AWS Well-Architected review
Well-Architected is a framework for discussing architecture trade-offs, not a compliance certification or automatic guarantee. Begin with foundations and review one bounded workload with its owner.
Six pillars and evidence
| Pillar | Question | Useful evidence |
|---|---|---|
| Operational excellence | Can the team safely change and operate the workload? | Deployment rollback, ownership, runbooks, incident learning |
| Security | Who can access what, and how is misuse detected? | Identity/policy review, key ownership, audit events, response test |
| Reliability | Can the workload meet its recovery objectives? | Failure tests, restore drills, dependency limits |
| Performance efficiency | Does resource choice meet workload demand? | Representative load tests, p95/p99 latency, bottleneck analysis |
| Cost optimization | Are costs allocated and proportional to value? | Unit cost, metered dimensions, utilization, commitment risk |
| Sustainability | Is resource demand reduced without violating objectives? | Utilization, idle reduction, data retention, managed-resource efficiency |
Do not optimize one pillar in isolation. A single-AZ database may cost less but fail the stated availability requirement; longer retention can improve recovery while increasing storage and privacy exposure.
Review exercise
Consider a document-upload API with Lambda, S3 and SQS. Record confidentiality, maximum acceptable processing latency, durability, peak requests, RPO/RTO, operational ownership and cost-per-document target.
For each pillar, identify one risk, the evidence needed to assess it, and a proposed improvement. Examples: replay a duplicate queue event, simulate an unavailable dependency, restore a deleted test object, and measure processing lag during a burst. Assign an owner, acceptance test and review date to each change.
Choose the first three actions by impact and urgency, not by a count of best-practice checkboxes. Explain one rejected alternative and what requirement would change the choice.
Operations handoff
Store the resource inventory, architecture decisions, cost assumptions, alarms and restore procedure with the workload. Re-review after major changes or incidents. A successful deployment is not proof that access, rollback and recovery work.