This site uses cookies and by using the site you are consenting to this. We utilize cookies to optimize our brand’s web presence and website experience. To learn more about cookies, click here to read our privacy statement.

Cloud Infrastructure Pipelines – Part 2

Author: John Merline Posted In: Cloud, DevOps

Part 1 suggested pipelines with two-stage templating. But where do the inputs for the first stage come from? A file generated with “fill-in-the-blanks” still requires manual edits in the first templating stage. To explore the tradeoffs of further templating, a quick detour to the world of fashion may help.


When discussing infrastructure automation, it is common to talk of tailoring environments:

  1. An “off the rack” product or service offers only in predetermined sizes.
  2. The “made to measure” approach modifies a blueprint or template based on a handful of parameters.
  3. A “custom” solution measures every parameter to fully tailor the solution.
  4. The term “bespoke” (British for “spoken for”) implies a more hand construction and less automation.

The equivalent for cloud automation might be:

  1. Selecting the App Service Plan when using the Azure WebApp determines the size and performance of the environment.  
  2. Using parameters with AWS Cloud Formation or Azure Resource Manager templates to tailor the environment.
  3. Creating Terraform or Ansible scripts that defines the infrastructure –  IP addressing, Load Balancers, Security Groups, Gateways, Instances, etc.
  4. Using the console to configure every option, maybe export a template of the environment to restore if necessary.

So what level of tailoring is practical at scale? Probably somewhere between “make to measure” and fully “custom”.  A pipeline that generates custom scripts from parameter inputs and templates seems the best trade-off. And it allows for ongoing customization (like new firewalls rules) with updates to the templates and regeneration.


Cloud Infrastructure Pipelines have multiple stages:

  • Populate takes inputs from various sources to create custom files and directory structure for the new project
  • Generate uses template languages and variable file inputs to create tailored scripting for the project
  • Build executes the generated code within each of the multiple subdirectories of the new project

Three Stage Pipelines

The inputs for the YAML variables might come from individual parameters, a text field, or service request. If the inputs are too specific, then it might appear easier to create a custom scripts but that does not scale. Instead, limit the inputs to questions like:

  • Public facing?
  • Sensitive data? 
  • Multiple regions or availability zones?
  • Multiple applications per infrastructure?

For resource sizing questions (networks, databases, etc.),  think T-shirts (Large, Medium, Small, …) and map those sizes to whatever is appropriate for the type of resource. Inputs can also be used to select global firewall rules or allow access to protocols like SSH or RDP. Bundle global rules into modules if the scripting language supports it. Updates to the modules can be propagated across the code base without regenerating scripts. Application specific firewall rules might still be handled with the same templating framework in a separate pipeline with its own service requests.

The second stage of templating combines the YAML variable files created by the first stage with Jinja templates to generate the code used to build the infrastructure. But items like resource names that are only calculated during the generate stage often need to query or save data to external systems. Depending on the templating language, there are options like filter plugins in Jinja or custom resources for Cloud Formation, that can invoke code to perform external actions. For example, assigning an IP address range to a dynamically generated subnet name.

Part 3 of this series covers managing state and validating the pipeline.