Blocks

Terragrunt HCL configuration uses configuration blocks when there’s a structural configuration that needs to be defined for Terragrunt.

Think of configuration blocks as a way to control different systems used by Terragrunt, whereas attributes are used to define values for those systems.

terraform

The terraform block is used to configure how Terragrunt will interact with OpenTofu/Terraform. This includes specifying where to find the OpenTofu/Terraform configuration files, any extra arguments to pass to the tofu/terraform binary, and any hooks to run before or after calling OpenTofu/Terraform.

The terraform block supports the following arguments:

• source (attribute): Specifies where to find OpenTofu/Terraform configuration files. This parameter supports the same syntax as the module source parameter for OpenTofu/Terraform module blocks except for the Terraform registry (see below note), including local file paths, Git URLs, and Git URLS with ref parameters. Terragrunt will download all the code in the repo (i.e. the part before the double-slash //) so that relative paths work correctly between modules in that repo.

  • The source parameter can be configured to pull OpenTofu/Terraform modules from any Terraform module registry using the tfr protocol. The tfr protocol expects URLs to be provided in the format tfr://REGISTRY_HOST/MODULE_SOURCE?version=VERSION. For example, to pull the terraform-aws-modules/vpc/aws module from the public Terraform registry, you can use the following as the source parameter: tfr://registry.terraform.io/terraform-aws-modules/vpc/aws?version=3.3.0.
  • If you wish to access a private module registry (e.g., Terraform Cloud/Enterprise), you can provide the authentication to Terragrunt as an environment variable with the key TG_TF_REGISTRY_TOKEN. This token can be any registry API token.
  • The tfr protocol supports a shorthand notation where the REGISTRY_HOST can be omitted to default to the public registry. The default registry depends on the wrapped executable: for Terraform, it is registry.terraform.io, and for Opentofu, it is registry.opentofu.org. Additionally, if the environment variable TG_TF_DEFAULT_REGISTRY_HOST is set, this value will be used as the default registry host instead, overriding the standard defaults for the wrapped executable.
  • If you use tfr:/// (note the three /). For example, the following will fetch the terraform-aws-modules/vpc/aws module from the public registry: tfr:///terraform-aws-modules/vpc/aws?version=3.3.0.
  • You can also use submodules from the registry using //. For example, to use the iam-policy submodule from the registry module terraform-aws-modules/iam, you can use the following: tfr:///terraform-aws-modules/iam/aws//modules/iam-policy?version=4.3.0.

• include_in_copy (attribute): A list of glob patterns (e.g., ["*.txt"]) that should always be copied from the same directory containing terragrunt.hcl into the OpenTofu/Terraform working directory. When you use the source param in your Terragrunt config and run terragrunt <command>, Terragrunt will download the code specified at source into a scratch folder (.terragrunt-cache, by default), copy the code in your current working directory into the same scratch folder, and then run tofu <command> (or terraform <command>) in that scratch folder. By default, Terragrunt excludes hidden files and folders during the copy step. This feature allows you to specify glob patterns of files that should always be copied from the Terragrunt working directory. Additional notes:

  • The path should be specified relative to the source directory.
  • This list is also used when using a local file source (e.g., source = "../modules/vpc"). For example, if your OpenTofu/Terraform module source contains a hidden file that you want to copy over (e.g., a .python-version file), you can specify that in this list to ensure it gets copied over to the scratch copy (e.g., include_in_copy = [".python-version"]).

• exclude_from_copy (attribute): A list of glob patterns (e.g., ["*.txt"]) that should always be skipped from the same directory containing terragrunt.hcl when copying into the OpenTofu/Terraform working directory. All examples valid for include_in_copy can be used here.

  Note that using include_in_copy and exclude_from_copy are not mutually exclusive. If a file matches a pattern in both include_in_copy and exclude_from_copy, it will not be included. If you would like to ensure that the file is included, make sure the patterns you use for include_in_copy do not match the patterns in exclude_from_copy.

  Note that if you wish to exclude files from being copied from a terraform module source, you should use the before_hook feature.

• copy_terraform_lock_file (attribute): In certain use cases, you don’t want to check the terraform provider lock file into your source repository from your working directory as described in Lock File Handling. This attribute allows you to disable the copy of the generated or existing .terraform.lock.hcl from the temp folder into the working directory. Default is true.

• extra_arguments (block): Nested blocks used to specify extra CLI arguments to pass to the tofu/terraform binary. Learn more about its usage in the Keep your CLI flags DRY use case overview. Supports the following arguments:

  • arguments (required) : A list of CLI arguments to pass to tofu/terraform.
  • commands (required) : A list of tofu/terraform sub commands that the arguments will be passed to.
  • env_vars (optional) : A map of key value pairs to set as environment variables when calling tofu/terraform.
  • required_var_files (optional): A list of file paths to OpenTofu/Terraform vars files (.tfvars) that will be passed in to terraform as -var-file=<your file>.
  • optional_var_files (optional): A list of file paths to OpenTofu/Terraform vars files (.tfvars) that will be passed in to tofu/terraform like required_var_files, only any files that do not exist are ignored.

• before_hook (block): Nested blocks used to specify command hooks that should be run before tofu/terraform is called. Hooks run from the directory with the OpenTofu/Terraform module, except for hooks related to read-config and init-from-module. These hooks run in the terragrunt configuration directory (the directory where terragrunt.hcl lives). Supports the following arguments:

  • commands (required) : A list of tofu/terraform sub commands for which the hook should run before.
  • execute (required) : A list of command and arguments that should be run as the hook. For example, if execute is set as ["echo", "Foo"], the command echo Foo will be run.
  • working_dir (optional) : The path to set as the working directory of the hook. Terragrunt will switch directory to this path before running the hook command. Defaults to the terragrunt configuration directory for read-config and init-from-module hooks, and the OpenTofu/Terraform module directory for other command hooks.
  • run_on_error (optional) : If set to true, this hook will run even if a previous hook hit an error, or in the case of “after” hooks, if the OpenTofu/Terraform command hit an error. Default is false.
  • suppress_stdout (optional) : If set to true, the stdout output of the executed commands will be suppressed. This can be useful when there are scripts relying on OpenTofu/Terraform’s output and any other output would break their parsing.
  • if (optional) : hook will be skipped when the argument is set or evaluates to false.

• after_hook (block): Nested blocks used to specify command hooks that should be run after tofu/terraform is called. Hooks run from the terragrunt configuration directory (the directory where terragrunt.hcl lives). Supports the same arguments as before_hook.

• error_hook (block): Nested blocks used to specify command hooks that run when an error is thrown. The error must match one of the expressions listed in the on_errors attribute. Error hooks are executed after the before/after hooks.

In addition to supporting before and after hooks for all OpenTofu/Terraform commands, the following specialized hooks are also supported:

• read-config (after hook only): read-config is a special hook command that you can use with the after_hook subblock to run an action immediately after terragrunt finishes loading the config. This hook will run on every invocation of terragrunt. Note that you can only use this hook with after_hooks. Any before_hooks with the command read-config will be ignored. The working directory for hooks associated with this command will be the terragrunt config directory.

• init-from-module and init: Terragrunt has two stages of initialization: one is to download remote configurations using go-getter; the other is Auto-Init, which configures the backend and downloads provider plugins and modules. If you wish to run a hook when Terragrunt is using go-getter to download remote configurations, use init-from-module for the command. If you wish to execute a hook when Terragrunt is using tofu init/terraform init for Auto-Init, use init for the command. For example, an after_hook for the command init-from-module will run after terragrunt clones the module, while an after_hook for the command init will run after terragrunt runs tofu init/terraform init on the cloned module.

  • Hooks for both init-from-module and init only run if the requisite stage needs to run. That is, if terragrunt detects that the module is already cloned in the terragrunt cache, this stage will be skipped and thus the hooks will not run. Similarly, if terragrunt detects that it does not need to run init in the auto init feature, the init stage is skipped along with the related hooks.
  • The working directory for hooks associated with init-from-module will run in the terragrunt config directory, while the working directory for hooks associated with init will be the OpenTofu/Terraform module.

