- OpenTF Settings
- Backends
- gcs
Backend Type: gcs
Stores the state as an object in a configurable prefix in a pre-existing bucket on Google Cloud Storage (GCS). The bucket must exist prior to configuring the backend.
This backend supports state locking.
It is highly recommended that you enable Object Versioning on the GCS bucket to allow for state recovery in the case of accidental deletions and human error.
Example Configuration
terraform {
backend "gcs" {
bucket = "tf-state-prod"
prefix = "opentf/state"
}
}
Data Source Configuration
data "terraform_remote_state" "foo" {
backend = "gcs"
config = {
bucket = "opentf-state"
prefix = "prod"
}
}
resource "local_file" "foo" {
content = data.terraform_remote_state.foo.outputs.greeting
filename = "${path.module}/outputs.txt"
}
Authentication
IAM Changes to buckets are eventually consistent and may take upto a few minutes to take effect. OpenTF will return 403 errors till it is eventually consistent.
Running OpenTF on your workstation.
If you are using OpenTF on your workstation, you will need to install the Google Cloud SDK and authenticate using User Application Default Credentials.
User ADCs do expire and you can refresh them by running gcloud auth application-default login
.
Running OpenTF on Google Cloud
If you are running OpenTF on Google Cloud, you can configure that instance or cluster to use a Google Service Account. This will allow OpenTF to authenticate to Google Cloud without having to bake in a separate credential/authentication file. Make sure that the scope of the VM/Cluster is set to cloud-platform.
Running OpenTF outside of Google Cloud
If you are running OpenTF outside of Google Cloud, generate a service account key and set the GOOGLE_APPLICATION_CREDENTIALS
environment variable to
the path of the service account key. OpenTF will use that key for authentication.
Impersonating Service Accounts
OpenTF can impersonate a Google Service Account as described here. A valid credential must be provided as mentioned in the earlier section and that identity must have the roles/iam.serviceAccountTokenCreator
role on the service account you are impersonating.
Encryption
Take care of your encryption keys because state data encrypted with a lost or deleted key is not recoverable. If you use customer-supplied encryption keys, you must securely manage your keys and ensure you do not lose them. You must not delete customer-managed encryption keys in Cloud KMS used to encrypt state. However, if you accidentally delete a key, there is a time window where you can recover it.
Customer-supplied encryption keys
To get started, follow this guide: Use customer-supplied encryption keys
If you want to remove customer-supplied keys from your backend configuration or change to a different customer-supplied key, OpenTF cannot perform a state migration automatically and manual intervention is necessary instead. This intervention is necessary because Google does not store customer-supplied encryption keys, any requests sent to the Cloud Storage API must supply them instead (see Customer-supplied Encryption Keys). At the time of state migration, the backend configuration loses the old key's details and OpenTF cannot use the key during the migration process.
To migrate your state away from using customer-supplied encryption keys or change the key used by your backend, you need to perform a rewrite (gsutil CLI) or cp (gcloud CLI) operation to remove use of the old customer-supplied encryption key on your state file. Once you remove the encryption, you can successfully run opentf init -migrate-state
with your new backend configuration.
Customer-managed encryption keys (Cloud KMS)
To get started, follow this guide: Use customer-managed encryption keys
If you want to remove customer-managed keys from your backend configuration or change to a different customer-managed key, OpenTF can manage a state migration without manual intervention. This ability is because GCP stores customer-managed encryption keys and are accessible during the state migration process. However, these changes do not fully come into effect until the first write operation occurs on the state file after state migration occurs. In the first write operation after state migration, the file decrypts with the old key and then writes with the new encryption method. This method is equivalent to the rewrite operation described in the customer-supplied encryption keys section. Because of the importance of the first write to state after state migration, you should not delete old KMS keys until any state file(s) encrypted with that key update.
Customer-managed keys do not need to be sent in requests to read files from GCS buckets because decryption occurs automatically within GCS. This process means that if you use the terraform_remote_state
data source to access KMS-encrypted state, you do not need to specify the KMS key in the data source's config
object.
To use customer-managed encryption keys, you need to create a key and give your project's GCS service agent permission to use it with the Cloud KMS CryptoKey Encrypter/Decrypter predefined role.
Configuration Variables
We recommend using environment variables to supply credentials and other sensitive data. If you use -backend-config
or hardcode these values directly in your configuration, OpenTF includes these values in both the .terraform
subdirectory and in plan files. Refer to Credentials and Sensitive Data for details.
The following configuration options are supported:
bucket
- (Required) The name of the GCS bucket. This name must be globally unique. For more information, see Bucket Naming Guidelines.credentials
/GOOGLE_BACKEND_CREDENTIALS
/GOOGLE_CREDENTIALS
- (Optional) Local path to Google Cloud Platform account credentials in JSON format. If unset, the path uses Google Application Default Credentials. The provided credentials must have the Storage Object Admin role on the bucket. Warning: if using the Google Cloud Platform provider as well, it will also pick up theGOOGLE_CREDENTIALS
environment variable.impersonate_service_account
/GOOGLE_BACKEND_IMPERSONATE_SERVICE_ACCOUNT
/GOOGLE_IMPERSONATE_SERVICE_ACCOUNT
- (Optional) The service account to impersonate for accessing the State Bucket. You must haveroles/iam.serviceAccountTokenCreator
role on that account for the impersonation to succeed. If you are using a delegation chain, you can specify that using theimpersonate_service_account_delegates
field.impersonate_service_account_delegates
- (Optional) The delegation chain for an impersonating a service account as described here.access_token
- (Optional) A temporary [OAuth 2.0 access token] obtained from the Google Authorization server, i.e. theAuthorization: Bearer
token used to authenticate HTTP requests to GCP APIs. This is an alternative tocredentials
. If both are specified,access_token
will be used over thecredentials
field.prefix
- (Optional) GCS prefix inside the bucket. Named states for workspaces are stored in an object called<prefix>/<name>.tfstate
.encryption_key
/GOOGLE_ENCRYPTION_KEY
- (Optional) A 32 byte base64 encoded 'customer-supplied encryption key' used when reading and writing state files in the bucket. For more information see Customer-supplied Encryption Keys.kms_encryption_key
/GOOGLE_KMS_ENCRYPTION_KEY
- (Optional) A Cloud KMS key ('customer-managed encryption key') used when reading and writing state files in the bucket. Format should beprojects/{{project}}/locations/{{location}}/keyRings/{{keyRing}}/cryptoKeys/{{name}}
. For more information, including IAM requirements, see Customer-managed Encryption Keys.storage_custom_endpoint
/GOOGLE_BACKEND_STORAGE_CUSTOM_ENDPOINT
/GOOGLE_STORAGE_CUSTOM_ENDPOINT
- (Optional) A URL containing three parts: the protocol, the DNS name pointing to a Private Service Connect endpoint, and the path for the Cloud Storage API (/storage/v1/b
, see here). You can either use a DNS name automatically made by the Service Directory or a custom DNS name made by you. For example, if you create an endpoint calledxyz
and want to use the automatically-created DNS name, you should set the field value ashttps://storage-xyz.p.googleapis.com/storage/v1/b
. For help creating a Private Service Connect endpoint using OpenTF, see this guide.