Blurt

since until

Spawning Google Cloud Infrastructures with Deployment Manager

Like Amazon’s CloudFormation, Google has a solution for the devops fellas and gals that need to spawn cloud infrastructures at the press of a button.

The beauty of AWS’s CloudFormation is that it allows one to treat a infrastructure setup as a deliverable pretty much in a similar way we consider source-code a deliverable. Heck, it even allows one to keep track of infrastructure by checking in templates into your version management tool of choice (probably git) and managing changes to infrastructures.

Installing the CLI Tool

The gcloud tool installs some components by default, but as Google Cloud Deployment Manager is still in alpha, as of Jebruari 2015, one will need to explicitly install the preview component.

Get a proper overview of all components by listing them:

gcloud components list

Install the developer preview by running the following command and following the instructions.

gcloud components update preview

The Basics

The API of the Deployment Manager is currently at version 2 as of the 27th of Januari 2015. So instead of the deployment-manager command group, the dm-v2 group will be used in this post.

The deployment manager knows configuration files, templates, manifests and deployments.

A deployment is a collection of resources that one uses to spawn an infrastructure. Which means, all machines, buckets, databases, networks, load balancers… the whole shebang of possible resources that you need.

Deployments are only possible if we there is a configuration. The configuration basically contains a description of that goodness one needs spawned. To allow one to structure configurations in a manner that is more pleasant to the DRY writers, one could use templates within configurations by simply importing and referencing them.

Manifests are expansions of configurations, and the templates imported within, that represent the configuration with all variables expanded exactly as it will be executed.

Configuration

Imagine the example-conf.yaml file looking like this:

resources:
- name: genesis
  type: compute.v1.instance
  properties:
    zone: europe-west1-c
    machineType: zones/europe-west1-c/f1-micro
    disks:
    - deviceName: boot
      type: PERSISTENT
      boot: true
      autoDelete: true
      initializeParams:
        diskName: genesis-bootdisk
        sourceImage: https://www.googleapis.com/compute/v1/projects/coreos-cloud/global/images/coreos-stable-522-5-0-v20150114
      networkInterfaces:
      - network: global/networks/default
- name: bina-docker-images
  type: storage.v1.bucket

Basically we have a file containing two resources, a f1-micro VM instance and a storage bucket. As a result we can spawn this setup by simply submitting the configuration description to the deployment manager:

gcloud preview dm-v2 deployments list \
  --config example-conf.yaml \
  --deployment blog_post_demo \
  --project bina

Templates

The former example may already simplify workflows enough as is, but somehow complex infrastructures would be quite an exercise in repeating oneself unless there is a way to reuse predefined components.

The former resource set could be extracted into a template which will be named simple_vm.jinja for the sake of demonstration. This file will simply describe a VM instance with pretty much the same properties as the genesis machine demonstrated in our previous example.

resources:
- name: 
  type: compute.v1.instance
  properties:
    zone: europe-west1-c
    machineType: zones/europe-west1-c/
    disks:
    - deviceName: boot
      type: PERSISTENT
      boot: true
      autoDelete: true
      initializeParams:
        diskName: -bootdisk
        sourceImage: https://www.googleapis.com/compute/v1/projects/coreos-cloud/global/images/coreos-stable-522-5-0-v20150114
      networkInterfaces:
      - network: global/networks/default

Observe how we use env["name"] and properties["machineType"] to allow reuse of the same template whilst providing the flexibility of specifying just the machineType and name.

The env["name"] attribute refers to the name of the resource as specified in the configuration file. In the following example this attribute will change as such to provide the VM’s with names held by some remarkable bot specimen including Optimus Prime, Ultra Magnus, Sentinel Prime and Megatron.

The created disk assigned to the different machines will also be named to bear the name of the VM it belongs to followed by the appropriate suffix (-bootdisk) which will leave us with the bootdisk optimus-bootdisk for the VM named optimus.

The previously described template may be reused by referencing it in the configuration file.

imports: ["simple_vm.jinja"]
resources:
- name: optimus
  type: simple_vm_template.jinja
  properties:
    machineType: n1-highmem-16
- name: megatron
  type: simple_vm_template.jinja
  properties:
    machineType: n1-highcpu-16
- name: ultra
  type: simple_vm_template.jinja
  properties:
    machineType: f1-micro
- name: sentinel
  type: simple_vm_template.jinja
  properties:
    machineType: f1-micro
- name: bina-store
  type: storage.v1.bucket

For the sake of demonstration, the bucket has been left in the configuration file just to demonstrate that our template is used as just another type in our configuration file. One can use Google Cloud’s types and out template types interchangeably.

Notice that the zone can easily be specified in the configuration file as a property and referred to from the template as could many other properties.

Sidenotes

  • When using templates ensure the filename ends with a clear extension .jinja or .py. I have tried filenames ending in .jinja.yaml in order to help vi in determining the syntax for highlighting, but that only confused the deployment manager.