Complete Example:

terragrunt.hcl

terraform {
  # Pull the OpenTofu/Terraform configuration at the github repo "acme/infrastructure-modules", under the subdirectory
  # "networking/vpc", using the git tag "v0.0.1".
  source = "git::git@github.com:acme/infrastructure-modules.git//networking/vpc?ref=v0.0.1"

  # For any OpenTofu/Terraform commands that use locking, make sure to configure a lock timeout of 20 minutes.
  extra_arguments "retry_lock" {
    commands  = get_terraform_commands_that_need_locking()
    arguments = ["-lock-timeout=20m"]
  }

  # You can also specify multiple extra arguments for each use case. Here we configure terragrunt to always pass in the
  # `common.tfvars` var file located by the parent terragrunt config.
  extra_arguments "custom_vars" {
    commands = [
      "apply",
      "plan",
      "import",
      "push",
      "refresh"
    ]

    required_var_files = ["${get_parent_terragrunt_dir()}/common.tfvars"]
  }

  # The following are examples of how to specify hooks

  # Before apply or plan, run "echo Foo".
  before_hook "before_hook_1" {
    commands     = ["apply", "plan"]
    execute      = ["echo", "Foo"]
  }

  # Before apply, run "echo Bar". Note that blocks are ordered, so this hook will run after the previous hook to
  # "echo Foo". In this case, always "echo Bar" even if the previous hook failed.
  before_hook "before_hook_2" {
    commands     = ["apply"]
    execute      = ["echo", "Bar"]
    run_on_error = true
  }

  # Note that you can use interpolations in subblocks. Here, we configure it so that before apply or plan, print out the
  # environment variable "HOME".
  before_hook "interpolation_hook_1" {
    commands     = ["apply", "plan"]
    execute      = ["echo", get_env("HOME", "HelloWorld")]
    run_on_error = false
  }

  # After running apply or plan, run "echo Baz". This hook is configured so that it will always run, even if the apply
  # or plan failed.
  after_hook "after_hook_1" {
    commands     = ["apply", "plan"]
    execute      = ["echo", "Baz"]
    run_on_error = true
  }

  # After an error occurs during apply or plan, run "echo Error Hook executed". This hook is configured so that it will run
  # after any error, with the ".*" expression.
  error_hook "error_hook_1" {
    commands  = ["apply", "plan"]
    execute   = ["echo", "Error Hook executed"]
    on_errors = [
      ".*",
    ]
  }

  # A special after hook to always run after the init-from-module step of the Terragrunt pipeline. In this case, we will
  # copy the "foo.tf" file located by the parent terragrunt.hcl file to the current working directory.
  after_hook "init_from_module" {
    commands = ["init-from-module"]
    execute  = ["cp", "${get_parent_terragrunt_dir()}/foo.tf", "."]
  }

  # A special after_hook. Use this hook if you wish to run commands immediately after terragrunt finishes loading its
  # configurations. If "read-config" is defined as a before_hook, it will be ignored as this config would
  # not be loaded before the action is done.
  after_hook "read-config" {
    commands = ["read-config"]
    execute  = ["bash", "script/get_aws_credentials.sh"]
  }
}

Local File Path Example with allowed hidden files:

terragrunt.hcl

terraform {
  # Pull the OpenTofu/Terraform configuration from the local file system. Terragrunt will make a copy of the source folder in the
  # Terragrunt working directory (typically `.terragrunt-cache`).
  source = "../modules/networking/vpc"

  # Always include the following file patterns in the Terragrunt copy.
  include_in_copy = [
    ".security_group_rules.json",
    "*.yaml",
  ]
}

A note about using modules from the registry

The key design of Terragrunt is to act as a preprocessor to convert shared service modules in the registry into a root module. In OpenTofu/Terraform, modules can be loosely categorized into two types:

• Root Module: An OpenTofu/Terraform module that is designed for running tofu init/terraform init and the other workflow commands (apply, plan, etc.). This is the entrypoint module for deploying your infrastructure. Root modules are identified by the presence of key blocks that setup configuration about how OpenTofu/Terraform behaves, like backend blocks (for configuring state) and provider blocks (for configuring how OpenTofu/Terraform interacts with the cloud APIs).
• Shared Module: A OpenTofu/Terraform module that is designed to be included in other OpenTofu/Terraform modules through module blocks. These modules are missing many of the key blocks that are required for running the workflow commands of OpenTofu/Terraform.

Terragrunt further distinguishes shared modules between service modules and modules:

• Shared Service Module: An OpenTofu/Terraform module that is designed to be standalone and applied directly. These modules are not root modules in that they are still missing the key blocks like backend and provider, but aside from that do not need any additional configuration or composition to deploy. For example, the terraform-aws-modules/vpc module can be deployed by itself without composing with other modules or resources.
• Shared Module: An OpenTofu/Terraform module that is designed to be composed with other modules. That is, these modules must be embedded in another OpenTofu/Terraform module and combined with other resources or modules. For example, the consul-security-group-rules module

Terragrunt started off with features that help directly deploy Root Modules, but over the years have implemented many features that allow you to turn Shared Service Modules into Root Modules by injecting the key configuration blocks that are necessary for OpenTofu/Terraform modules to act as Root Modules.

Modules on the Terraform Registry are primarily designed to be used as Shared Modules. That is, you won’t be able to git clone the underlying repository and run tofu init/terraform init or apply directly on the module without modification. Unless otherwise specified, almost all the modules will require composition with other modules/resources to deploy. When using modules in the registry, it helps to think about what blocks and resources are necessary to operate the module, and translating those into Terragrunt blocks that generate them.

Note that often, Terragrunt may not be able to deploy modules from the registry. While Terragrunt has features to turn any Shared Module into a Root Module, there are two key technical limitations that prevent Terragrunt from converting ALL shared modules:

• Every complex input must have a type associated with it. Otherwise, OpenTofu/Terraform will interpret the input that Terragrunt passes through as string. This includes list and map.
• Derived sensitive outputs must be marked as sensitive. Refer to the terraform tutorial on sensitive variables for more information on this requirement.

If you run into issues deploying a module from the registry, chances are that module is not a Shared Service Module, and thus not designed for use with Terragrunt. Depending on the technical limitation, Terragrunt may be able to support the transition to root module. Please always file an issue on the terragrunt repository with the module + error message you are encountering, instead of the module repository.

remote_state

The remote_state block is used to configure how Terragrunt will set up the remote state configuration of your OpenTofu/Terraform code. You can read more about Terragrunt’s remote state functionality in Keep your remote state configuration DRY use case overview.

The remote_state block supports the following arguments:

• backend (attribute): Specifies which remote state backend will be configured. This should be one of the available backends that Opentofu/Terraform supports.

• disable_init (attribute): When true, skip automatic initialization of the backend by Terragrunt. Some backends have support in Terragrunt to be automatically created if the storage does not exist. Currently, s3 and gcs are the two backends with support for automatic creation. Defaults to false.

• disable_dependency_optimization (attribute): When true, disable optimized dependency fetching for terragrunt modules using this remote_state block. See the documentation for dependency block for more details.

• generate (attribute): Configure Terragrunt to automatically generate a .tf file that configures the remote state backend. This is a map that expects two properties:

  • path: The path where the generated file should be written. If a relative path, it’ll be relative to the Terragrunt working dir (where the OpenTofu/Terraform code lives).

  • if_exists (attribute): What to do if a file already exists at path.

    Valid values are:

    • overwrite (overwrite the existing file)
    • overwrite_terragrunt (overwrite the existing file if it was generated by terragrunt; otherwise, error)
    • skip (skip code generation and leave the existing file as-is)
    • error (exit with an error)

• config (attribute): An arbitrary map that is used to fill in the backend configuration in OpenTofu/Terraform. All the properties will automatically be included in the OpenTofu/Terraform backend block (with a few exceptions: see below).

• encryption (attribute): A map that is used to configure state and plan encryption in OpenTofu. The properties will be transformed into an encryption block in the OpenTofu terraform block. The properties are specific to the respective key_provider (see below).

