Updated December 6, 2017
On both Google Cloud and Amazon Web Services (AWS), you can manage your cloud environments using an infrastructure-as-code approach. Google Cloud offers Deployment Manager, and AWS offers CloudFormation.
Treating your infrastructure as code involves defining your environment through scripts, templates, and configuration files, and keeping the files in a source code repository. Using this approach, you can:
- Control the version of your configuration.
- Deploy consistent, reproducible configurations.
- Review an audit trail of changes to your configuration.
- Use the configuration as part of a continuous deployment system.
- Easily fail back to a previous known-good configuration.
Service model comparison
The following table maps high-level CloudFormation features, terms, and concepts to Deployment Manager features, terms, and concepts:
|Feature||AWS CloudFormation||Google Cloud Deployment Manager|
|Deployed collection of resources||Stack||Deployment|
|Deployment files||Template file||Configuration files, template files, and schema files|
|Syntax||JSON, YAML||YAML, Jinja, Python|
|Composition and reuse||Nested stacks||Templates|
|Identification of individual resources||Logical ID||Name|
|Scope of deployment locality||Regional||Global|
|Default maximum number of stacks or deployments||200||1000|
|Graphical user interface||Yes||No|
Templates and configuration
The following table compares available template and configuration features in CloudFormation to those in Deployment Manager:
|Feature||AWS CloudFormation||Google Cloud Deployment Manager|
|Maximum size of template or configuration||460.8 KB||1 MB|
In AWS CloudFormation, you create a template that defines a set of actions against various services, such as creating an S3 bucket or launching an EC2 instance. An AWS template can be expressed in YAML or JSON, and AWS CloudFormation can invoke a template from either an S3 bucket or your local machine. When AWS CloudFormation invokes a template, it makes calls to the APIs of the services you have defined in the template to launch and configure the services.
To use the services defined in a template, you must have appropriate AWS IAM permissions for each service. You can also set IAM policies to restrict or allow the use of CloudFormation.
Deployment Manager uses three types of files to define a deployment:
- A configuration file, which defines the structure of your deployment and configures the resources to be deployed. This file is written in YAML. You can use this file as a standalone configuration for your deployment, or you can combine it with imported template files for increased flexibility.
- Template files, which are composable, reusable resource definitions. Template files are written in Jinja 2.7.3 or Python 2.7. You can add loops to create multiple instances of the same resource without having to explicitly declare each instance.
- Schema files, which define sets of rules that a configuration file must meet in order to use a particular template.
You pass the configuration file (which will expand any templates that it
references) to the Deployment Manager service through the
tool or API. You can also launch predefined configurations by using
Google Cloud Marketplace.
By default, Deployment Manager uses the credentials of the GCP API service account to authenticate to other APIs. You must grant the user using Deployment Manager the appropriate permissions to use Deployment Manager. You can use predefined IAM roles to limit the permissions appropriately. Deployment Manager allows you to automate the creation of projects themselves as well as the resources contained within the projects.
A Deployment Manager configuration file consists of the following sections:
imports: - path: network.jinja resources: - name: network type: network.jinja properties: region: us-central1 subnets: - range: 10.177.0.0/17 name: web - range: 10.177.128.0/17 name: data - name: web-instance type: compute.v1.instance properties: zone: us-central1-f machineType: https://www.googleapis.com/compute/v1/projects/myproject/zones/us-central1-f/machineTypes/f1-micro disks: - deviceName: boot type: PERSISTENT boot: true autoDelete: true initializeParams: sourceImage: https://www.googleapis.com/compute/v1/projects/debian-cloud/global/images/family/debian-9 networkInterfaces: - network: https://www.googleapis.com/compute/v1/projects/myproject/global/networks/default accessConfigs: - name: External NAT type: ONE_TO_ONE_NAT
The above Google Cloud example configuration consists of the following parts:
Imports. This references the path to composable templates that are expanded as part of the actual deployment. This is optional, because you might have a simple deployment where you do not need to create a configuration where you have placed individual resources into composite template files
Resources. This is a list of the resources that you want deployed as part of your configuration. This is the only mandatory requirement for a configuration. Each resource is defined by the following constituent parts:
- Name. A name that you define for this resource. In the example
above we have two resources called
- Type. Specifies the base type of the resource. The example
shows two types of resource. The
web-instanceis of type
compute.v1.instance. The second resource refers to the import file declared in the
importssection of the configuration file. In the example it is a Jinja file type, but you can also import Python files.
- Properties. These are the properties associated with a resource. The
web-instancedefines a Compute Engine instance of machine type
f1-micro,which has a persistent disk that can be auto deleted when you delete the instance. The instance is to be deployed into
us-central1-fregion. The second resource is defined in a separate template file called
network.jinja. For this resource, you define the values for the properties in the configuration file. The values are passed through to the actual template file. This composability lends itself to building complex deployments with reusable resource definitions, to which you can pass values through the configuration files. For example, you could use the same template file to create more subnets with different IP ranges and names.
A schema file defines a set of rules for how to use a template. Defining the rules in a single schema gives your users a way to interact with the templates you write. They can use the schema file without needing to examine the templates that constitute a configuration or to learn which properties need to be set and which have default values. For more information on schema files, see the Deployment Manager documentation.
