Netflix: From Zero to Production-Ready in Minutes (QCon 2017)

Netflix: From Zero to Production-Ready in Minutes (QCon 2017)

1. 1. From Zero to Production-Ready in Minutes Tim Bozarth @timbozarth
2. 2. Dev Experience: Level up your Eng Effectiveness
3. 3. Agenda 1. It was the best of times... 2. Best practices made easy 3. Goodbye hand-written clients 4. From NIH to OSS
4. 4. It was the best of times… (ie: The story of the skeletons in our closet) 1
5. 5. About Netflix.. 100m+ members 1000+ developers 190+ countries 1/3 US download traffic 500+ microservices Over 100,000 VMs
6. 6. Runtime Platform Enable developers to productively create and integrate software in the Netflix ecosystem.
7. 7. Major Investments in Platform
8. 8. High Availability = Winning Moments of Truth
9. 9. High Availability =
10. 10. Challenges: Hard to take advantage of evolving best practices Owning client-side logic is complex and stressful Non-Java experience is hard
11. 11. Challenges:
12. 12. Productivity++ (availability is table stakes)
13. 13. Complexity is the mind killer.
14. 14. Runtime Platform Enable developers to productively create and integrate software in the Netflix ecosystem.
15. 15. Best-practices made easy (Better living through less complexity) 2
16. 16. Generators
17. 17. Generators What: Gives you a deployed app on the “paved road” in minutes.
18. 18. Generators Why: To make it easy to adopt, understand, and build production-ready apps.
19. 19. + Best Practices
20. 20. Historically: “Let’s go!”
21. 21. With Generators: “Let’s go!”
22. 22. + + + + + =
23. 23. But wait! There’s more! (Consistency)
24. 24. Components != PaaS
25. 25. Goodbye hand-written client libraries 3
26. 26. @Netflix every service owner is responsible for a client
27. 27. Clients defend themselves from failure (and the foundation to much of Netflix’s micro-service success)
28. 28. Your service Their service Your Client
29. 29. RPC Internals Platform Integration Serialization & Deserialization Bespoke business logic Your Client Your service Includes integration with Metrics, Caching, Discovery, Fallbacks, etc...
30. 30. Your Service RPC Internals Dependencies RPC Internals Platform Integration Serialization & Deserialization Bespoke business logic Server Logic Platform Integration Serialization & Deserialization Your Client Their Server
31. 31. Problems Server-API changes are a nightmare So much hand-written RPC-related code No cross-language client story
32. 32. These are solvable problems
33. 33. +
34. 34. 2 big wins: Code Generation New Abstraction Layer
35. 35. PROTO
36. 36. Your Service gRPC generated interfaces Depend encies gRPC Generated Client Bespoke business logic (please no) Server Logic Your Client Their Server Service Proto
37. 37. aching, Circuit-breake Fallbacks, Failure Injection, Discovery, equest-context-tracin etrics, Retries, Hedge
38. 38. Interceptors encapsulate common patterns (outside the user’s typical concern domain)
39. 39. Client Defense Examples: • Fallbacks • Advanced Caching • Retries • Failure Injection • Hedged Requests • Circuit Breakers (Hystrix) • Common analytics & event-logs • ... and much more
40. 40. Complex, multi-tier caching took a lot of code.
41. 41. (In proto) (In client config)
42. 42. gRPC ❤ languages!
43. 43. NIH → OSS 4
44. 44. Value Effort
45. 45. Value Effort
46. 46. With every step comes the decision to take another.
47. 47. Inertia is a powerful force, and a terrible strategy.
48. 48. Favor commodity when it’s not our core competency (oh right! AWS!)
49. 49. Wrapping up… Ω
50. 50. Everything discussed is done gRPC = 10%+ of Netflix RPC 800+ projects made with generators 100+ services currently deployed from generators This stuff = Default for 6-12 months
51. 51. Code generation is the short & long term solution IDLs = micro-services’ best friend Don’t build stuff you don’t need to