  For example, if you had the following remote_state block:

  terragrunt.hcl

  remote_state {
    backend = "s3"
    config = {
      bucket = "mybucket"
      key    = "path/to/my/key"
      region = "us-east-1"
    }
  }

  This is equivalent to the following OpenTofu/Terraform code:

  main.tf

  terraform {
    backend "s3" {
      bucket = "mybucket"
      key    = "path/to/my/key"
      region = "us-east-1"
    }
  }

Note that remote_state can also be set as an attribute. This is useful if you want to set remote_state dynamically. For example, if in common.hcl you had:

common.hcl

remote_state {
  backend = "s3"
  config = {
    bucket = "mybucket"
    key    = "path/to/my/key"
    region = "us-east-1"
  }
}

Then in a terragrunt.hcl file, you could dynamically set remote_state as an attribute as follows:

terragrunt.hcl

locals {
  # Load the data from common.hcl
  common = read_terragrunt_config(find_in_parent_folders("common.hcl"))
}

# Set the remote_state config dynamically to the remote_state config in common.hcl
remote_state = local.common.remote_state

backend

Note that Terragrunt does special processing of the config attribute for the s3 and gcs remote state backends, and supports additional keys that are used to configure the automatic initialization feature of Terragrunt.

For the s3 backend, the following additional properties are supported in the config attribute:

• region - (Optional) The region of the S3 bucket.
• profile - (Optional) This is the AWS profile name as set in the shared credentials file.
• endpoint - (Optional) A custom endpoint for the S3 API.
• endpoints: (Optional) A configuration map for custom service API (starting with Terraform 1.6).
  • s3 - (Optional) A custom endpoint for the S3 API. Overrides endpoint argument.
  • dynamodb - (Optional) A custom endpoint for the DynamoDB API. Overrides dynamodb_endpoint argument.
• encrypt - (Optional) Whether to enable server-side encryption of the state file. If disabled, a log warning will be issued in the console output to notify the user. If skip_bucket_ssencryption is enabled, the log will be written as a debug log.
• role_arn - (Optional) The role to be assumed.
• shared_credentials_file - (Optional) This is the path to the shared credentials file. If this is not set and a profile is specified, ~/.aws/credentials will be used.
• external_id - (Optional) The external ID to use when assuming the role.
• session_name - (Optional) The session name to use when assuming the role.
• dynamodb_table - (Optional) The name of a DynamoDB table to use for state locking and consistency. The table must have a primary key named LockID. If not present, locking will be disabled.
• use_lockfile - (Optional) When true, enables native S3 locking using S3 object conditional writes for state locking. This feature requires OpenTofu >= 1.10. Can be used simultaneously with dynamodb_table during migration (both locks must be acquired successfully), but typically used as a replacement for DynamoDB locking.
• skip_bucket_versioning: When true, the S3 bucket that is created to store the state will not be versioned.
• skip_bucket_ssencryption: When true, the S3 bucket that is created to store the state will not be configured with server-side encryption.
• skip_bucket_accesslogging: DEPRECATED If provided, will be ignored. A log warning will be issued in the console output to notify the user.
• skip_bucket_root_access: When true, the S3 bucket that is created will not be configured with bucket policies that allow access to the root AWS user.
• skip_bucket_enforced_tls: When true, the S3 bucket that is created will not be configured with a bucket policy that enforces access to the bucket via a TLS connection.
• skip_bucket_public_access_blocking: When true, the S3 bucket that is created will not have public access blocking enabled.
• disable_bucket_update: When true, disable update S3 bucket if not equal configured in config block
• enable_lock_table_ssencryption: When true, the synchronization lock table in DynamoDB used for remote state concurrent access will be configured with server-side encryption.
• s3_bucket_tags: A map of key value pairs to associate as tags on the created S3 bucket.
• dynamodb_table_tags: A map of key value pairs to associate as tags on the created DynamoDB remote state lock table.
• accesslogging_bucket_tags: A map of key value pairs to associate as tags on the created S3 bucket to store de access logs.
• disable_aws_client_checksums: When true, disable computing and checking checksums on the request and response, such as the CRC32 check for DynamoDB. See #1059 for issue where this is a useful workaround.
• accesslogging_bucket_name: (Optional) When provided as a valid string, create an S3 bucket with this name to store the access logs for the S3 bucket used to store OpenTofu/Terraform state. If not provided, or string is empty or invalid S3 bucket name, then server access logging for the S3 bucket storing the Opentofu/Terraform state will be disabled. Note: When access logging is enabled supported encryption for state bucket is only AES256. Reference: S3 server access logging
• accesslogging_target_object_partition_date_source: (Optional) When provided as a valid string, it configures the PartitionDateSource option. This option is part of the TargetObjectKeyFormat and PartitionedPrefix AWS configurations, allowing you to configure the log object key format for the access log files. Reference: Logging requests with server access logging.
• accesslogging_target_prefix: (Optional) When provided as a valid string, set the TargetPrefix for the access log objects in the S3 bucket used to store Opentofu/Terraform state. If set to emptystring, then TargetPrefix will be set to empty string. If attribute is not provided at all, then TargetPrefix will be set to default value TFStateLogs/. This attribute won’t take effect if the accesslogging_bucket_name attribute is not present.
• skip_accesslogging_bucket_acl: When set to true, the S3 bucket where access logs are stored will not be configured with bucket ACL.
• skip_accesslogging_bucket_enforced_tls: When set to true, the S3 bucket where access logs are stored will not be configured with a bucket policy that enforces access to the bucket via a TLS connection.
• skip_accesslogging_bucket_public_access_blocking: When set to true, the S3 bucket where access logs are stored will not have public access blocking enabled.
• skip_accesslogging_bucket_ssencryption: When set to true, the S3 bucket where access logs are stored will not be configured with server-side encryption.
• bucket_sse_algorithm: (Optional) The algorithm to use for server-side encryption of the state bucket. Defaults to aws:kms.
• bucket_sse_kms_key_id: (Optional) The KMS Key to use when the encryption algorithm is aws:kms. Defaults to the AWS Managed aws/s3 key.
• assume_role: (Optional) A configuration map to use when assuming a role (starting with Terraform 1.6 for Terraform). Override top level arguments
  • role_arn - (Required) The role to be assumed.
  • duration - (Optional) The duration the credentials will be valid.
  • external_id - (Optional) The external ID to use when assuming the role.
  • policy - (Optional) Policy JSON to further restrict the role.
  • policy_arns - (Optional) A list of policy ARNs to further restrict the role.
  • session_name - (Optional) The session name to use when assuming the role.
  • source_identity - (Optional) The source identity to use when assuming the role.
  • tags - (Optional) A map of key value pairs used as assume role session tags.
  • transitive_tag_keys - (Optional) A list of tag keys that to be passed.
• assume_role_with_web_identity - (Optional) A configuration map to use when assuming a role with a web identity token.
  • role_arn - (Required) The role to be assumed.
  • duration - (Optional) The duration the credentials will be valid.
  • policy - (Optional) Policy JSON to further restrict the role.
  • policy_arns - (Optional) A list of policy ARNs to further restrict the role.
  • session_name - (Optional) The session name to use when assuming the role.
  • web_identity_token - (Required) The web identity token to use when assuming the role.
  • web_identity_token_file - (Optional) The path to the file containing the web identity token to use when assuming the role.

For the gcs backend, the following additional properties are supported in the config attribute:

• skip_bucket_creation: When true, Terragrunt will skip the auto initialization routine for setting up the GCS bucket for use with remote state.
• skip_bucket_versioning: When true, the GCS bucket that is created to store the state will not be versioned.
• enable_bucket_policy_only: When true, the GCS bucket that is created to store the state will be configured to use uniform bucket-level access.
• project: The GCP project where the bucket will be created.
• location: The GCP location where the bucket will be created.
• gcs_bucket_labels: A map of key value pairs to associate as labels on the created GCS bucket.
• credentials: Local path to Google Cloud Platform account credentials in JSON format.
• access_token: A temporary [OAuth 2.0 access token] obtained from the Google Authorization server. Example with S3:

root.hcl

# Configure OpenTofu/Terraform state to be stored in S3, in the bucket "my-tofu-state" in us-east-1 under a key that is
# relative to included terragrunt config. For example, if you had the following folder structure:
#
# .
# ├── root.hcl
# └── child
#     ├── main.tf
#     └── terragrunt.hcl
#
# And the following is defined in the root terragrunt.hcl config that is included in the child, the state file for the
# child module will be stored at the key "child/tofu.tfstate".
#
# Note that since we are not using any of the skip args, this will automatically create the S3 bucket
# "my-tofu-state" and DynamoDB table "my-lock-table" if it does not already exist.
remote_state {
  backend = "s3"
  config = {
    bucket         = "my-tofu-state"
    key            = "${path_relative_to_include()}/tofu.tfstate"
    region         = "us-east-1"
    encrypt        = true
    dynamodb_table = "my-lock-table"
  }
}

