The 12Factor App

Transcript

1. @dmathieu Software Engineer @heroku The 12Factor App 1

2. Twelve Factor • Can apply to any language and platform!

   • Speeds up deployment! • Makes scaling easier! • Keeps apps clean 2

3. Let’s take a look 3

4. One codebase to rule them all 4

5. Codebase Production Staging Feature A Feature B 5

6. 6 Multiple codebases is a distributed system

7. 7 Create libraries

8. Explicitly declare and isolate dependencies 8

9. 9 source 'https://rubygems.org' ! ruby '2.1.3' ! gem 'rails' gem

   'pg' gem 'puma'

10. 10

11. 11 curl imagemagick

12. 12 Config never goes in source control

13. Config Resource strings to databases Credentials to S3, Facebook, Twitter,

    … Security Tokens 13

14. $ cat .env! AWS_TOKEN=QWERTYUIOP! AWS_SECRET=12345 ! GITHUB_APP_ID=1! GITHUB_SECRET=67890 14

15. $ foreman run rails console! >> puts ENV[‘AWS_TOKEN’]! > “QWERTYUIOP”

    15

16. You can use a yml config file* *Just don’t store

    it in source control

17. Treat backing services as attached resources 17

18. 18 DATABASE_URL= postgres://example.com:5032

19. Production deploy PostgreSQL Email Service Amazon S3

20. 20 One codebase per app Explicitly declare and isolate dependencies

    Don’t put config in source control Treat backing services as attached resources

21. Build Release Run 21

22. Codebase Build Run Config 22

23. Codebase Build Run Config 23

24. Execute the app as one or more stateless processes 24

25. web.1 web.2 Database 25

26. 26 No sticky sessions

27. 27 Ephemeral filesystem

28. 28 Export services via port binding

29. 29 Self-contained app

30. 30 An app can be the backing service for an

    other

31. Scale out via the process model 31

32. web.1 web.2 web.3 worker.1 worker.2 clock.1 32

33. 33 You can still do your own internal multiplexing

34. 34 Build, release, run One or more stateless processes Scale

    out via the process model Export services via port binding

35. Maximize robustness with fast startup and graceful shutdown 35

36. 36 Handle a request

37. 37 60s to boot is a lot

38. 38 Use the SIGTERM signal

39. 39 Send jobs back to the queue

40. 40 Build, release, run One or more stateless processes Scale

    out via the process model Fast startup

41. Keep development, staging, and production as similar as possible 41

42. 42 Use the same kind of machine

43. 43 With the same OS version

44. 44 And the same database server

45. Logging 45

46. 46 Production errors will happen

47. 47 Tail logs to find the cause

48. 48 Then use a logging service to dig in deeper

49. action=create_user email=foo@bar.com action=log_user email=alice@heroku.com 49

50. count#api-call-200=1 50

51. Don’t write logs to a file 51

52. Run admin/ management tasks as one-off processes 52

53. 53 Not in your app’s main processes

54. 54 Run console instances the same way

55. 55 Keep all your environments as similar as possible Run

    management tasks as one-off processes Logging

56. Do you like? • Minimizing new developer overhead?! • Running

    in multiple environments?! • Easily scaling without tooling, architecture or development headaches?! • Having the latest update available to users at a moments notice? Read more at 12factor.net 56

57. Questions? damien@heroku.com 12factor.net 57 @dmathieu