The Well-Architected Framework is a set of best practices which form a repeatable process for designing solution architecture, to help identify potential issues and optimize workloads.
Figure: The Well-Architected Framework includes the five pillars of architectural excellence. Surrounding the Well-Architected Framework are six supporting elements
There are trade-offs to be made between these pillars. E.g. improving reliability by adding Azure regions and backup points will increase the cost.
Thinking about architecting workloads can be hard – you need to think about many different issues and trade-offs, with varying contexts between them. WAF gives you a consistent process for approaching this to make sure nothing gets missed and all the variables are considered.
Just like Agile, this is intended to be applied for continuous improvement throughout development and not just an initial step when starting a new project. It is less about architecting the perfect workload and more about maintaining a well-architected state and an understanding of optimizations that could be implemented.
Assess your workload against the 5 Pillars of WAF with the Microsoft Azure Well-Architected Review and add any recommendations from the assessment results to your backlog.
Figure: Some recommendations will be checked, others go to the backlog so the Product Owner can prioritize
Figure: Recommended actions results show things to be improved
✅ Figure: Good example - WAF is very visible to the Product Owner on the backlog
One review at the end finds problems when it is too late to fix them. Do it at least twice:
For an AI project, also do the Azure Well-Architected AI workload assessment.
| Sprint | What happens |
|---|---|
| 12 | First Well-Architected Review, one week before go-live. It finds "single region, no failover". Go-live goes ahead. There is no time to change the design |
❌ Figure: Bad example - The only review is at the end, so the findings come too late
| Sprint | What happens |
|---|---|
| 1 | Review of the design. Milestone "Design". "Single region, no failover" goes on the backlog as a PBI |
| 12 | Go-Live review. Milestone "Before go-live". The score is compared with Sprint 1 |
✅ Figure: Good example - Reviewed at the start and before go-live, with a milestone each time