child/terragrunt.hcl

include "root" {
  path   = find_in_parent_folders("root.hcl")
}

child/main.tf

terraform {
  backend "s3" {}
}

Example with GCS:

root.hcl

# Configure OpenTofu/Terraform state to be stored in GCS, in the bucket "my-tofu-state" in the "my-tofu" GCP project in
# the eu region under a key that is relative to included terragrunt config. This will also apply the labels
# "owner=terragrunt_test" and "name=tofu_state_storage" to the bucket if it is created by Terragrunt.
#
# For example, if you had the following folder structure:
#
# .
# ├── root.hcl
# └── child
#     ├── main.tf
#     └── terragrunt.hcl
#
# And the following is defined in the root terragrunt.hcl config that is included in the child, the state file for the
# child module will be stored at the key "child/tofu.tfstate".
#
# Note that since we are not using any of the skip args, this will automatically create the GCS bucket
# "my-tofu-state" if it does not already exist.
remote_state {
  backend = "gcs"

  config = {
    project  = "my-tofu"
    location = "eu"
    bucket   = "my-tofu-state"
    prefix   = "${path_relative_to_include()}/tofu.tfstate"

    gcs_bucket_labels = {
      owner = "terragrunt_test"
      name  = "tofu_state_storage"
    }
  }
}

child/terragrunt.hcl

include "root" {
  path   = find_in_parent_folders("root.hcl")
}

child/main.tf

terraform {
  backend "gcs" {}
}

Example with S3 using native S3 locking (OpenTofu >= 1.10):

# Configure OpenTofu/Terraform state to be stored in S3 with native S3 locking instead of DynamoDB.
# This uses S3 object conditional writes for state locking, which requires OpenTofu >= 1.10.
remote_state {
  backend = "s3"
  config = {
    bucket       = "my-tofu-state"
    key          = "${path_relative_to_include()}/tofu.tfstate"
    region       = "us-east-1"
    encrypt      = true
    use_lockfile = true
  }
}

Example with S3 using both DynamoDB and native S3 locking during migration (OpenTofu >= 1.10):

# Configure OpenTofu/Terraform state with dual locking during migration from DynamoDB to S3 native locking.
# Both locks must be successfully acquired before operations can proceed.
# After the migration period, remove dynamodb_table to use only S3 native locking.
# Note: This won't delete the DynamoDB table, it will just be unused.
# You can delete it manually after the migration period.
remote_state {
  backend = "s3"
  config = {
    bucket         = "my-tofu-state"
    key            = "${path_relative_to_include()}/tofu.tfstate"
    region         = "us-east-1"
    encrypt        = true
    dynamodb_table = "my-lock-table"  # Remove this after migration period
    use_lockfile   = true             # New native S3 locking
  }
}

encryption

The encryption map needs a key_provider property, which can be set to one of pbkdf2, aws_kms or gcp_kms.

Documentation for each provider type and its possible configuration can be found in the OpenTofu docs.

A terragrunt.hcl file configuring PBKDF2 encryption could look like this:

terragrunt.hcl

remote_state {
  backend = "s3"
  config = {
    bucket = "mybucket"
    key    = "path/to/my/key"
    region = "us-east-1"
  }

  encryption = {
    key_provider = "pbkdf2"
    passphrase   = get_env("PBKDF2_PASSPHRASE")
  }
}

This would result in the following OpenTofu code:

main.tf

terraform {
  backend "s3" {
    bucket = "mybucket"
    key    = "path/to/my/key"
    region = "us-east-1"
  }
  encryption {
    key_provider "pbkdf2" "default" {
      passphrase = "SUPERSECRETPASSPHRASE"
    }
    method "aes_gcm" "default" {
      keys = key_provider.pbkdf2.default
    }
    state {
      method = method.aes_gcm.default
    }
    plan {
      method = method.aes_gcm.default
    }
  }
}

include

The include block is used to specify inheritance of Terragrunt configuration files. The included config (also called the parent) will be merged with the current configuration (also called the child) before processing. You can learn more about the inheritance properties of Terragrunt in the Filling in remote state settings with Terragrunt section of the “Keep your remote state configuration DRY” use case overview.

You can have more than one include block, but each one must have a unique label. It is recommended to always label your include blocks. Bare includes (include block with no label - e.g., include {}) are currently supported for backward compatibility, but is deprecated usage and support may be removed in the future.

include blocks support the following arguments:

• name (label): You can define multiple include blocks in a single terragrunt config. Each include block must be labeled with a unique name to differentiate it from the other includes. e.g., if you had a block include "remote" {}, you can reference the relevant exposed data with the expression include.remote.
• path (attribute): Specifies the path to a Terragrunt configuration file (the parent config) that should be merged with this configuration (the child config).
• expose (attribute, optional): Specifies whether or not the included config should be parsed and exposed as a variable. When true, you can reference the data of the included config under the variable include. Defaults to false. Note that the include variable is a map of include labels to the parsed configuration value.
• merge_strategy (attribute, optional): Specifies how the included config should be merged. Valid values are: no_merge (do not merge the included config), shallow (do a shallow merge - default), deep (do a deep merge of the included config).

NOTE: At this time, Terragrunt only supports a single level of include blocks. That is, Terragrunt will error out if an included config also has an include block defined. If you are interested in this feature, please follow #1566 to be notified when nested include blocks are supported.

Special case for shallow merge: When performing a shallow merge, all attributes and blocks are merged shallowly with replacement, except for dependencies blocks (NOT dependency block). dependencies blocks are deep merged: that is, all the lists of paths from included configurations are concatenated together, rather than replaced in override fashion.

Examples:

Single include

root.hcl

# If you have the following folder structure, and the following contents for ./child/terragrunt.hcl, this will include
# and merge the configurations in the root.hcl file.
#
# .
# ├── root.hcl
# └── child
#     ├── main.tf
#     └── terragrunt.hcl
remote_state {
  backend = "s3"
  config = {
    bucket         = "my-tofu-state"
    key            = "${path_relative_to_include()}/tofu.tfstate"
    region         = "us-east-1"
    encrypt        = true
    dynamodb_table = "my-lock-table"
  }
}

child/terragrunt.hcl

include "root" {
  path   = find_in_parent_folders("root.hcl")
  expose = true
}

inputs = {
  remote_state_config = include.root.remote_state
}

child/main.tf

terraform {
  backend "s3" {}
}

Multiple includes

root.hcl

# If you have the following folder structure, and the following contents for ./child/terragrunt.hcl, this will include
# and merge the configurations in the root.hcl, while only loading the data in the region.hcl
# configuration.
#
# .
# ├── root.hcl
# ├── region.hcl
# └── child
#     └── terragrunt.hcl
remote_state {
  backend = "s3"
  config = {
    bucket         = "my-tofu-state"
    key            = "${path_relative_to_include()}/tofu.tfstate"
    region         = "us-east-1"
    encrypt        = true
    dynamodb_table = "my-lock-table"
  }
}

region.hcl

locals {
  region = "production"
}

child/terragrunt.hcl

include "remote_state" {
  path   = find_in_parent_folders("root.hcl")
  expose = true
}

include "region" {
  path           = find_in_parent_folders("region.hcl")
  expose         = true
  merge_strategy = "no_merge"
}

inputs = {
  remote_state_config = include.remote_state.remote_state
  region              = include.region.locals.region
}

child/main.tf

terraform {
  backend "s3" {}
}

Limitations on accessing exposed config

In general, you can access all attributes on include when they are exposed (e.g., include.locals, include.inputs, etc.).

However, to support run --all, Terragrunt is unable to expose all attributes when the included config has a dependency block. To understand this, consider the following example:

root.hcl

dependency "vpc" {
  config_path = "${get_terragrunt_dir()}/../vpc"
}

inputs = {
  vpc_name = dependency.vpc.outputs.name
}

terragrunt.hcl

include "root" {
  path   = find_in_parent_folders("root.hcl")
  expose = true
}

dependency "alb" {
  config_path = (
    include.root.inputs.vpc_name == "mgmt"
    ? "../alb-public"
    : "../alb-private"
  )
}

inputs = {
  alb_id = dependency.alb.outputs.id
}

In the child terragrunt.hcl, the dependency path for the alb depends on whether the VPC is the mgmt VPC or not, which is determined by the dependency.vpc in the root config. This means that the output from dependency.vpc must be available to parse the dependency.alb config.

