The Terragrunt configuration file uses the same HCL syntax as Terraform itself in terragrunt.hcl.
Terragrunt also supports JSON-serialized HCL in a terragrunt.hcl.json file:
where terragrunt.hcl is mentioned you can always use terragrunt.hcl.json instead.
The following is a reference of all the supported blocks and attributes in the configuration file:
The terraform block is used to configure how Terragrunt will interact with Terraform. This includes specifying where
to find the Terraform configuration files, any extra arguments to pass to the terraform CLI, and any hooks to run
before or after calling Terraform.
The terraform block supports the following arguments:
source (attribute): Specifies where to find Terraform configuration files. This parameter supports the exact same syntax as the
module source parameter for 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.
source parameter can be configured to pull 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.TG_TF_REGISTRY_TOKEN.
This token can be any registry API token.tfr protocol supports a shorthand notation where the REGISTRY_HOST can be omitted to default to the public
registry (registry.terraform.io) 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.//. 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.extra_arguments (block): Nested blocks used to specify extra CLI arguments to pass to the terraform CLI. 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 terraform.commands (required) : A list of 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 terraform.required_var_files (optional): A list of file paths to 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 terraform vars files (.tfvars) that will be passed in to
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 terraform is called.
Hooks run from the directory with the terraform module, except for hooks related to terragrunt-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 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 prior to running the hook command. Defaults to the terragrunt configuration directory for
terragrunt-read-config and init-from-module hooks, and the 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 Terraform command hit an error. Default is false.after_hook (block): Nested blocks used to specify command hooks that should be run after terraform is called.
Hooks run from the terragrunt configuration directory (the directory where terragrunt.hcl lives). Supports the same
arguments as before_hook.In addition to supporting before and after hooks for all terraform commands, the following specialized hooks are also supported:
terragrunt-read-config (after hook only): terragrunt-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 terragrunt-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
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 terraform init on the cloned module.
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.init-from-module will run in the terragrunt config directory,
while the working directory for hooks associated with init will be the terraform module.Example:
terraform {
# Pull the 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 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 = ["${get_parent_terragrunt_dir()}/common.tfvars"]
}
# The following are examples of how to specify hooks
# Before apply or plan, run "echo Foo".
before_hook "before_hook_1" {
commands = ["apply", "plan"]
execute = ["echo", "Foo"]
}
# Before apply, run "echo Bar". Note that blocks are ordered, so this hook will run after the previous hook to
# "echo Foo". In this case, always "echo Bar" even if the previous hook failed.
before_hook "before_hook_2" {
commands = ["apply"]
execute = ["echo", "Bar"]
run_on_error = true
}
# Note that you can use interpolations in subblocks. Here, we configure it so that before apply or plan, print out the
# environment variable "HOME".
before_hook "interpolation_hook_1" {
commands = ["apply", "plan"]
execute = ["echo", get_env("HOME", "HelloWorld")]
run_on_error = false
}
# After running apply or plan, run "echo Baz". This hook is configured so that it will always run, even if the apply
# or plan failed.
after_hook "after_hook_1" {
commands = ["apply", "plan"]
execute = ["echo", "Baz"]
run_on_error = true
}
# A special after hook to always run after the init-from-module step of the Terragrunt pipeline. In this case, we will
# copy the "foo.tf" file located by the parent terragrunt.hcl file to the current working directory.
after_hook "init_from_module" {
commands = ["init-from-module"]
execute = ["cp", "${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 "terragrunt-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 "terragrunt-read-config" {
commands = ["terragrunt-read-config"]
execute = ["bash", "script/get_aws_credentials.sh"]
}
}
The key design of Terragrunt is to act as a preprocessor to convert shared service modules in the registry into a root module. In Terraform, modules can be loosely categorized into two types:
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 Terraform behaves, like backend blocks (for
configuring state) and provider blocks (for configuring how Terraform interacts with the cloud APIs).module
blocks. These modules are missing many of the key blocks that are required for running the workflow commands of
terraform.Terragrunt further distinguishes shared modules between service modules and modules:
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.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 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 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 in many cases, 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:
type associated with it. Otherwise, Terraform will interpret the input that
Terragrunt passes through as string. This includes list and map.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.
