Day 5 - Week 5 Review: Global Architecture Integrated Scenarios

Reflecting on Week 5's 5 days, one axis runs through: How to route global users to correct region/endpoint, auto or manually switch on failure, all within budget. DR strategies (Day 1), Route 53 7 routing types (Day 2), CloudFront advanced (Day 3), Global Accelerator, Route 53 Health Check/DNSSEC/Resolver (Day 4). These five layers learned independently but exam combines them into single scenarios.

Today isn't repetition. Compress each concept's decision tree, mathematically organize most-frequent trade-off comparison mistakes, finish with 12 complex enterprise scenarios.

DR Strategy Integration: Aurora Global RPO/RTO Mathematics

Revisit DR strategy numbers. Understanding "why this number" matters beyond memorization.

Aurora Global Database Mathematics

Aurora Global Database positions between Pilot Light and Warm Standby as DR tool.

RPO: Aurora Global uses async replication, Replication Lag typically <1 second. Thus RPO ~1 second. Large write storms (e.g., batch inserts millions) can increase Lag to seconds/tens of seconds.

RTO: Managed Planned Failover vs Unplanned Failover differences exist.

• Managed Planned Failover (planned switch): Aurora fully syncs Secondary with Primary before promotion. RTO typically under 30 seconds on supported versions. Zero data loss.
• Unplanned Failover (Primary region failure): Manual Secondary promotion needed. promote-read-replica-db-cluster API or console. RTO typically a few minutes (DB promotion) + app reconnection.

# Aurora Global Unplanned Failover: promote Secondary to new Primary
aws rds promote-read-replica-db-cluster \
  --db-cluster-identifier secondary-cluster-ap-northeast-2 \
  --region ap-northeast-2
 
# After promotion, change app connection string to Secondary cluster endpoint

Replication Lag Monitoring: Use AuroraGlobalDBReplicationLag CloudWatch metric as DR decision basis. Can condition R53 ARC Safety Rule: "Lag > 5 seconds, forbid failover."

💡 Related Theory: Aurora Global Database async replication based on WAL (Write-Ahead Logging). Aurora compatible PostgreSQL streams WAL segments to Secondary region. Streaming occurs directly at Aurora Storage Layer so overhead lower and Lag shorter than DB engine-level replication (MySQL binlog-based). Aurora Global Database's <1 second Lag superior to general RDS Cross-Region Read Replica (seconds/tens of seconds).

R53 ARC Safety Rule: Auto Failover Safety Guards

Auto failover fast but risky. Real incident: "Auto failover while DB Replication Lag large loses 5 minutes data." R53 ARC prevents via manual switch + SafetyRule.

R53 ARC SafetyRule Example:
MUST_EXIST: After switch, minimum 1 cell must stay ACTIVE
           → prevent complete traffic blackout

ASSERTION: Allow switch only when Aurora Lag < 3 seconds
          → control data loss threshold

SafetyRule two types: Gating Control + Assertion Control:

• Gating Control: prerequisite evaluated before other Routing Control on/off
• Assertion Control: quantity-based conditions guaranteeing min/max ACTIVE state

📚 Case Study: Amazon internal systems origin. 2019 Amazon Shopping Cart service auto-failover caused Secondary data inconsistency. After, Amazon adopted "human confirms DB Lag, switch manually when safe" principle, becoming R53 ARC design philosophy. ARC product of "automation isn't always correct" lesson.

CloudFront + Global Accelerator Selection Formula

Final organization of two most-confused services.

Selection Algorithm:
1. UDP protocol?
   YES → AGA (CloudFront doesn't support UDP)

2. Need 2 static IPs? (firewall whitelist, BYOIP)
   YES → AGA

3. Caching critical? (static content, Cache Hit Rate > 50%)
   YES → CloudFront

4. Need WAF / Signed URL / Signed Cookies / FLE / Lambda@Edge?
   YES → CloudFront

5. Need immediate failover bypassing DNS cache?
   YES → AGA (bypass Route 53 DNS cache, packet-level switch)

6. Basic HTTP API, no caching/static IP needed?
   → Route 53 LBR cheapest

Cost Comparison (100GB traffic/month):

• Route 53 LBR: $0.0000004 per query → 1M queries = $0.40
• CloudFront: Data Transfer Out $0.085/GB × 100GB = $8.50 (add Origin fees without cache)
• AGA: fixed $18/month + $0.015~0.035/GB × 100GB = $20~$21.50

Small traffic makes AGA $18 fixed cost burden, but static IP + packet-level switch requirement leaves no choice.