This causes problems when performing a run --all apply operation. During a run --all operation, Terragrunt first parses all the dependency blocks to build a dependency tree of the Terragrunt modules to figure out the order of operations. If all the paths are static references, then Terragrunt can determine all the dependency paths before any module has been applied. In this case there is no problem even if other config blocks access dependency, as by the time Terragrunt needs to parse those blocks, the upstream dependencies would have been applied during the run --all apply.

However, if those dependency blocks depend on upstream dependencies, then there is a problem as Terragrunt would not be able to build the dependency tree without the upstream dependencies being applied.

Therefore, to ensure that Terragrunt can build the dependency tree in a run --all operation, Terragrunt enforces the following limitation to exposed include config:

If the included configuration has any dependency blocks, only locals and include are exposed and available to the child include and dependency blocks. There are no restrictions for other blocks in the child config (e.g., you can reference inputs from the included config in child inputs).

Otherwise, if the included config has no dependency blocks, there is no restriction on which exposed attributes you can access.

For example, the following alternative configuration is valid even if the alb dependency is still accessing the inputs attribute from the included config:

root.hcl

inputs = {
  vpc_name = "mgmt"
}

terragrunt.hcl

include "root" {
  path   = find_in_parent_folders("root.hcl")
  expose = true
}

dependency "vpc" {
  config_path = "../vpc"
}

dependency "alb" {
  config_path = (
    include.root.inputs.vpc_name == "mgmt"
    ? "../alb-public"
    : "../alb-private"
  )
}

inputs = {
  vpc_name = dependency.vpc.outputs.name
  alb_id   = dependency.alb.outputs.id
}

What is deep merge?

When the merge_strategy for the include block is set to deep, Terragrunt will perform a deep merge of the included config. For Terragrunt config, deep merge is defined as follows:

• For simple types, the child overrides the parent.
• For lists, the two attribute lists are combined together in concatenation.
• For maps, the two maps are combined together recursively. That is, if the map keys overlap, then a deep merge is performed on the map value.
• For blocks, if the label is the same, the two blocks are combined together recursively. Otherwise, the blocks are appended like a list. This is similar to maps, with block labels treated as keys.

However, due to internal implementation details, some blocks are not deep mergeable. This will change in the future, but for now, terragrunt performs a shallow merge (that is, block definitions in the child completely override the parent definition). The following blocks have this limitation: - remote_state - generate

Similarly, the locals block is deliberately omitted from the merge operation by design. That is, you will not be able to access parent config locals in the child config, and vice versa in a merge. However, you can access the parent locals in child config if you use the expose feature.

Finally, dependency blocks have special treatment. When doing a deep merge, dependency blocks from both child and parent config are accessible in both places. For example, consider the following setup:

root.hcl

dependency "vpc" {
  config_path = "../vpc"
}

inputs = {
  vpc_id = dependency.vpc.outputs.vpc_id
  db_id  = dependency.mysql.outputs.db_id
}

terragrunt.hcl

include "root" {
  path           = find_in_parent_folders("root.hcl")
  merge_strategy = "deep"
}

dependency "mysql" {
  config_path = "../mysql"
}

inputs = {
  security_group_id = dependency.vpc.outputs.security_group_id
}

In the example, note how the parent is accessing the outputs of the mysql dependency even though it is not defined in the parent. Similarly, the child is accessing the outputs of the vpc dependency even though it is not defined in the child.

Full example:

root.hcl

remote_state {
  backend = "s3"
  config = {
    encrypt = true
    bucket = "__FILL_IN_BUCKET_NAME__"
    key = "${path_relative_to_include()}/tofu.tfstate"
    region = "us-west-2"
  }
}

dependency "vpc" {
  # This will get overridden by child terragrunt.hcl configs
  config_path = ""

  mock_outputs = {
    attribute     = "hello"
    old_attribute = "old val"
    list_attr     = ["hello"]
    map_attr = {
      foo = "bar"
    }
  }
  mock_outputs_allowed_terraform_commands = ["apply", "plan", "destroy", "output"]
}

inputs = {
  attribute     = "hello"
  old_attribute = "old val"
  list_attr     = ["hello"]
  map_attr = {
    foo = "bar"
    test = dependency.vpc.outputs.new_attribute
  }
}

terragrunt.hcl

include "root" {
  path           = find_in_parent_folders("root.hcl")
  merge_strategy = "deep"
}

remote_state {
  backend = "local"
}

dependency "vpc" {
  config_path = "../vpc"
  mock_outputs = {
    attribute     = "mock"
    new_attribute = "new val"
    list_attr     = ["mock"]
    map_attr = {
      bar = "baz"
    }
  }
}

inputs = {
  attribute     = "mock"
  new_attribute = "new val"
  list_attr     = ["mock"]
  map_attr = {
    bar = "baz"
  }

  dep_out = dependency.vpc.outputs
}

# Merged terragrunt.hcl

# Child override parent completely due to deep merge limitation
remote_state {
  backend = "local"
}

# mock_outputs are merged together with deep merge
dependency "vpc" {
  config_path = "../vpc"       # Child overrides parent
  mock_outputs = {
    attribute     = "mock"     # Child overrides parent
    old_attribute = "old val"  # From parent
    new_attribute = "new val"  # From child
    list_attr     = [
      "hello",                 # From parent
      "mock",                  # From child
    ]
    map_attr = {
      foo = "bar"              # From parent
      bar = "baz"              # From child
    }
  }

  # From parent
  mock_outputs_allowed_terraform_commands = ["apply", "plan", "destroy", "output"]
}

# inputs are merged together with deep merge
inputs = {
  attribute     = "mock"       # Child overrides parent
  old_attribute = "old val"    # From parent
  new_attribute = "new val"    # From child
  list_attr     = [
    "hello",                 # From parent
    "mock",                  # From child
  ]
  map_attr = {
    foo = "bar"                                   # From parent
    bar = "baz"                                   # From child
    test = dependency.vpc.outputs.new_attribute   # From parent, referencing dependency mock output from child
  }

  dep_out = dependency.vpc.outputs                # From child
}

locals

The locals block is used to define aliases for Terragrunt expressions that can be referenced elsewhere in configuration.

The locals block does not have a defined set of arguments that are supported. Instead, all the arguments passed into locals are available under the reference local.<local name> throughout the file where the locals block is defined.

Example:

terragrunt.hcl

# Make the AWS region a reusable variable within the configuration
locals {
  aws_region = "us-east-1"
}

inputs = {
  region = local.aws_region
  name   = "${local.aws_region}-bucket"
}

Complex locals

Some local variables can be complex types, such as list or map.

For example:

terragrunt.hcl

locals {
  # Define a list of regions
  regions = ["us-east-1", "us-west-2", "eu-west-1"]

  # Define a map of regions to their corresponding bucket names
  region_to_bucket_name = {
    us-east-1 = "east-bucket"
    us-west-2 = "west-bucket"
    eu-west-1 = "eu-bucket"
  }

  # The first region is accessed like this
  first_region = local.regions[0]

  # The bucket name for us-east-1 is accessed like this
  us_east_1_bucket = local.region_to_bucket_name["us-east-1"]
}

These complex types can also arise when using values derived from reading other files.

