Bake Machine Images on AWS Using Packer
Overview of baking images in Spinnaker
SpinnakerTM uses the open source Packer tool to bake images, which is included in the Spinnaker Rosco microservice. For example, if you are deploying to AWS, you can use Spinnaker to bake Amazon Machine Images (AMIs) from the artifacts that were produced by your CI tool.
This section focuses on configuring Packer scripts to build machine images (such as AMIs). If you’re only deploying to Kubernetes, you can skip this section.
Configuring AWS bake credentials
If you’ve configured Spinnaker to deploy to AWS, then you’ve likely set up a set of IAM credentials and permissions for Clouddriver to be able to deploy to AWS (using the AWS API). Since Packer is run from a different microservice in Spinnaker (Rosco), configuration of Bake credentials is separate from the configuration of Deploy credentials.
AMI baking primarily occurs in a single AWS account; you can configure baked AMIs to be shared to additional desired AWS accounts
AWS bake credentials using IAM credentials
If you’ve configured Spinnaker to interact with AWS using explicit credentials (an AWS Access Key and Secret Access Key), you can likewise configure Rosco to use a set of AWS credentials. What you essentially need are an IAM user with permissions to do the things that Packer needs to do, and then you can pass those credentials to Rosco.
The AWS account that you’re baking in must also match an account configured as a Managed Account, and that Managed Account must be configured as the primary AWS account within Spinnaker.
This User must have all permissions necessary to bake (for example, PowerUserAccess and associated PassRoles)
This User may be, but does not have to be, the same as the Managing Account User.
Spinnaker will always bake with this user. If you need to deploy to other accounts, update your Packer template to support sharing the baked image with other accounts. For example, add this to your builder
configuration in your packer template (and add the custom packer template following the instructions in the Spinnaker Packer documentation:
"ami_users": ["222222222222","333333333333"]
Create an IAM user
If you don’t have an IAM user with the desired permissions, you can create one by doing the following:
Log into the AWS account where Spinnaker lives, into the browser-based AWS Console
Navigate to the IAM page (click on “Services” at the top, then on “IAM” under “Security, Identity, & Compliance”)
Click on “Users” on the left side
Click on “Add user”
Specify a logical user name, such as “SpinnakerBake”
Check the “Programmatic access” checkbox
Select “Attach existing policies directly”
Find and attach the AWS “PowerUserAccess” Policy
Click “Next: Tags”
Optionally, add tags that will identify this user
Click “Next: Review”
Click “Create user”
Copy the “Access key ID” and “Secret access key” (you’ll have to click “Show”).
Click “Close”
Click on the “User name” for the user that you just created.
Click on “Add inline policy” (on the right).
Click on the “JSON” tab, and paste in this:
{ "Version": "2012-10-17", "Statement": [ { "Action": [ "iam:PassRole" ], "Resource": [ "*" ], "Effect": "Allow" } ] }
Click on “Review Policy”
Call it “PassRole” and then click “Create Policy”
Add IAM user credentials to Rosco
Using the “Access key ID” and “Secret access key” for your generated user, run this command in Halyard:
# You will be prompted for the secret access key
hal config provider aws bakery edit --aws-access-key "YOUR_ACCESS_KEY" --aws-secret-key
Then deploy your changes with this:
hal deploy apply
AWS bake credentials using IAM profiles
If you’re using IAM Instance Roles, you need to provide credentials to Spinnaker to use to Bake by adding additional policies to the EC2 instances where Spinnaker is running. The AWS account that you’re baking in must also be configured as a Managed Account, and that Managed Account must be configured as the primary AWS account within Spinnaker.
