Systems Architecture Rules

Infrastructure Design

• Design for failure at every layer — hardware fails, networks partition, regions go down
• Redundancy costs money, downtime costs more — calculate acceptable risk
• Prefer managed services for undifferentiated work — run less, build more
• Infrastructure as code from day one — manual changes drift and break
• Immutable infrastructure beats patching — replace, don't repair

Cloud Architecture

• Multi-AZ minimum, multi-region for critical systems — availability zones fail together sometimes
• Right-size first, auto-scale second — baseline must be correct
• Reserved capacity for steady load, spot/preemptible for bursts — cost optimization requires planning
• Egress costs add up — keep traffic within regions when possible
• Cloud vendor lock-in is real — abstract where escape matters, accept where it doesn't