For example:

region.hcl

locals {
  region = "us-east-1"
}

unit/terragrunt.hcl

locals {
  # Load the data from region.hcl
  region_hcl = read_terragrunt_config(find_in_parent_folders("region.hcl"))

  # Access the region from the loaded file
  region = local.region_hcl.locals.region
}

inputs = {
  bucket_name = "${local.region}-bucket"
}

Similarly, you might want to define this shared data using other serialization formats, like JSON or YAML:

region.yml

region: us-east-1

unit/terragrunt.hcl

locals {
  # Load the data from region.json
  region_yml = yamldecode(file(find_in_parent_folders("region.yml")))

  # Access the region from the loaded file
  region = local.region_yml.region
}

inputs = {
  bucket_name = "${local.region}-bucket"
}

Computed locals

When reading Terragrunt HCL configurations, you might read in a computed configuration:

computed.hcl

locals {
  computed_value = run_cmd("--terragrunt-quiet", "python3", "-c", "print('Hello,')")
}

unit/terragrunt.hcl

locals {
  # Load the data from computed.hcl
  computed = read_terragrunt_config(find_in_parent_folders("computed.hcl"))

  # Access the computed value from the loaded file
  computed_value = "${local.computed.locals.computed_value} world!" # <-- This will be "Hello, world!"
}

Note that this can be a powerful feature, but it can easily lead to performance issues if you are not careful, as each read will require a full parse of the HCL file and potentially execute expensive computation.

Use this feature judiciously.

dependency

The dependency block is used to configure module dependencies. Each dependency block exports the outputs of the target module as block attributes you can reference throughout the configuration. You can learn more about dependency blocks in the Dependencies between modules section of the “Execute Opentofu/Terraform commands on multiple modules at once” use case overview.

You can define more than one dependency block. Each label you provide to the block identifies another dependency that you can reference in your config.

The dependency block supports the following arguments:

• name (label): You can define multiple dependency blocks in a single terragrunt config. As such, each block needs a name to differentiate between the other blocks, which is what the first label of the block is used for. You can reference the specific dependency output by the name. E.g if you had a block dependency "vpc", you can reference the outputs and inputs of this dependency with the expressions dependency.vpc.outputs and dependency.vpc.inputs.
• config_path (attribute): Path to a Terragrunt module (folder with a terragrunt.hcl file) that should be included as a dependency in this configuration.
• enabled (attribute): When false, excludes the dependency from execution. Defaults to true.
• skip_outputs (attribute): When true, skip calling terragrunt output when processing this dependency. If mock_outputs is configured, set outputs to the value of mock_outputs. Otherwise, outputs will be set to an empty map. Put another way, setting skip_outputs means “use mocks all the time if mock_outputs are set.”
• mock_outputs (attribute): A map of arbitrary key value pairs to use as the outputs attribute when no outputs are available from the target module, or if skip_outputs is true. However, it’s generally recommended not to set skip_outputs if using mock_outputs, because skip_outputs means “use mocks all the time if they are set” whereas mock_outputs means “use mocks only if real outputs are not available.” Use locals instead when skip_outputs = true.
• mock_outputs_allowed_terraform_commands (attribute): A list of Terraform commands for which mock_outputs are allowed. If a command is used where mock_outputs is not allowed, and no outputs are available in the target module, Terragrunt will throw an error when processing this dependency.
• mock_outputs_merge_with_state (attribute): DEPRECATED. Use mock_outputs_merge_strategy_with_state. When true, mock_outputs and the state outputs will be merged. That is, the mock_outputs will be treated as defaults and the real state outputs will overwrite them if the keys clash.
• mock_outputs_merge_strategy_with_state (attribute): Specifies how any existing state should be merged into the mocks. Valid values are
  • no_merge (default) - any existing state will be used as is. If the dependency does not have an existing state (it hasn’t been applied yet), then the mocks will be used
  • shallow - the existing state will be shallow merged into the mocks. Mocks will only be used where the output does not already exist in the dependency’s state
  • deep_map_only - the existing state will be deeply merged into the mocks. If an output is a map, the mock key will be used where that key does not exist in the state. Lists will not be merged

Example:

terragrunt.hcl

# Run `terragrunt output` on the module at the relative path `../vpc` and expose them under the attribute
# `dependency.vpc.outputs`
dependency "vpc" {
  config_path = "../vpc"

  # Configure mock outputs for the `validate` command that are returned when there are no outputs available (e.g the
  # module hasn't been applied yet.
  mock_outputs_allowed_terraform_commands = ["validate"]
  mock_outputs = {
    vpc_id = "fake-vpc-id"
  }
}

# Another dependency, available under the attribute `dependency.rds.outputs`
dependency "rds" {
  config_path = "../rds"
}

inputs = {
  vpc_id = dependency.vpc.outputs.vpc_id
  db_url = dependency.rds.outputs.db_url
}

IMPORTANT: The dependency.<name>.inputs field has been deprecated and removed. You can only access dependency outputs via dependency.<name>.outputs. If you were previously using dependency.<name>.inputs, you should refactor your configuration to use dependency.<name>.outputs instead.

Can I speed up dependency fetching?

dependency blocks are fetched in parallel at each source level, but will serially parse each recursive dependency. For example, consider the following chain of dependencies:

account --> vpc --> securitygroup --> ecs
                                      ^
                                     /
                              ecr --

In this chain, the ecr and securitygroup module outputs will be fetched concurrently when applying the ecs module, but the outputs for account and vpc will be fetched serially as terragrunt needs to recursively walk through the tree to retrieve the outputs at each level.

This recursive parsing happens due to the necessity to parse the entire terragrunt.hcl configuration (including dependency blocks) in full before being able to call tofu output/terraform output.

However, terragrunt includes an optimization to only fetch the lowest level outputs (securitygroup and ecr in this example) provided that the following conditions are met in the immediate dependencies:

• The remote state is managed using remote_state blocks.
• The dependency optimization feature flag is enabled (disable_dependency_optimization = false, which is the default).
• The remote_state block itself does not depend on any dependency outputs (locals and include are ok).
• You are not relying on before_hook, after_hook, or extra_arguments to the tofu init/terraform init call. NOTE: terragrunt will not automatically detect this and you will need to explicitly opt out of the dependency optimization flag.

If these conditions are met, terragrunt will only parse out the remote_state blocks and use that to pull down the state for the target module without parsing the dependency blocks, avoiding the recursive dependency retrieval.

dependencies

The dependencies block is used to enumerate all the Terragrunt modules that need to be applied in order for this module to be able to apply. Note that this is purely for ordering the operations when using run --all commands of OpenTofu/Terraform. This does not expose or pull in the outputs like dependency blocks.

The dependencies block supports the following arguments:

• paths (attribute): A list of paths to modules that should be marked as a dependency.

Example:

terragrunt.hcl

# When applying this terragrunt config in an `run --all` command, make sure the modules at "../vpc" and "../rds" are
# handled first.
dependencies {
  paths = ["../vpc", "../rds"]
}

generate

The generate block can be used to arbitrarily generate a file in the terragrunt working directory (where tofu/terraform is called). This can be used to generate common OpenTofu/Terraform configurations that are shared across multiple OpenTofu/Terraform modules. For example, you can use generate to generate the provider blocks in a consistent fashion by defining a generate block in the parent terragrunt config.

The generate block supports the following arguments:

• name (label): You can define multiple generate blocks in a single terragrunt config. As such, each block needs a name to differentiate between the other blocks.

• path (attribute): The path where the generated file should be written. If a relative path, it’ll be relative to the Terragrunt working dir (where the OpenTofu/Terraform code lives).

• if_exists (attribute): What to do if a file already exists at path.

  Valid values are:

  • overwrite (overwrite the existing file)
  • overwrite_terragrunt (overwrite the existing file if it was generated by terragrunt; otherwise, error)
  • skip (skip code generation and leave the existing file as-is)
  • error (exit with an error)

• if_disabled (attribute): What to do if a file already exists at path and disable is set to true (skip by default)

  Valid values are:

  • remove (remove the existing file)
  • remove_terragrunt (remove the existing file if it was generated by terragrunt; otherwise, error)
  • skip (skip removing and leave the existing file as-is).

• comment_prefix (attribute): A prefix that can be used to indicate comments in the generated file. This is used by terragrunt to write out a signature for knowing which files were generated by terragrunt. Defaults to #. Optional.

• disable_signature (attribute): When true, disables including a signature in the generated file. This means that there will be no difference between overwrite_terragrunt and overwrite for the if_exists setting. Defaults to false. Optional.

• contents (attribute): The contents of the generated file.

• disable (attribute): Disables this generate block.