🔍 Deeper Dive: CloudFront + AGA combo pattern. When both static IP (AGA Anycast) + HTTP caching (CloudFront) needed, place AGA frontend forwarding to CloudFront Origin. Whitelist AGA IP at corporate firewall, CloudFront handles WAF/caching/Lambda@Edge. Hybrid SaaS pattern needing both "static IP enterprise customers + public CDN."

Route 53 Health Check Comprehensive Summary

Failover routing: Primary Health Check becoming Unhealthy switches all queries to Secondary. Must guarantee "Secondary also Healthy." Attach Health Check to Secondary too; if Secondary Unhealthy, Route 53 won't switch and returns Primary (lesser evil).

⚠️ Pitfall Lesson: If Route 53 feels slow to switch due to DNS cache, sum Health Check detection time (default 30-second interval, 3 consecutive failures = 90 seconds) + DNS TTL. Default adds up to 90 + 300 = 390 seconds (6.5 minutes). Fast switching needs Health Check interval 10 seconds (Fast) and TTL ≤60 seconds.

Savings Plans Cost Calculation Integrated Review

DR and global architecture cost optimization belongs SAP domain 4 (continuous improvement).

Compute Savings Plans vs EC2 Instance Savings Plans:

DR environments risking region changes = Compute Savings Plans safer. Switching us-east-1 to us-west-2 on failover still applies Compute Savings Plans.

Cost Calculation Example:

Current: m5.xlarge On-Demand 10 × $0.192/hour × 720 hours = $1,382/month

Compute Savings Plans 1 year:
30% discount → $0.134/hour
Savings: $1,382 - ($0.134 × 10 × 720) = $1,382 - $965 = $417/month saved

Compute Savings Plans 3 years:
50% discount → $0.096/hour
Savings: $1,382 - ($0.096 × 10 × 720) = $1,382 - $691 = $691/month saved

💡 Related Theory: Savings Plans structure similar to finance Futures Contracts. Commit future computing use at fixed price, receive discount. Spot Instance opposite: "cheaply buy spare capacity at spot market." Optimal cost combining both — Savings Plans cover base load, Spot handles bursts.

Well-Architected 6 Pillar Checklist: Global Architecture Application

🔍 Deeper Dive: 2022 AWS Well-Architected adds 6th Pillar "Sustainability." SAP-C02 explicitly addresses this. Global architecture sustainability optimization core: "don't operate more regions than needed." If Active-Active exceeds RTO/RPO requirements, reducing to Warm Standby lowers resource usage and aligns with sustainability.

Key Comparison Matrix Final Edition

DR Strategy Decision Tree

RTO > 4 hours, RPO > 1 hour?
  YES → Backup & Restore (cheapest)

RTO 30 min to 4 hours, RPO minutes?
  "Only replicate DB, quick app start via AMI"
  YES → Pilot Light

RTO minutes to 30 min?
  "Full stack must always run at small scale"
  YES → Warm Standby

RTO ~0, simultaneous serving?
  YES → Active-Active (most expensive)

Route 53 + CloudFront + AGA Integrated Decision Tree

Route users to correct region?
  Data sovereignty enforce → Geolocation
  Performance optimize → Latency-Based Routing
  Distance + traffic adjust → Geoproximity (Traffic Flow)
  DR Primary/Secondary → Failover

Endpoint acceleration needed additionally?
  HTTP + cache + WAF → CloudFront
  UDP / static IP / DNS-bypass failover → AGA
  Both → AGA frontend + CloudFront backend

Failover within region?
  HTTP error-based instant retry → CloudFront Origin Failover
  Health Check-based region switch → Route 53 / AGA
  Manual safe switch → R53 ARC Routing Control