The Next Generation of Microservices — YOW 2017 Brisbane
2 likes909 views
The document discusses the evolution of microservices architectures over time as techniques were developed to address challenges introduced by highly distributed systems. Early approaches involved implementing circuit breakers, timeouts, and sharing state across services to handle failures. Later, service discovery and client-side load balancing were introduced. Today, sidecars and service meshes move reliability patterns lower in the stack and provide stability, security, and other features to services through a common platform. However, these abstractions also introduce new tradeoffs around opacity, development cycles, and homogeneity.
The Next Generation of Microservices — YOW 2017 Brisbane
- 1. The Next Generation of Microservices Phil Calçado http://philcalcado.com @pcalcado
- 3. highly distributed application architecture We can think of microservices as
- 4. Highly distributed application architectures are just a flavor of distributed computing
- 5. Technical Challenges Organization Challenges Introducing distribution into your architecture brings up new
- 6. Introducing distribution into your architecture brings up new
- 8. The example we will work with today: Recovering from temporary failure
- 11. "I know, let’s use microservices”
- 12. Timeouts Your code looks like this now Biz Logic Timeouts
- 13. server client
- 14. Timeouts Your code looks like this now Biz Logic Timeouts RPC code
- 15. server client 🔥 🔥 🔥
- 18. "I know, let’s implement circuit breakers and timeouts”
- 19. Circuit breakers Timeouts Your code looks like this now Biz Logic Timeouts Telemetry RPC code
- 20. server client client client client client client
- 22. 🔥 wait ☺ 😰 🔥 🔥 ☺ wait 😰 ☺ 🔥 🔥 wait 😰 ☺ 🔥 🔥 wait 😰 ☺ 🔥 🔥 …
- 23. 🔥 🗑
- 24. "I know, let’s have all circuit breakers share state”
- 25. "I know, let’s have all circuit breakers share state” Answer to the question “how did my application ended up importing a Zookeeper library again?"
- 26. Circuit breakers Timeouts Your code looks like this now Biz Logic Timeouts Telemetry RPC code Distributed state
- 27. "I know, let’s have clients keep track of healthy instances”
- 28. server client 🔥 🔥 🔥 server server serverserver 🔥 🔥 🔥 💩 🤔 Which instance should we talk to? ☺
- 29. DNS be like… '
- 30. Circuit breakers Timeouts Your code looks like this now Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery
- 31. And that’s just for RPC reliability.
- 32. What makes 2017 different?
- 34. Circuit breakers Timeouts Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery SDK
- 35. Application Transport Internet Network Circuit breakers Telemetry RPC code Distributed state Client-side Service Discovery }TCP/IP }? One way to think about it
- 36. But how does one go about adding a new layer to the TCP/IP stack?
- 37. Sidecars to the rescue Circuit breakers Timeouts Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery SCARY OUTSIDE WORLD
- 38. Sidecars to the rescue Circuit breakers Timeouts Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery SCARY OUTSIDE WORLDSidecar
- 39. Sidecars to the rescue
- 41. How does that impact your service?
- 42. Circuit breakers Timeouts Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery Service Platform
- 45. Timeouts Biz Logic Stability Service Platform Capacity Security Availability
- 47. Allows our services to pretend some of those are true
- 48. It’s not that these patterns aren’t used anymore, it’s just that the dumb work moved down the stack.
- 49. New, optmised, protocols are quite opaque (e.g. gRPC and friends) Works better with metadata-rich protocols
- 50. It makes it even harder to fully replicate production earlier in the development cycle Coupled to the platform
- 51. Not everything will be part of the mesh, i.e. the network still isn’t homogeneous Leaky abstraction
- 52. Q&A