The example template uses a public image to create the instance. GCP takes a global approach, so when you select an image type to launch an instance, you do not need to create a map of the image type per region, as you would for AWS CloudFormation. Using Google Cloud, you declare the machine type and the source image. The source image is available to all projects.
Composition and reuse
Both AWS CloudFormation and Google Cloud Deployment Manager encourage you to compose your own stacks or deployments and to reuse stacks or deployments.
AWS CloudFormation relies on the concept of nested stacks. With nested stacks,
you can create a template for the resources that you want to reuse. You store
the template that you want to call from the master template in an S3 bucket and
AWS::CloudFormation::Stack type. You pass the S3 URL of the template as
You can use parameters to easily customise a template each time you use it to create a stack. You can use intrinsic functions together with parameters to create conditional resources.
You can use outputs to export a value through the logical ID assigned to the value, so that you can use the value elsewhere. For example, you can export the names of load balancers and S3 buckets, import values into other stacks, and expose the values in the console.
CloudFormation creates resources in parallel and determines which resources can
be deployed in parallel, but you might need to use the creation policy and
DependsOn attributes to help enforce the order in which resources are created.
For example, the deployment might require EBS volumes to be created before the
instance is created that uses the volumes.
You can use custom resources to include non-AWS resources as part of your
CloudFormation stack. To implement a custom resource, you supply a service token
Properties section of the template.
The service token specifies where CloudFormation should send requests.
Google Cloud Deployment Manager
Deployment Manager allows you to import templates into your configuration
file. You declare the path to the templates, which can be stored locally, or in a
Cloud Storage bucket, or at a publicly accessible URL
such as GitHub, in the
import statement. This allows you to scale and reuse
templates as part of your configuration.
You can customize Deployment Manager templates by passing variables through to your templates. You can use two types of variables with Deployment Manager: template variables and environment variables. Using template variables, you define the customizable value in the configuration file and pass that through to the template files. Environment variables are predefined and automatically assigned. Environment variables are designed to be used for items that are not directly related to resources you want to deploy, such as project IDs and deployment names.
Both Jinja and Python allow you to take advantage of their native conditionals and looping functions.
Deployment Manager relies on the concept of template modules. You can use template modules to incorporate template snippets that can be used in multiple templates. For example, you can use a template module to always generate instance names based on their environment and function.
Deployment Manager creates all resources in parallel. You can use references to enforce the order in which resources are created. For example, if an instance references a new network in a configuration, you can ensure that Deployment Manager always creates the network before creating the instance, no matter where in the configuration file the network is declared.
Outputs in Deployment Manager are designed to expose key properties of resources in an easily consumable way, so that they can be used by other templates or be exposed in the configuration. Outputs are declared as key-value pairs and can consist of a static string, a reference to a property, a template variable, or an environment variable.
If the Google Cloud–supported types don't meet your needs, you can create a type provider and create additional types that you can register with Deployment Manager. To create a type provider, you must supply an API descriptor document, which can be an OpenAPI specification or a Google Discovery document. When you create and register a type provider, all resources provided by the API and supported by a RESTful interface with create, read, update, and delete (CRUD) operations are exposed as types that you can use in your deployment.
You can extend the base types to create a composite type. A composite type is a combination of multiple base types that are configured to work together. For example, a network load-balanced managed instance group is a composite type. A network load balancer requires multiple Google Cloud resources and some configuration between resources. You can set up these resources in a configuration once and register the type with Deployment Manager.
Metadata, helper scripts, and startup scripts
AWS CloudFormation allows you to use a set of Python scripts that help
you start services and install software on EC2 instances. The scripts are
available from a repository or installed on an Amazon Linux AMI, and you
can call them from your templates. You can add metadata or retrieve metadata on
your instances by using the
GCP Deployment Manager allows you to use
or add metadata to virtual machine instances in your deployment by specifying
the metadata in your template or configuration. The
metadata key for your startup script must be
startup-script, and the value
is the contents of your
The example snippet below shows how this is done:
resources: - name: my-first-vm-template type: compute.v1.instance properties: zone: us-central1-f machineType: ...[snip]... metadata: items: - key: startup-script value: "STARTUP-SCRIPT-CONTENTS"
Updating stacks and deployments
You can update CloudFormation stacks by passing new parameter values or updated templates. You can use change sets to see which stack settings and resources will change before you apply the change set.
Cloud Deployment Manager
With Deployment Manager you can also update deployments and preview potential changes to assess the effect your changes will have. Deployment Manager does not instantiate any actual resources when you preview a configuration. Instead, it expands the full configuration and creates shell resources. This gives you the opportunity to see the deployment before committing to it.
Deployment Manager also provides policies that you can use when you update a deployment to help you control how Deployment Manager manages updates. Some resources are immutable and might not have an update method. As with AWS CloudFormation, you should check the API methods for the resource to understand the implications of carrying out an update.
You can also look at manifests prior to doing any updates to understand what a deployment actually looks like. You can do this through the command-line interface or the console.
There is no charge for AWS CloudFormation or Google Cloud Deployment Manager. In both cases you are charged only for the resources you launch.
Check out the other Google Cloud Platform for AWS Professionals articles: