Building data-driven APIs at the edge

As presented at Serverless in the Park on 05 May 2022

1. Building data-driven APIs at the edge Rob Sutter (@rts_rob) Serverless in the Park 2022
2. What you will learn • What are APIs at the edge? • How are edge functions di ff erent? • When should you use them? • How to choose a data store? • A note about authentication. Rob Sutter (@rts_rob)
3. Introduction
4. Who am I? • Head of DA at Fauna • Former AWS Serverless DA • Previous SaaS startup founder • Global active-active architecture • Mobile and web users Rob Sutter (@rts_rob)
5. What are APIs at the edge? Rob Sutter (@rts_rob)
6. What are APIs at the edge? Rob Sutter (@rts_rob)
7. Architecture
8. Classic serverless functions Example - AWS Lambda • Designed for single-region workloads • Isolated as Firecracker virtual machines • Long running - currently up to 15 minutes (wall time) • Big environments: • 10,240 MB of memory • Up to six vCPUs • Up to 10GB code with container image packaging format Rob Sutter (@rts_rob)
9. Edge functions Example - Cloud fl are Workers • Originally designed for inbound HTTP requests at the edge • V8 isolates form the security boundary • Short-lived - up to 50ms of CPU time for Bundled Usage Model • Small environments: • Up to 128MB of memory • Shared processes in a single VM • 1MB code limit Rob Sutter (@rts_rob)
10. Other changes at the edge • API Gateway is typically built in • Typically only JavaScript • Also WASM, so you can get Rust - kinda • Fewer related services
11. Use cases
12. What are good use cases? Rob Sutter (@rts_rob)
13. What use cases are less appropriate? Rob Sutter (@rts_rob)
14. What are the advantages? Rob Sutter (@rts_rob)
15. What are the disadvantages? Rob Sutter (@rts_rob)
16. Data stores
17. Choosing a data store • Single-region data stores • Multi-region data stores • Edge data stores
18. Single-region data stores Rob Sutter (@rts_rob)
19. Single-region data stores Advantages • Simpler to provision • Simpler to operate • Colocated with single-region assets Disadvantages • Latency for distant workloads Rob Sutter (@rts_rob)
20. Single-region data stores Rob Sutter (@rts_rob) Region Location Distance (km) Speed of Light (ms) Observed latency (ms) 1 Des Moines, IA 0 0 0 2 Washington, DC 1600 5.337 16 3 New York, NY 1900 6.338 14 4 Charleston, SC 1920 6.404 21 5 Dallas, TX 1200 4.003 15 6 Denver, CO 1100 3.669 9 7 Los Angeles, CA 2800 9.340 16 8 San Francisco, CA 2900 9.673 23 9 Seattle, WA 2800 9.340 24 Simple Mean 1802 6.011 15.333
21. Single-region data stores • AWS Aurora • MongoDB • PlanetScale • PostgreSQL • Xata Rob Sutter (@rts_rob)
22. Multi-region data stores Rob Sutter (@rts_rob)
23. Multi-region data stores Advantages • Balanced latency • Transactions available • Various consistency models Disadvantages • Replication cost • Replication complexity • Maintaining consistency • Transactions (again) Rob Sutter (@rts_rob)
24. Multi-region data stores • DynamoDB Global Tables • Fauna • Cloud fl are Workers Durable Objects Rob Sutter (@rts_rob)
25. Edge data stores Rob Sutter (@rts_rob)
26. Edge data stores Advantages • Lowest latency • Single programming model • Integrated security model Disadvantages • Eventual consistency • Lack of transactions • Data visualization tooling Rob Sutter (@rts_rob)
27. Edge data stores • Cloud fl are Workers KV Rob Sutter (@rts_rob)
28. Authentication
29. Authentication at the backend Rob Sutter (@rts_rob)
30. Authentication at the edge Rob Sutter (@rts_rob)
31. Distributed protection Rob Sutter (@rts_rob)
32. Review • What are APIs at the edge? • How are edge functions di ff erent? • When should you use them? • How to choose a data store? • A note about authentication. Rob Sutter (@rts_rob)
33. Rob Sutter (@rts_rob) Thank you!