stacker makes use of a YAML formatted config file to define the different CloudFormation stacks that make up a given environment.

The configuration file has a loose definition, with only a few top-level keywords. Other than those keywords, you can define your own top-level keys to make use of other YAML features like anchors & references to avoid duplicating config. (See YAML anchors & references for details)

Top Level Keywords


You can provide a namespace to create all stacks within. The namespace will be used as a prefix for the name of any stack that stacker creates, and makes it unnecessary to specify the fully qualified name of the stack in output lookups.

In addition, this value will be used to create an S3 bucket that stacker will use to upload and store all CloudFormation templates.

In general, this is paired with the concept of Environments to create a namespace per environment:

namespace: ${namespace}

Namespace Delimiter

By default, stacker will use ‘-‘ as a delimiter between your namespace and the declared stack name to build the actual CloudFormation stack name that gets created. Since child resources of your stacks will, by default, use a portion of your stack name in the auto-generated resource names, the first characters of your fully-qualified stack name potentially convey valuable information to someone glancing at resource names. If you prefer to not use a delimiter, you can pass the namespace_delimiter top level key word in the config as an empty string.

See the CloudFormation API Reference for allowed stack name characters

S3 Bucket

Stacker, by default, pushes your CloudFormation templates into an S3 bucket and points CloudFormation at the template in that bucket when launching or updating your stacks. By default it uses a bucket named stacker-${namespace}, where the namespace is the namespace provided the config.

If you want to change this, provide the stacker_bucket top level key word in the config.

The bucket will be created in the same region that the stacks will be launched in. If you want to change this, or if you already have an existing bucket in a different region, you can set the stacker_bucket_region to the region where you want to create the bucket.

S3 Bucket location prior to 1.0.4:
There was a “bug” early on in stacker that created the s3 bucket in us-east-1, no matter what you specified as your –region. An issue came up leading us to believe this shouldn’t be the expected behavior, so we fixed the behavior. If you executed a stacker build prior to V 1.0.4, your bucket for templates would already exist in us-east-1, requiring you to specify the stacker_bucket_region top level keyword.


Deprecation of fallback to legacy template bucket. We will first try the region you defined using the top level keyword under stacker_bucket_region, or what was specified in the –region flag. If that fails, we fallback to the us-east-1 region. The fallback to us-east-1 will be removed in a future release resulting in the following botocore excpetion to be thrown:

TemplateURL must reference a valid S3 object to which you have access.

To avoid this issue, specify the stacker_bucket_region top level keyword as described above. You can specify this keyword now to remove the deprecation warning.

If you want stacker to upload templates directly to CloudFormation, instead of first uploading to S3, you can set stacker_bucket to an empty string. However, note that template size is greatly limited when uploading directly. See the CloudFormation Limits Reference.

Module Paths

When setting the classpath for blueprints/hooks, it is sometimes desirable to load modules from outside the default sys.path (e.g., to include modules inside the same repo as config files).

Adding a path (e.g. ./) to the sys_path top level key word will allow modules from that path location to be used.

Service Role

By default stacker doesn’t specify a service role when executing changes to CloudFormation stacks. If you would prefer that it do so, you can set service_role to be the ARN of the service that stacker should use when executing CloudFormation changes.

This is the equivalent of setting RoleARN on a call to the following CloudFormation api calls: CreateStack, UpdateStack, CreateChangeSet.

See the AWS documentation for AWS CloudFormation Service Roles.

Remote Packages

The package_sources top level keyword can be used to define remote sources for blueprints (e.g., retrieving stacker_blueprints on github at tag v1.0.2).

The only required key for a git repository config is uri, but branch, tag, & commit can also be specified:

    - uri:
    - uri:
      tag: 1.0.0
        - stacker_blueprints
    - uri:
      branch: staging
    - uri:
      commit: 12345678

If no specific commit or tag is specified for a repo, the remote repository will be checked for newer commits on every execution of Stacker.

