Infrastructure as Code

Infrastructure that exists only in the console cannot be reviewed, cannot be rebuilt, and cannot be explained six months later. This topic replaces that with a file: a template that describes the end state, a service that reaches it, and a workflow that shows you what a change will do before it does it. It is the largest skill in the deployment domain, and it is where most of the domain's scenario questions live.

What This Topic Covers

• What a CloudFormation template contains: the template sections, parameters with real guardrails, intrinsic functions, the dependency graph CloudFormation builds from your references, and the resource attributes that decide what survives a delete
• Sharing values between stacks with exports, Fn::GetStackOutput, and nested stacks, and keeping credentials out of templates with dynamic references
• Operating a running stack: the three update behaviors, change sets, stack policies, signals and rolling Auto Scaling updates, rollback options and rollback triggers, drift detection, and importing existing resources
• A method for failed deployments: reading stack events from the first failure, mapping a stack status to the recovery it allows, and diagnosing permission errors, subnet address exhaustion, service quotas, missing signals, and rollbacks that themselves fail
• The AWS CDK: construct levels, apps and stacks, bootstrapping, the synth-diff-deploy workflow, and what changes for an operator when the template is generated rather than written
• Terraform for CloudOps: state and why it needs a locked remote backend on AWS, the write-plan-apply workflow, how it differs from CloudFormation on rollback and drift, and the Git branch-and-pull-request practice that makes any of these tools safe in a team

Why It Matters

Skill 3.1.2 and 3.1.3 of the SOA-C03 exam guide cover creating resources with CloudFormation and the CDK and remediating deployment issues, and skill 3.1.6 names Terraform and Git directly. Questions in this area are rarely definitional. They give you a stack in ROLLBACK_COMPLETE, an update that silently replaced a database, a wait condition that timed out, or a subnet that ran out of addresses, and ask for the next action.

On the job the same skills decide how much a bad afternoon costs. An engineer who reads a change set before executing it does not lose a production database to a one-word property change, and an engineer who knows which stack statuses permit which operations recovers a broken deployment in minutes instead of escalating it.