Containers and Micro-Services

Presented at the Container Summit, hosted by Joyent, in New York City on April 8th, 2015.

Transcript

1. Containers and Microservices Like Peanut Butter & Bananas
2. #nodebf @walmartlabs @jasonpincin • microservice platform • 900+ processes • multiple DCs Analytics Mobile Device Config • 1,540 processes • multiple DCs • 150,000 msg/sec peak
3. Container Discovery Depicted
4. The Problem Soaring complexity, difficult to deploy, even more difficult to test
5. Changing Course • Reduce complexity • Avoid developer gridlock • Ship constantly • Fewer meetings
6. Scalability & Availability • Simple to increase capacity • Failure tolerant • Failure isolation • Efficiency
7. How? Make things smaller!
8. First Try
9. Smaller!
11. Microservice Properties • Simple, does one thing • API exposed to network • Independently deployable • Disposable • Collections form applications
12. Build it
13. Did it work? • Reduce complexity • Avoid developer gridlock • Ship constantly • Fewer meetings • Simple to increase capacity • Failure tolerant • Failure isolation • Efficiency
14. ☑ Gridlock ☑ Constantly shipping ☑ Scalable ☑ Failure management ☑ Efficiency Succeeded in…
15. More Microservice Benefits • Simplified & asynchronous delivery pipeline • Implementation agnostic; avoids commitment to technology stack • Loose coupling
16. ❌ Fewer meetings ❌ Reduced complexity (it shifted) Fell short at…
17. Microservice Operational Challenges • The network • Process management • Deployments
18. Where?
19. Virtualization • Heavy • Performance Penalty • Limited/No benefit to architecture
20. Zones • Light • Uniquely addressable • Bare-metal speed • Resource controls
21. Continued • DTrace, MDB • Service Management Facility (SMF) • ZFS
22. Containers? Containers!
23. Zone Images (aka Containers) • Relatively small • Easy to build • Bash-scriptable deploy
24. Zone Images cont. • Robust server support • Limited/No tooling/support for other platforms • Required additional plumbing to be applicable to microservice deployment process • Led to arbitrary process division
25. Docker
26. Docker • All the advantages of zones • Able to run same container in all environments • Big and growing ecosystem • Excellent tooling
27. Docker (but…) • Required Linux (back to virtualization) • Another layer on top of that • Concerned about performance, overhead, and efficiency that led to zones to begin with • Giving up DTrace, MDB, etc
28. Triton • Native speed • Keep the Debugging toolchain • Single Docker host means simplified, consistent provisioning and deployment - let platform figure it out • Microservices’ BFF
29. Whales all the way down