Example:

terragrunt.hcl

# When using this terragrunt config, terragrunt will generate the file "provider.tf" with the aws provider block before
# calling to OpenTofu/Terraform. Note that this will overwrite the `provider.tf` file if it already exists.
generate "provider" {
  path      = "provider.tf"
  if_exists = "overwrite"
  contents = <<EOF
provider "aws" {
  region              = "us-east-1"
  version             = "= 2.3.1"
  allowed_account_ids = ["1234567890"]
}
EOF
}

Note that generate can also be set as an attribute. This is useful if you want to set generate dynamically. For example, if in common.hcl you had:

common.hcl

generate "provider" {
  path      = "provider.tf"
  if_exists = "overwrite"
  contents = <<EOF
provider "aws" {
  region              = "us-east-1"
  version             = "= 2.3.1"
  allowed_account_ids = ["1234567890"]
}
EOF
}

Then in a terragrunt.hcl file, you could dynamically set generate as an attribute as follows:

terragrunt.hcl

locals {
  # Load the data from common.hcl
  common = read_terragrunt_config(find_in_parent_folders("common.hcl"))
}

# Set the generate config dynamically to the generate config in common.hcl
generate = local.common.generate

engine

The engine block is used to configure experimental Terragrunt engine configuration. More details in engine section.

feature

The feature block is used to configure feature flags in HCL for a specific Terragrunt Unit.

Each feature flag must include a default value.

Feature flags can be overridden via the --feature CLI option.

terragrunt.hcl

feature "string_flag" {
  default = "test"
}

feature "run_hook" {
  default = false
}

terraform {
  before_hook "feature_flag" {
    commands = ["apply", "plan", "destroy"]
    execute  = feature.run_hook.value ? ["sh", "-c", "feature_flag_script.sh"] : [ "sh", "-c", "exit", "0" ]
  }
}

inputs = {
  string_feature_flag = feature.string_flag.value
}

Setting feature flags through CLI:

terragrunt --feature run_hook=true apply

terragrunt --feature run_hook=true --feature string_flag=dev apply

Setting feature flags through env variables:

export TG_FEATURE=run_hook=true
terragrunt apply

export TG_FEATURE=run_hook=true,string_flag=dev
terragrunt apply

Note that the default value of the feature block is evaluated as an expression dynamically.

What this means is that the value of the flag can be set via a Terragrunt expression at runtime. This is useful for scenarios where you want to integrate with external feature flag services like LaunchDarkly, AppConfig, etc.

terragrunt.hcl

feature "feature_name" {
  default = run_cmd("--terragrunt-quiet", "<command-to-fetch-feature-flag-value>")
}

Feature flags are used to conditionally control Terragrunt behavior at runtime, including the inclusion or exclusion of units. More on that in the exclude block.

exclude

The exclude block in Terragrunt provides advanced configuration options to dynamically determine when and how specific units in the Terragrunt dependency graph are excluded. This feature allows for fine-grained control over which actions are executed and can conditionally exclude dependencies.

Syntax:

terragrunt.hcl

exclude {
    if                   = <boolean>         # Boolean to determine exclusion.
    no_run               = <boolean>         # Boolean to prevent the unit from running (even when not using `--all`).
    actions              = ["<action>", ...] # List of actions to exclude (e.g., "plan", "apply", "all", "all_except_output").
    exclude_dependencies = <boolean>         # Boolean to determine if dependencies should also be excluded.
}

Attributes:

Attribute	Type	Description
if	boolean	Condition to dynamically determine whether the unit should be excluded.
actions	list(string)	Specifies which actions to exclude when the condition is met. Options: plan, apply, all, all_except_output etc.
exclude_dependencies	boolean	Indicates whether the dependencies of the excluded unit should also be excluded (default: false).
no_run	boolean	When true and if is true, prevents the unit from running entirely for both single unit commands and run --all commands, but only when the current action matches the actions list.

Examples:

terragrunt.hcl

exclude {
    if = feature.feature_name.value # Dynamically exclude based on a feature flag.
    actions = ["plan", "apply"]     # Exclude `plan` and `apply` actions.
    exclude_dependencies = false    # Do not exclude dependencies.
}

In this example, the unit is excluded for the plan and apply actions only when feature.feature_name.value evaluates to true. Dependencies are not excluded.

terragrunt.hcl

exclude {
    if = feature.is_dev_environment.value # Exclude only for development environments.
    actions = ["all"]                     # Exclude all actions.
    exclude_dependencies = true           # Exclude dependencies along with the unit.
}

This configuration ensures the unit and its dependencies are excluded from all actions in the Terragrunt graph when the feature is_dev_environment evaluates to true.

terragrunt.hcl

exclude {
    if = true                       # Explicitly exclude.
    actions = ["all_except_output"] # Allow `output` actions nonetheless.
    exclude_dependencies = false    # Dependencies remain active.
}

This setup is useful for scenarios where output evaluation is still needed, even if other actions like plan or apply are excluded.

terragrunt.hcl

exclude {
    if      = true
    no_run  = true
    actions = ["plan"]
}

This configuration prevents the unit from running when if is true AND the current action is “plan”. The no_run attribute works for both single unit commands (like terragrunt plan) and run --all commands (like terragrunt run --all plan), but only when the current action matches the actions list.

Consider using this for units that are expensive to continuously update, and can be opted in when necessary.

errors

The errors block contains all the configurations for handling errors.

It supports different nested configuration blocks like retry and ignore to define specific error-handling strategies.

Retry Configuration

The retry block within the errors block defines rules for retrying operations when specific errors occur. This is useful for handling intermittent errors that may resolve after a short delay or multiple attempts.

Example: Retry Configuration

terragrunt.hcl

errors {
    retry "retry_example" {
        retryable_errors = [".*Error: transient.*"] # Matches errors containing 'Error: transient'
        max_attempts = 5                           # Retry up to 5 times
        sleep_interval_sec = 10                    # Wait 10 seconds between retries
    }
}

Parameters:

• retryable_errors: A list of regex patterns to match errors that are eligible to be retried.

  e.g. ".*Error: transient.*" matches errors containing Error: transient.

• max_attempts: The maximum number of retry attempts.

  e.g. 5 retries.

• sleep_interval_sec: Time (in seconds) to wait between retries.

  e.g. 10 seconds.

Ignore Configuration

The ignore block within the errors block defines rules for ignoring specific errors. This is useful when certain errors are known to be safe and should not prevent the run from proceeding.

Example: Ignore Configuration

terragrunt.hcl

errors {
    ignore "ignore_example" {
        ignorable_errors = [
            ".*Error: safe-to-ignore.*", # Ignore errors containing 'Error: safe-to-ignore'
            "!.*Error: critical.*"      # Do not ignore errors containing 'Error: critical'
        ]
        message = "Ignoring safe-to-ignore errors" # Optional message displayed when ignoring errors
        signals = {
            safe_to_revert = true # Indicates the operation is safe to revert on failure
        }
    }
}

Parameters:

• ignorable_errors: A list of regex patterns to define errors to ignore.
  • "Error: safe-to-ignore.*": Ignores errors containing Error: safe-to-ignore.
  • "!Error: critical.*": Ensures errors containing Error: critical are not ignored.
• message (Optional): A warning message displayed when an error is ignored.
  • Example: "Ignoring safe-to-ignore errors".
• signals (Optional): Key-value pairs used to emit signals to external systems.
  • Example: safe_to_revert = true indicates it is safe to revert the operation if it fails.

Populating values into the signals attribute results in a JSON file named error-signals.json being emitted on failure. This file can be inspected in CI/CD systems to determine the recommended course of action to address the failure.

Example:

If an error occurs and the author of the unit has signaled safe_to_revert = true, the CI/CD system could follow a standard process:

• Identify all units with files named error-signals.json.
• Checkout the previous commit for those units.
• Apply the units in their previous state, effectively reverting their updates.

This approach ensures consistent and automated error handling in complex pipelines.

Combined Example

Below is a combined example showcasing both retry and ignore configurations within the errors block.

terragrunt.hcl

errors {
    # Retry block for transient errors
    retry "transient_errors" {
        retryable_errors = [".*Error: transient network issue.*"]
        max_attempts = 3
        sleep_interval_sec = 5
    }

    # Ignore block for known safe-to-ignore errors
    ignore "known_safe_errors" {
        ignorable_errors = [
            ".*Error: safe warning.*",
            "!.*Error: do not ignore.*"
        ]
        message = "Ignoring safe warning errors"
        signals = {
            alert_team = false
        }
    }
}

Take note that:

• All retry and ignore configurations must be defined within a single errors block.
• Conditional logic can be used within ignorable_errors to enable or disable rules dynamically.