These policies must have all permissions necessary to bake (for example, PowerUserAccess and associated PassRoles)
Spinnaker will always bake with the EC2 instance role (unless you specify explicit baking creds). If you need to deploy to other accounts, update your Packer template to support sharing the baked image with other accounts. For example, add this to your builder
configuration in your packer template (and add the custom packer template following the instructions in the Spinnaker Packer documentation:
"ami_users": ["222222222222","333333333333"]
If you don’t configure Rosco with explicit AWS credentials to use, Packer will default to the AWS permissions available to the Rosco container. In general, this means that Packer will use the IAM Role/Profile attached to the Kubernetes nodes where Spinnaker is running. In order for this to work, the IAM Role/Profile attached to your Kubernetes cluster will need a set of permissions to be able to create and interact with EC2 instances (and assign roles to those EC2 instances).
First, identify the IAM Profile attached to the Kubernetes cluster where Spinnaker is running. If you’re running EKS, this will be the role attached to the EKS EC2 worker nodes.
Go to the Role Summary for the IAM Profile, and click “Attach policies”
Find and select the AWS “PowerUserAccess” Policy
Click “Attach policy”
Click on “Add inline policy” (on the right).
Click on the “JSON” tab, and paste in this:
{ "Version": "2012-10-17", "Statement": [ { "Action": [ "iam:PassRole" ], "Resource": [ "*" ], "Effect": "Allow" } ] }
Click on “Review Policy”
Call it “PassRole” and then click “Create Policy”
This role should be immediately available to your Rosco instance.
Configuring AWS networks
In addition to providing permissions for Rosco’s packer to do Bake stages, if your AWS account doesn’t have a default VPC / subnet, you can specify a default subnet to bake in with this:
hal config provider aws bakery edit --aws-subnet subnet-0123456789
hal deploy apply
Or, on a per-bake basis, you can specify what VPC and subnet to bake by adding these two extended attributes to the Bake stage:
aws_vpc_id
:vpc-123456
aws_subnet_id
:subnet-0123456789
What exactly are Packer scripts?
Spinnaker works best when deploying immutable artifacts to immutable machine images. When working with machine images, packer scripts are used during the Bake Stage to create an immutable machine image.
Spinnaker has a microservice called Rosco, which uses Packer to bake machine images (such as Amazon Machine Images or AMIs). Out of the box, it comes with the packer templates and scripts listed here.
By default, Rosco performs the following actions in a “Bake” stage:
- Takes a list of desired packages specified in the pipeline definition
- Identifies the
deb
files produced by your CI pipeline and matches those to the desired package - Creates a set of Packer variables using the name/repository of the matched
deb
file(s) - Bakes an AMI using the packer template (visible here
- Runs the
install_packages.sh
script (visible here) to install the identifieddeb
packages into the AMI - Make the AMI ID available to later stages in the Spinnaker pipeline (such as deployments)
In a bake stage configuration, you can specify other packer templates to use.
If your app is using zip, tarballs or you need some customization, you need to create a new Packer script (see below).
Adding custom Packer scripts to Armory
Out of the box, Armory comes with these built-in Packer templates and scripts: https://github.com/spinnaker/rosco/tree/master/rosco-web/config/packer
If you’d like to add additional Packer template or script files, you can add them via the Armory Operator .
Add any Packer template and supporting scripts as string-formatted entries under the spec.spinnakerConfig.files
section of the SpinnakerService
config.
If you have a template named example-packer-config.json containing the following:
{
"packerSetting" : "someValue"
}
and a script file named my-custom-script.sh containing the following:
#!/bash/bash -e
echo "Hello world!"
make the following entries into the SpinnakerService
config:
files:
profiles__rosco__packer__example-packer-config.json: |
{
"packerSetting" : "someValue"
}
profiles__rosco__packer__my-custom-script.sh: |
#!/bash/bash -e
echo "Hello world!"
NOTE: The Armory Operator interprets the double underscores in the file names as slashes indicating the directory path where the Operator saves the file.
See the “Export Packer template files” step in the Migrating from Halyard to Operator section of the Armory Operator document for more examples.
Feedback
Was this page helpful?
Thank you for letting us know!
Sorry to hear that. Please tell us how we can improve.
Last modified December 9, 2022: (77a2e500)