fork.