The remote_state block is used to configure how Terragrunt will set up the remote state configuration of your
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
backend types that 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 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 Terraform. All the
properties will automatically be included in the Terraform backend block (with a few exceptions: see below). For
example, if you had the following remote_state block:
remote_state {
backend = "s3"
config = {
bucket = "mybucket"
key = "path/to/my/key"
region = "us-east-1"
}
}
This is equivalent to the following terraform code:
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:
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:
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
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.encrypt - (Optional) Whether to enable server side encryption of the state file.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.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.enable_lock_table_ssencryption: When true, the synchronization lock table in DynamoDB used for remote state concurrent access will not 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.disable_aws_client_checksums: When true, disable computing and checking checksums on the request and response,
such as the CRC32 check for DynamoDB. This can be used to workaround
https://github.com/gruntwork-io/terragrunt/issues/1059.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 Terraform state. If not provided, or string is empty or invalid S3 bucket name, then server access logging for the S3 bucket storing the terraform state will be disabled.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 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.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.Example with S3:
# Configure terraform state to be stored in S3, in the bucket "my-terraform-state" in us-east-1 under a key that is
# relative to included terragrunt config. For example, if you had the following folder structure:
#
# .
# ├── terragrunt.hcl
# └── child
# └── 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/terraform.tfstate".
#
# Note that since we are not using any of the skip args, this will automatically create the S3 bucket
# "my-terraform-state" and DynamoDB table "my-lock-table" if it does not already exist.
remote_state {
backend = "s3"
config = {
bucket = "my-terraform-state"
key = "${path_relative_to_include()}/terraform.tfstate"
region = "us-east-1"
encrypt = true
dynamodb_table = "my-lock-table"
}
}
Example with GCS:
# Configure terraform state to be stored in GCS, in the bucket "my-terraform-state" in the "my-terraform" GCP project in
# the eu region under a key that is relative to included terragrunt config. This will also apply the labels
# "owner=terragrunt_test" and "name=terraform_state_storage" to the bucket if it is created by Terragrunt.
#
# For example, if you had the following folder structure:
#
# .
# ├── terragrunt.hcl
# └── child
# └── terragrunt.hcl
#
# And the following is defined in the root terragrunt.hcl config that is included in the child, the state file for the
# child module will be stored at the key "child/terraform.tfstate".
#
# Note that since we are not using any of the skip args, this will automatically create the GCS bucket
# "my-terraform-state" if it does not already exist.
remote_state {
backend = "gcs"
config = {
project = "my-terraform"
location = "eu"
bucket = "my-terraform-state"
prefix = "${path_relative_to_include()}/terraform.tfstate"
gcs_bucket_labels = {
owner = "terragrunt_test"
name = "terraform_state_storage"
}
}
}
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
https://github.com/gruntwork-io/terragrunt/issues/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
# If you have the following folder structure, and the following contents for ./child/terragrunt.hcl, this will include
# and merge the items in the terragrunt.hcl file at the root.
#
# .
# ├── terragrunt.hcl
# └── child
# └── terragrunt.hcl
include "root" {
path = find_in_parent_folders()