Evaluation Order:

• Ignore Rules: Errors are checked against the ignore rules first. If an error matches, it is ignored and will not trigger a retry.

• Retry Rules: Once ignore rules are applied, the retry rules handle any remaining errors.

Note: Only the first matching rule is applied. If there are multiple conflicting rules, any matches after the first one are ignored.

Errors during source fetching

In addition to handling errors during OpenTofu/Terraform runs, the errors block will also handle errors that occur during source fetching.

This can be particularly useful when fetching from artifact repositories that may be temporarily unavailable.

Example:

terragrunt.hcl

terraform {
  source = "https://unreliable-source.com/module.zip"
}

errors {
    retry "source_fetch" {
        retryable_errors = [".*Error: transient network issue.*"]
        max_attempts = 3
        sleep_interval_sec = 5
    }
}

unit

The unit block is used to define a deployment unit within a Terragrunt stack file (terragrunt.stack.hcl). Each unit represents a distinct infrastructure component that should be deployed as part of the stack.

Purpose: Define a single, deployable piece of infrastructure. Use case: When you want to create a single piece of isolated infrastructure (e.g. a specific VPC, database, or application). Result: Generates a single terragrunt.hcl file in the specified path.

The unit block supports the following arguments:

• name (label): A unique identifier for the unit. This is used to reference the unit elsewhere in your configuration.
• source (attribute): Specifies where to find the Terragrunt configuration files for this unit. This follows the same syntax as the source parameter in the terraform block.
• path (attribute): The relative path where this unit should be deployed within the stack directory (.terragrunt-stack). Also take note of the no_dot_terragrunt_stack attribute below, which can impact this.
• values (attribute, optional): A map of values that will be passed to the unit as inputs.
• no_dot_terragrunt_stack (attribute, optional): A boolean flag (true or false). When set to true, the unit will not be placed inside the .terragrunt-stack directory but will instead be generated in the same directory where terragrunt.stack.hcl is located. This allows for a soft adoption of stacks, making it easier for users to start using terragrunt.stack.hcl without modifying existing directory structures, or performing state migrations.
• no_validation (attribute, optional): A boolean flag (true or false) that controls whether Terragrunt should validate the unit’s configuration. When set to true, Terragrunt will skip validation checks for this unit.

Example:

terragrunt.stack.hcl

unit "vpc" {
  source = "git::git@github.com:acme/infrastructure-units.git//networking/vpc?ref=v0.0.1"
  path   = "vpc"
  values = {
    vpc_name = "main"
    cidr     = "10.0.0.0/16"
  }
}

Note that each unit must have a unique name and path within the stack.

When values are specified, generated units will have access to those values via a special terragrunt.values.hcl file generated next to the terragrunt.hcl file of the unit.

• terragrunt.stack.hcl
• Directory.terragrunt-stack

  • Directoryvpc

    • terragrunt.values.hcl
    • terragrunt.hcl

The terragrunt.values.hcl file will contain the values specified in the values block as top-level attributes:

.terragrunt-stack/vpc/terragrunt.values.hcl

vpc_name = "main"
cidr     = "10.0.0.0/16"

The unit will be able to leverage those values via values variables.

.terragrunt-stack/vpc/terragrunt.hcl

inputs = {
  vpc_name = values.vpc_name
  cidr     = values.cidr
}

Example usage of no_dot_terragrunt_stack attribute:

terragrunt.stack.hcl

unit "vpc" {
  source = "git::git@github.com:acme/infrastructure-units.git//networking/vpc?ref=v0.0.1"
  path   = "vpc"
  values = {
    vpc_name = "main"
    cidr     = "10.0.0.0/16"
  }
}

unit "rds" {
  source = "git::git@github.com:acme/infrastructure-units.git//database/rds?ref=v0.0.1"
  path   = "rds"
  values = {
    engine   = "postgres"
    version  = "13"
  }
  no_dot_terragrunt_stack = true
}

With the above configuration, the resulting directory structure will be:

• terragrunt.stack.hcl
• Directory.terragrunt-stack

  • Directoryvpc

    • terragrunt.values.hcl
    • terragrunt.hcl
• Directoryrds

  • terragrunt.values.hcl
  • terragrunt.hcl

The vpc unit is placed inside .terragrunt-stack, as expected. The rds unit is generated in the same directory as terragrunt.stack.hcl, rather than inside .terragrunt-stack, due to no_dot_terragrunt_stack = true.

Notes:

• The source value can be updated dynamically using the --source-map flag, just like terraform.source.

• A pre-created terragrunt.values.hcl file can be provided in the unit source (sibling to the terragrunt.hcl file used as the source of the unit). If present, this file will be used as the default values for the unit. However, if the values attribute is defined in the unit block, the generated terragrunt.values.hcl will replace the pre-existing file.

Comparison: unit vs stack blocks

Aspect	unit block	stack block
Purpose	Define a single infrastructure component	Define a reusable collection of components
When to use	For specific, one-off infrastructure pieces	For patterns of infrastructure pieces that you want provisioned together
Generated output	A directory with a single terragrunt.hcl file	A directory with a terragrunt.stack.hcl file

stack

The stack block is used to define a stack of deployment units in a Terragrunt configuration file (terragrunt.stack.hcl). Stacks allow for nesting, enabling the organization of infrastructure components into modular, reusable groups, reducing redundancy and improving maintainability.

Purpose: Define a collection of related units that can be reused. Use case: When you have a common, multi-unit pattern (like “dev environment” or “three-tier web application”) that you want to deploy multiple times. Result: Generates another terragrunt.stack.hcl file that can contain more units or stacks.

Stacks are designed to be nestable, helping to mitigate the risk of stacks becoming too large or too repetitive. When a stack is generated, it can include nested stacks, ensuring that the configuration scales efficiently.

The stack block supports the following arguments:

• name (label): A unique identifier for the stack. This is used to reference the stack elsewhere in your configuration.
• source (attribute): Specifies where to find the Terragrunt configuration files for this stack. This follows the same syntax as the source parameter in the terraform block.
• path (attribute): The relative path within .terragrunt-stack where this stack should be generated.If an absolute path is provided here, Terragrunt will generate the stack in that location, instead of generating it in a path relative to the .terragrunt-stack directory. Also take note of the no_dot_terragrunt_stack attribute below, which can impact this.
• values (attribute, optional): A map of custom values that can be passed to the stack. These values can be referenced within the stack’s configuration files, allowing for customization without modifying the stack source.
• no_dot_terragrunt_stack (attribute, optional): A boolean flag (true or false). When set to true, the stack will not be placed inside the .terragrunt-stack directory but will instead be generated in the same directory where terragrunt.stack.hcl is located. This allows for a soft adoption of stacks, making it easier for users to start using terragrunt.stack.hcl without modifying existing directory structures, or performing state migrations.
• no_validation (attribute, optional): A boolean flag (true or false) that controls whether Terragrunt should validate the stack’s configuration. When set to true, Terragrunt will skip validation checks for this stack.

Example:

terragrunt.stack.hcl

stack "services" {
    source = "github.com/gruntwork-io/terragrunt-stacks//stacks/mock/services?ref=v0.0.1"
    path   = "services"
    values = {
        project = "dev-services"
        cidr    = "10.0.0.0/16"
    }
}

github.com/gruntwork-io/terragrunt-stacks//stacks/mock/services/terragrunt.stack.hcl

# ...
unit "vpc" {
  # ...
  values = {
    cidr = values.cidr
  }
}

In this example, the services stack is defined with path services, which will be generated at .terragrunt-stack/services. The stack is also provided with custom values for project and cidr, which can be used within the stack’s configuration files. Terragrunt will recursively generate a stack using the contents of the .terragrunt-stack/services/terragrunt.stack.hcl file until the entire stack is fully generated.

Notes:

• The source value can be updated dynamically using the --source-map flag, just like terraform.source.

• A pre-created terragrunt.values.hcl file can be provided in the stack source (sibling to the terragrunt.stack.hcl file used as the source of the stack). If present, this file will be used as the default values for the stack. However, if the values attribute is defined in the stack block, the generated terragrunt.values.hcl will replace the pre-existing file.