[mercari GEARS 2025] Mercari’s CDN Migration from Fastly to Cloudflare

Transcript

1. Mercari’s CDN Migration from Fastly to Cloudflare
 Hatappi
 Mercari / Software Engineer

2. Hatappi
 
 
 • Joined Mercari in 2019
 • Led CDN migration from Fastly to Cloudflare on Platform Network team
 • Now SRE in Cross-border team,
 driving global expansion
 Software Engineer

3. Key Takeaways for Today
 1. Motivation for migrating to Cloudflare
 2. How we approached the Migration
 3. How development experience changed

4. Agenda
 1. Why we migrated to Cloudflare
 2. How we migrated
 3. Building toward CDN as a Service
 4. How developer experience changed

5. Why we migrated
 to Cloudflare

6. CDN Operations Before Migration
 Change
 Request
 Apply
 Fastly
 Developers
 Platform Network
 / SRE

7. Key Points in Adopting Cloudflare
 ⚙ Reduced Operation Costs
 • Most features can be configured via API
 • Preliminary research showed APIs could cover nearly all use cases
 🔒 Enhanced Security
 • DDoS Protection, WAF, Rate Limiting, etc.
 🌐 Toward CDN as a Service
 • Self-service enables rapid feature utilization 
 and supports business expansion

8. How we migrated

9. Migration steps
 🔍 Step1: Assessment
 • Analyze features implemented
 in Fastly
 • Research migration methods
 to Cloudflare
 ⚙ Step2: Implementation
 • Implement required features based on investigation
 • Standardize reusable features
 🧪 Step3: Testing
 • Verify functionality
 • Identify gaps and issues
 🚀 Step4: Migration
 • Gradually migrate traffic

10. 🔍 Step 1: Assessment
 Examples
 • Vary header
 ◦ Fastly: Supported
 ◦ Cloudflare: Only supported for image (as of Sep 2025)
 • Default cache TTL on CDN
 ◦ Fastly: Supported
 ◦ Cloudflare: Not configurable (as of Sep 2025)
 There are differences between Fastly and Cloudflare. In particular, cache-related behavior requires careful attention.
11. 🔍 Step 1: Assessment
 Examples
 • Vary header
 ◦ Use Cache Keys to achieve the same purpose
 • Setting default cache TTL on CDN
 ◦ Reconfirm necessity
 ◦ For content under specific conditions, set TTL with Cache Rule.
 Cloudflare Workers is powerful, but check for simpler solutions first.
12. ⚙ Step2: Implementation
 1. Use Cloudflare Terraform provider
 2. Build reusable Terraform module (CDN Kit)
 ◦ Implement using standardized methods,
 not by directly modifying Cloudflare resources
 3. Prioritize migration over new features
 ◦ Because many services need migration,
 adding new features would slow down progress

13. ⚙ Step2: Implementation
 • Auto-generate CDN Kit settings
 from Fastly settings
 • Highlight VCL code requiring migration review
 Building a Migration Support Tool
14. 🧪 Step3: Testing
 Use the existing Fastly service as the baseline
15. 🚀 Step4: Migration
 • Gradual traffic migration
 ◦ limit potential impact of issues
 • Smooth rollback
 ◦ quickly revert if issues occur, minimize business impact
 To ensure safe and smooth migration
16. 🚀 Step4: Migration
 • Adjust DNS weights
 to shift traffic step by step
 • Short TTLs ensure quick reflection
 of changes
 • Rollback is simple by just reverting the weight
 DNS Weighted Routing for Gradual Migration
17. 🚀 Step4: Migration
 • Track error rates
 at each traffic shift
 to ensure safe migration
 • Monitor Cache Hit Rate, Bandwidth, and other metrics
 Monitoring During Migration
18. Migrated 100% 🎉
 What’s Next?

19. Building toward 
 CDN as a Service

20. Building CDN as a Service
 CDN Kit
 Mercari standardized 
 Terraform module
 for CDN configuration
 CF IAM
 Automated permission management for Cloudflare Dashboard

21. CDN Kit
 🔧 Provide a unified Terraform module for Cloudflare at Mercari
 🎯 Simplify configuration to reduce developer burden
 ✨ With minimal configuration, developers can:
 • Configure DNS, WAF, and other resources
 • Export access logs seamlessly to BigQuery
 • Auto-generate Datadog monitors
 Mercari standardized Terraform module for CDN configuration
22. CDN Kit
 Change
 Request
 Apply
 Fastly
 Developers
 Platform Network
 / SRE
 Before
23. CDN Kit
 Update their own zones anytime
 Cloudflare
 Developers /
 SRE
 Platform Network
 After CDN Kit
 Implements common settings and best practices

24. CDN Kit
 Example
25. CDN Kit
 
 • CDN Kit manages zone-scoped resources (e.g. Cache Rules)
 • To use account-scoped resources (e.g. Cloudflare Workers)
 with CDN Kit, we needed to figure out how to manage them
 Challenge: Handling account-scoped resources in CDN Kit Solution: Prefix resource names with the zone name to treat them as zone-scoped resources

26. CF IAM
 
 
 Automating Permission Management for Cloudflare Dashboard 🧭 Provide a more intuitive workflow than raw log analysis 
 🔐 Enable controlled access for incident investigations
 🚦 Detect and verify blocked traffic to ensure legitimate requests

27. CF IAM
 
 
 • Grant permissions per zone, since each has its specific owning team
 • Set default access to Read Only, as most changes go through CDN Kit
 Problem: Manual Permission Management Manual permission assignment is time-consuming and error-prone!!

28. CF IAM
 
 
 CDN Kit settings are defined under each service • Each service defines its owner team, mapped to its Cloudflare zones
 • Team members are managed
 in the same repository 
 via Team Kit, the Terraform module
 ├── service-a
 │ ├── development
 │ │ └── cdn
 │ │ └── a.example.com
 │ │ └── cdn_kit.tf
 │ └── production
 │ └── cdn
 │ └── b.example.com
 │ └── cdn_kit.tf
 └── service-b
 └── production
 └── cdn
 └── c.example.com
 └── cdn_kit.tf

29. CF IAM
 
 
 We already have all the necessary information…
 Looks like we can automate it !!

30. • Detect changes in CDN Kit 
 / Team Kit
 • Trigger GitHub Actions 
 to update permissions
 • Restore correct state if permissions are modified manually
 CF IAM
 
 

31. How developer experience changed

32. How Developer Experience Changed
 
 • 70%+ PRs released without Platform Network team involvement
 • Most PRs merged within 1 day
 ◦ Previously it took 2–5 business days
 • Developers investigate incidents using Cloudflare Dashboard
 without involving the Platform Network team
 • Inquiries evolved into higher-value discussions
 ◦ e.g. Cloudflare utilization & design

33. Conclusion

34. Conclusion
 
 
 🌐 Migrated to Cloudflare and standardized CDN configuration
 • Built internal Terraform module (CDN Kit)
 • Automated dashboard permission management
 ⚙ Realized CDN as a Service
 • Enabled developers to autonomously manage CDN
 󰠁 Developer experience improved
 ◦ 70%+ PRs released without Platform Network involvement
 ◦ Inquiries shifted to higher-value discussions
 ◦ Developers can independently investigate incidents

35. Thank You!