Automation and Infrastructure as Code
Automation centers on ROS (Resource Orchestration Service), Cloud Assistant, and the OpenAPI/CLI. ROS uses templates (JSON/YAML) to define a stack of resources with dependencies, parameters, and outputs. A candidate must understand stack lifecycle (create, update, delete), that updates can be non-destructive or replacement depending on resource changes, and that ROS can roll back on failure. Terraform and the Alibaba Cloud provider are also common.
Cloud Assistant runs commands or scripts on ECS instances without SSH; it requires the agent and is useful for batch operations, patching, and scheduled tasks. The CLI and SDKs call OpenAPI actions; understand that API calls are authenticated with AccessKey and that regions/endpoints matter.
Traps: assuming a stack update always replaces resources (some changes update in place, some force replacement); forgetting Cloud Assistant needs the agent and a supported OS; confusing user data (runs once at first boot) with Cloud Assistant (repeatable). Also, API rate limits and eventual consistency can cause transient errors - retries with backoff are expected.
Rule of thumb: "repeatable, idempotent, versioned" → ROS/Terraform; "run a one-off or scheduled command across many instances" → Cloud Assistant.
Worked example: A template change to an instance's image forces replacement, destroying the instance; use a rolling update strategy or snapshot first.