expose = true
}
inputs = {
remote_state_config = include.root.remote_state
}
Multiple includes
# If you have the following folder structure, and the following contents for ./child/terragrunt.hcl, this will include
# and merge the items in the terragrunt.hcl file at the root, while only loading the data in the region.hcl
# configuration.
#
# .
# ├── terragrunt.hcl
# ├── region.hcl
# └── child
# └── terragrunt.hcl
include "remote_state" {
path = find_in_parent_folders()
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.region
}
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 terragrunt.hcl
dependency "vpc" {
config_path = "${get_terragrunt_dir()}/../vpc"
}
inputs = {
vpc_name = dependency.vpc.outputs.name
}
child terragrunt.hcl
include "root" {
path = find_in_parent_folders()
expose = true
}
dependency "alb" {
config_path = (
include.root.inputs.vpc_name == "mgmt"
? "../alb-public"
: "../alb-private"
)
}
input = {
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 ourput 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 terragrunt.hcl
inputs = {
vpc_name = "mgmt"
}
child terragrunt.hcl
include "root" {
path = find_in_parent_folders()
expose = true
}
dependency "vpc" {
config_path = "../vpc"
}
dependency "alb" {
config_path = (
include.root.inputs.vpc_name == "mgmt"
? "../alb-public"
: "../alb-private"
)
}
input = {
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:
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:
parent config
dependency "vpc" {
config_path = "../vpc"
}
inputs = {
vpc_id = dependency.vpc.outputs.vpc_id
db_id = dependency.mysql.outputs.db_id
}
child config
include "root" {
path = find_in_parent_folders()
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:
parent terragrunt.hcl
remote_state {
backend = "s3"
config = {
encrypt = true
bucket = "__FILL_IN_BUCKET_NAME__"
key = "${path_relative_to_include()}/terraform.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
}
}
child terragrunt.hcl
include "root" {
path = find_in_parent_folders()
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
}
The locals block is used to define aliases for Terragrunt expressions that can be referenced within the configuration.
You can learn more about locals in the feature overview.
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.ARG_NAME throughout the Terragrunt configuration.
Example:
# 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"
}
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 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 of this dependency with the expression dependency.vpc.outputs.config_path (attribute): Path to a Terragrunt module (folder with a terragrunt.hcl file) that should be included
as a dependency in this configuration.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.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.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): When true, mock_outputs and the state outputs will be merged.Example:
# 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
}
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 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:
remote_state blocks.disable_dependency_optimization = false, which is the default).remote_state block itself does not depend on any dependency outputs (locals and include are ok).before_hook, after_hook, or extra_arguments to the 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.
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
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:
# 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"]
}
The generate block can be used to arbitrarily generate a file in the terragrunt working directory (where terraform
is called). This can be used to generate common terraform configurations that are shared across multiple 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 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).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.Example:
# When using this terragrunt config, terragrunt will generate the file "provider.tf" with the aws provider block before
# calling to 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:
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:
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
The inputs attribute is a map that is used to specify the input variables and their values to pass in to Terraform.
Each entry of the map will be passed to Terraform using the environment variable
mechanism. This means that each input
will be set using the form TF_VAR_variablename, with the value in json encoded format.
Note that because the values are being passed in with environment variables and json, the type information is lost
when crossing the boundary between Terragrunt and Terraform. You must specify the proper type
constraint on the variable in Terraform in
order for Terraform to process the inputs to the right type.
Example:
inputs = {
string = "string"
number = 42
bool = true
list_string = ["a", "b", "c"]
list_number = [1, 2, 3]
list_bool = [true, false]
map_string = {
foo = "bar"
}
map_number = {
foo = 42
bar = 12345
}
map_bool = {
foo = true
bar = false
baz = true
}
object = {
str = "string"
num = 42
list = [1, 2, 3]
map = {
foo = "bar"
}
}
from_env = get_env("FROM_ENV", "default")
}
The terragrunt download_dir string option can be used to override the default download directory.
The precedence is as follows: --terragrunt-download-dir command line option → TERRAGRUNT_DOWNLOAD env variable →
download_dir attribute of the terragrunt.hcl file in the module directory → download_dir attribute of the included
terragrunt.hcl.
It supports all terragrunt functions, i.e. path_relative_from_include().
