Build Images for Google Compute Platform

Requirements:

• Packer
• Google Compute Packer plugin
• Google Cloud SDK
• A Vanilla AMI
• Packer templates

The suggested approach is based on using Packer templates to customize the deployment and automate the upload and publish to GCP of cOS derivatives or cOS itself. For all the details and possibilties of Packer check the official documentation.

There are no cOS Vanilla images publicly available in GCP, however they can be easily build or downloaded and published to your working GCP project. See Build Raw Images and Importing a Google Cloud image manually to see how to upload a Vanilla image in your project.

Run the build with Packer

Publishing an image in GCP based on top of the latest cOS Vanilla image is fairly simple. In fact, it is only needed to set the User Application Default Credentials for GCP and the GCP project ID and then run a packer build process to trigger the deployment and register the resulting snapshot as an image. In such case the lastest cOS image will be deployed and configured with pure defaults. Consider:

# From the root of a cOS-toolkit repository checkout

> export GCP_PROJECT_ID=<your_gcp_project_id>

> cd packer
> packer build -only gcp.cos .

Packer authenticates automatically if the User Application Default Credentials are properly set in the host.

The -only gcp.cos flag is just to tell packer which of the sources to make use for the build. Note the packer/images.json.pkr.hcl file defines few other sources such as qemu, virtualbox and amazon-ebs.

Customize the build with a variables file

The packer template can be customized with the variables defined in packer/variables.pkr.hcl. These are the variables that can be set on run time using the -var key=value or -var-file=path flags. The variable file can be a json file including desired varibles. Consider the following example:

# From the packer folder of the cOS-toolkit repository checkout

> cat << EOF > test.json
{
    "gcp_project_id": "<your_gcp_project>"
    "gcp_cos_deploy_args": "elemental reset --docker-image <my-custom-image>",
    "gcp_disk_size": 20,
    "name": "MyTest"
}
EOF

> packer build -only gcp.cos -var-file=test.json .

The above example runs the cOS Vanilla image on a 20GB disk and calls the command elemental reset to deploy the main OS, once deployed an snapshot is created and published as an image in Google Compute Engine. The created artifact will be called MyTest, the name has no impact in the underlaying OS.

Available variables for customization

All the customizable variables are listed in packer/variables.pkr.hcl, variables with the aws_ prefix are the ones related to the AWS Packer template. These are some of the relevant ones:

• gcp_project_id: The project ID that will be used to launch instances and store images.

• gcp_cos_deploy_args: This the command that will be executed once the Vanilla image booted. In this stage it is expected that user sets a command to install the desired cOS or derivative image. By default it is set to elemental reset which will deploy the cOS image from the recovery partition. To deploy custom derivatives something like elemental reset --docker-image <my-derivative-img-ref> should be sufficient.

• gcp_disk_size: This sets the disk size of the VM that Packer launches for the build. During Vanilla image first boot the system will expand to the disk geometry. The layout is configurable with the user-data.

• gcp_user_data_file: This sets the user-data file that will be used for the aws instance during the build process. It defaults to aws/setup-disk.yaml and the defauklt file basically includes the disk expansion configuration. It adds a COS_STATE partition that should be big enough to store about three times the size of the image to deploy. Then it also creates a COS_PERSISTENT partition with all the rest of the available space in disk.

• gcp_source_image_family: This the family to choose the image for the build process. It defaults to cos-vanilla to pick the latest cOS Vanilla image available. Note Packer tries to find the image family first in the given working project (gcp_project_id).