4 years building services at peerTransfer

Transcript

1. 4 years building services Felipe Talavera
2. A.K.A. Services vs Monoliths
3. "Microservices" - yet another new term on the crowded streets of software architecture. http://martinfowler.com/articles/microservices.html -Martin Fowler
5. 1B$ process +500 schools almost 4 years
6. 0 sinatra 1 rails 0 daemons 0 libraries 1 server November 2010!
8. 0 sinatra 2 rails 1 daemons 0 libraries 2 servers January 2011!
9. HTTP REST API’s! ! Async process with a message broker
11. App per git repo! ! Apps are deployed separately ! Infra as code
12. Issues
13. Features involving multiple apps ! One rails app failure affect the other ! Complex monitoring/acceptance needed !
14. dependency resolution ! version tagging and deployment! ! handling performance degradation Manual
15. ! Error Handling cascading, trace, storm, retry
16. Infra on AWS June 2012!
17. http://martinfowler.com/bliki/MicroservicePrerequisites.html Rapid provisioning Basic Monitoring Rapid application deployment
18. 4 sinatra 4 rails 4 daemons 3 libraries 6 servers June 2012!
22. 10 sinatra 4 rails 8 daemons 7 libraries 21 servers October 2014!
23. teams around services
24. Accidental complexities
25. API consumers Auto generated consumer, ex: ActiveResource ! Cross app testing, consumer mock mode ! Recording traffic, VCR approach
26. A/B testing across apps ! Feature flagging
28. 40 sinatra 3 rails 15 daemons 20 libraries 40 servers July 2016!
29. ! Docker containers ! Consul ! Etcd ! Terraform Our next steps:!
30. Modern alternatives !
31. ! Hystrix/RxJava Latency and Fault Tolerance for Distributed Systems Java lib runs in JVM languages, scala, clojure, etc… from netflix http://blog.josephwilk.net/clojure/building-clojure-services-at-scale.html
32. ! Finagle Pipes-and-Filters meets Futures model. Thrift for RPC ZooKeeper for Service Discovery from twitter Finatra Scala lib
33. ! Zipkin A distributed tracing system
34. Apache Mesos Develop and run resource-efficient distributed systems ! ! Cluster manager that simplifies the complexity of running applications on a shared pool of servers.
35. Learned lessons! ! What things we would do differently?
36. Thanks!