Operations, cost, performance, and sustainability

A design is not finished until teams can see, run, and afford it.

Observability

Collect metrics, logs, traces, and business outcomes. CloudWatch provides metrics, logs, alarms, dashboards, and synthetics. X-Ray or OpenTelemetry traces requests across services. CloudTrail answers who changed AWS resources.

Alarm on symptoms users feel and on causes operators can act on. Every critical alarm needs an owner and a runbook.

Infrastructure and change

Use CloudFormation, CDK, Terraform, or another governed infrastructure-as-code tool. Review changes, scan them, deploy through environments, and detect drift. Use Systems Manager for fleet operations and Parameter Store where appropriate.

Cost

Tag and account-separate ownership. Use Cost Explorer, Budgets, Cost and Usage Reports, Savings Plans, Spot capacity, and Compute Optimizer. Track unit cost, such as cost per order, not only the monthly total.

The biggest wins usually come from architecture: deleting idle resources, right-sizing, using storage lifecycle policies, scheduling non-production, choosing efficient data transfer paths, and matching purchase models to demand.

Performance

Test with realistic load. Select the right compute, database, storage, and networking. Cache only with a clear freshness and invalidation policy. Use CloudFront close to global users and asynchronous work when a response need not wait.

Sustainability

Use fewer resources for the same outcome: scale with demand, eliminate idle capacity, use managed services, efficient processors such as Graviton where compatible, efficient code, sensible data retention, and Regions that meet both business and sustainability requirements.