Terragrunt prevent_destroy boolean flag allows you to protect selected Terraform module. It will prevent destroy or
destroy-all command to actually destroy resources of the protected module. This is useful for modules you want to
carefully protect, such as a database, or a module that provides auth.
Example:
terraform {
source = "git::git@github.com:foo/modules.git//app?ref=v0.0.3"
}
prevent_destroy = true
The terragrunt skip boolean flag can be used to protect modules you don’t want any changes to or just to skip modules
that don’t define any infrastructure by themselves. When set to true, all terragrunt commands will skip the selected
module.
Consider the following file structure:
root
├── terragrunt.hcl
├── prod
│ └── terragrunt.hcl
├── dev
│ └── terragrunt.hcl
└── qa
└── terragrunt.hcl
In some cases, the root level terragrunt.hcl file is solely used to DRY up your Terraform configuration by being
included in the other terragrunt.hcl files. In this case, you do not want the run-all commands to process the root
level terragrunt.hcl since it does not define any infrastructure by itself. To make the run-all commands skip the
root level terragrunt.hcl file, you can set skip = true:
skip = true
The skip flag must be set explicitly in terragrunt modules that should be skipped. If you set skip = true in a
terragrunt.hcl file that is included by another terragrunt.hcl file, only the terragrunt.hcl file that explicitly
set skip = true will be skipped.
The iam_role attribute can be used to specify an IAM role that Terragrunt should assume prior to invoking Terraform.
The precedence is as follows: --terragrunt-iam-role command line option → TERRAGRUNT_IAM_ROLE env variable →
iam_role attribute of the terragrunt.hcl file in the module directory → iam_role attribute of the included
terragrunt.hcl.
Example:
iam_role = "arn:aws:iam::ACCOUNT_ID:role/ROLE_NAME"
Notes:
iam_role can reference local variablesiam_role included from other HCL files through includeThe iam_assume_role_duration attribute can be used to specify the STS session duration, in seconds, for the IAM role that Terragrunt should assume prior to invoking Terraform.
The precedence is as follows: --terragrunt-iam-assume-role-duration command line option → TERRAGRUNT_IAM_ASSUME_ROLE_DURATION env variable →
iam_assume_role_duration attribute of the terragrunt.hcl file in the module directory → iam_assume_role_duration attribute of the included
terragrunt.hcl.
Example:
iam_assume_role_duration = 14400
The iam_assume_role_session_name attribute can be used to specify the STS session name, for the IAM role that Terragrunt should assume prior to invoking Terraform.
The precedence is as follows: --terragrunt-iam-assume-role-session-name command line option → TERRAGRUNT_IAM_ASSUME_ROLE_SESSION_NAME env variable →
iam_assume_role_session_name attribute of the terragrunt.hcl file in the module directory → iam_assume_role_session_name attribute of the included
terragrunt.hcl.
The terragrunt terraform_binary string option can be used to override the default terraform binary path (which is
terraform).
The precedence is as follows: --terragrunt-tfpath command line option → TERRAGRUNT_TFPATH env variable →
terragrunt.hcl in the module directory → included terragrunt.hcl
The terragrunt terraform_version_constraint string overrides the default minimum supported version of terraform.
Terragrunt only officially supports the latest version of terraform, however in some cases an old terraform is needed.
Example:
terraform_version_constraint = ">= 0.11"
The terragrunt terragrunt_version_constraint string can be used to specify which versions of the Terragrunt CLI can be used with your configuration. If the running version of Terragrunt doesn’t match the constraints specified, Terragrunt will produce an error and exit without taking any further actions.
Example:
terragrunt_version_constraint = ">= 0.23"
The terragrunt retryable_errors list can be used to override the default list of retryable errors with your own custom list.
To learn more about the retryable_errors attribute, see the auto-retry feature overview.
Example:
retryable_errors = [
"(?s).*Error installing provider.*tcp.*connection reset by peer.*",
"(?s).*ssh_exchange_identification.*Connection closed by remote host.*"
]
Install