For .tar.gz & zip archives on s3, specify a bucket & key:

    - bucket: mystackers3bucket
      key: archives/
        - stacker_blueprints
    - bucket: anothers3bucket
      key: public/public-blueprints-v2.tar.gz
      requester_pays: true
    - bucket: yetanothers3bucket
      key: sallys-blueprints-v1.tar.gz
      # use_latest defaults to true - will update local copy if the
      # last modified date on S3 changes
      use_latest: false

Use the paths option when subdirectories of the repo/archive should be added to Stacker’s sys.path.

Cloned repos/archives will be cached between builds; the cache location defaults to ~/.stacker but can be manually specified via the stacker_cache_dir top level keyword.

Remote Configs

Configuration yamls from remote configs can also be used by specifying a list of configs in the repo to use:

    - uri:
        - vpc.yaml

In this example, the configuration in vpc.yaml will be merged into the running current configuration, with the current configuration’s values taking priority over the values in vpc.yaml.

Dictionary Stack Names & Hook Paths

To allow remote configs to be selectively overriden, stack names & hook paths can optionally be defined as dictionaries, e.g.:

    path: stacker.hooks.route53.create_domain:
    required: true
    class_path: stacker_blueprints.vpc.VPC
    locked: false
    enabled: true
    class_path: stacker_blueprints.bastion.Bastion
    locked: false
    enabled: true

Pre & Post Hooks

Many actions allow for pre & post hooks. These are python methods that are executed before, and after the action is taken for the entire config. Only the following actions allow pre/post hooks:

  • build (keywords: pre_build, post_build)
  • destroy (keywords: pre_destroy, post_destroy)

There are a few reasons to use these, though the most common is if you want better control over the naming of a resource than what CloudFormation allows.

The keyword is a list of dictionaries with the following keys:

the python import path to the hook
If set, and the hook returns data (a dictionary), the results will be stored in the context.hook_data with the data_key as it’s key.
whether to stop execution if the hook fails
a dictionary of arguments to pass to the hook

An example using the create_domain hook for creating a route53 domain before the build action:

  - path: stacker.hooks.route53.create_domain
    required: true


CloudFormation supports arbitrary key-value pair tags. All stack-level, including automatically created tags, are propagated to resources that AWS CloudFormation supports. See AWS CloudFormation Resource Tags Type for more details. If no tags are specified, the stacker_namespace tag is applied to your stack with the value of namespace as the tag value.

If you prefer to apply a custom set of tags, specify the top-level keyword tags as a map. Example:

  "hello": world
  "my_tag:with_colons_in_key": ${dynamic_tag_value_from_my_env}
  simple_tag: simple value

If you prefer to have no tags applied to your stacks (versus the default tags that stacker applies), specify an empty map for the top-level keyword:

tags: {}


Mappings are dictionaries that are provided as Mappings to each CloudFormation stack that stacker produces.

These can be useful for providing things like different AMIs for different instance types in different regions:

      NAT: ami-ad227cc4
      ubuntu1404: ami-74e27e1c
      bastion: ami-74e27e1c
      NAT: ami-290f4119
      ubuntu1404: ami-5189a661
      bastion: ami-5189a661

These can be used in each blueprint/stack as usual.


Lookups allow you to create custom methods which take a value and are resolved at build time. The resolved values are passed to the Blueprints before it is rendered. For more information, see the Lookups documentation.

stacker provides some common lookups, but it is sometimes useful to have your own custom lookup that doesn’t get shipped with stacker. You can register your own lookups by defining a lookups key:


The key name for the lookup will be used as the type name when registering the lookup. The value should be the path to a valid lookup handler.

You can then use these within your config:

conf_value: ${custom some-input-here}


This is the core part of the config - this is where you define each of the stacks that will be deployed in the environment. The top level keyword stacks is populated with a list of dictionaries, each representing a single stack to be built.

A stack has the following keys:

The base name for the stack (note: the namespace from the environment will be prepended to this)
The python class path to the Blueprint to be used. Specify this or template_path for the stack.
Path to raw CloudFormation template (JSON or YAML). Specify this or class_path for the stack.
A short description to apply to the stack. This overwrites any description provided in the Blueprint. See:
A dictionary of Variables to pass into the Blueprint when rendering the CloudFormation template. Variables can be any valid YAML data structure.
(optional) If set to true, the stack is locked and will not be updated unless the stack is passed to stacker via the –force flag. This is useful for risky stacks that you don’t want to take the risk of allowing CloudFormation to update, but still want to make sure get launched when the environment is first created.
(optional) If set to false, the stack is disabled, and will not be built or updated. This can allow you to disable stacks in different environments.
(optional) When running an update in non-interactive mode, if a stack has protected set to true and would get changed, stacker will switch to interactive mode for that stack, allowing you to approve/skip the change.
(optional) a list of other stacks this stack requires. This is for explicit dependencies - you do not need to set this if you refer to another stack in a Parameter, so this is rarely necessary.
(optional) a dictionary of CloudFormation tags to apply to this stack. This will be combined with the global tags, but these tags will take precendence.

Here’s an example from stacker_blueprints, used to create a VPC:

  - name: vpc-example
    class_path: stacker_blueprints.vpc.VPC
    locked: false
    enabled: true
      InstanceType: t2.small
      SshKeyName: default
      ImageName: NAT
      AZCount: 2


Variables are values that will be passed into a Blueprint before it is rendered. Variables can be any valid YAML data structure and can leverage Lookups to expand values at build time.

The following concepts make working with variables within large templates easier:

YAML anchors & references

If you have a common set of variables that you need to pass around in many places, it can be annoying to have to copy and paste them in multiple places. Instead, using a feature of YAML known as anchors & references, you can define common values in a single place and then refer to them with a simple syntax.

For example, say you pass a common domain name to each of your stacks, each of them taking it as a Variable. Rather than having to enter the domain into each stack (and hopefully not typo’ing any of them) you could do the following:

domain_name: &domain

Now you have an anchor called domain that you can use in place of any value in the config to provide the value You use the anchor with a reference:

  - name: vpc
    class_path: stacker_blueprints.vpc.VPC
      DomainName: *domain

Even more powerful is the ability to anchor entire dictionaries, and then reference them in another dictionary, effectively providing it with default values. For example:

common_variables: &common_variables
  InstanceType: m3.medium
  AMI: ami-12345abc

Now, rather than having to provide each of those variables to every stack that could use them, you can just do this instead:

  - name: vpc
    class_path: stacker_blueprints.vpc.VPC
      << : *common_variables
      InstanceType: c4.xlarge # override the InstanceType in this stack

Using Outputs as Variables

Since stacker encourages the breaking up of your CloudFormation stacks into entirely separate stacks, sometimes you’ll need to pass values from one stack to another. The way this is handled in stacker is by having one stack provide Outputs for all the values that another stack may need, and then using those as the inputs for another stack’s Variables. stacker makes this easier for you by providing a syntax for Variables that will cause stacker to automatically look up the values of Outputs from another stack in its config. To do so, use the following format for the Variable on the target stack:

MyParameter: ${output OtherStack::OutputName}

Since referencing Outputs from stacks is the most common use case, output is the default lookup type. For more information see Lookups.

This example is taken from stacker_blueprints example config - when building things inside a VPC, you will need to pass the VpcId of the VPC that you want the resources to be located in. If the vpc stack provides an Output called VpcId, you can reference it easily:

domain_name: my_domain &domain

  - name: vpc
    class_path: stacker_blueprints.vpc.VPC
      DomainName: *domain
  - name: webservers
    class_path: stacker_blueprints.asg.AutoscalingGroup
      DomainName: *domain
      VpcId: ${output vpc::VpcId} # gets the VpcId Output from the vpc stack

Note: Doing this creates an implicit dependency from the webservers stack to the vpc stack, which will cause stacker to submit the vpc stack, and then wait until it is complete until it submits the webservers stack.


A pretty common use case is to have separate environments that you want to look mostly the same, though with some slight modifications. For example, you might want a production and a staging environment. The production environment likely needs more instances, and often those instances will be of a larger instance type. Environments allow you to use your existing stacker config, but provide different values based on the environment file chosen on the command line. For more information, see the Environments documentation.



Translators have been deprecated in favor of Lookups and will be removed in a future release.

Translators allow you to create custom methods which take a value, then modify it before passing it on to the stack. Currently this is used to allow you to pass a KMS encrypted string as a Parameter, then have KMS decrypt it before submitting it to CloudFormation. For more information, see the Translators documentation.