The Birth of DevOps

From 2½ Days to 2½ Seconds - the Birth of DevOps

This is an experience report from the mid-2000s, where a fifty-person programme team was being held up by test environments that took an average of 2½ days to prepare. By developing strategies like Blue/Green Deployments, the Single Deployable Artifact, Container Templates, the Build Production Line and many of the other principles that led to Continuous Delivery, we brought this down to only a few minutes, with a near-instantaneous cutover for the users. The legendary team included the likes of Julian "Build Doctor" Simpson, Chris Read, Sam Newman and both the authors of the Continuous Delivery book, only one of whom swore a lot. Dan North is not the father of DevOps but he may be its midwife. He definitely assisted at the birth.

Transcript

1. From 2½ days to 2½ seconds
   - the birth of DevOps
   Dan North
   @tastapod
   

   View Slide

2. Once upon a time…
   

   View Slide

3. Once upon a time…
   back in the mists of 2005
   

   View Slide

4. Once upon a time…
   back in the mists of 2005
   before Continuous Delivery™
   

   View Slide

5. Once upon a time…
   back in the mists of 2005
   before Continuous Delivery™
   a team was stuck.
   

   View Slide

6. Once upon a time…
   back in the mists of 2005
   before Continuous Delivery™
   a team was stuck.
   This is their story.
   

   View Slide

7. Life was tough
   back then
   

   View Slide

8. Life was tough
   back then
   2½ days to deploy a build
   

   View Slide

9. Life was tough
   back then
   2½ days to deploy a build
   Testers were frustrated!
   

   View Slide

10. Life was tough
    back then
    2½ days to deploy a build
    Testers were frustrated!
    Developers were frustrated!
    

    View Slide

11. Life was tough
    back then
    2½ days to deploy a build
    Testers were frustrated!
    Developers were frustrated!
    We were bottlenecked at testing :(
    

    View Slide

12. Life was tough
    back then
    2½ days to deploy a build
    Testers were frustrated!
    Developers were frustrated!
    We were bottlenecked at testing :(
    actually bottlenecked at deployment!
    

    View Slide

13. The causes were
    obvious
    

    View Slide

14. The causes were
    obvious
    “snowflake” environments
    

    View Slide

15. The causes were
    obvious
    “snowflake” environments
    too many moving parts
    

    View Slide

16. The causes were
    obvious
    “snowflake” environments
    too many moving parts
    multiple teams
    

    View Slide

17. The causes were
    obvious
    “snowflake” environments
    too many moving parts
    multiple teams
    contention for resources
    

    View Slide

18. Where do we start?
    

    View Slide

19. Where do we start?
    "Elevate the constraint"
    

    View Slide

20. Where do we start?
    "Elevate the constraint"
    Engage Operations as customer
    

    View Slide

21. Where do we start?
    "Elevate the constraint"
    Engage Operations as customer
    Understand the manual steps
    

    View Slide

22. Where do we start?
    "Elevate the constraint"
    Engage Operations as customer
    Understand the manual steps
    Identify where time is being sunk
    

    View Slide

23. Where do we start?
    "Elevate the constraint"
    Engage Operations as customer
    Understand the manual steps
    Identify where time is being sunk
    Focus on the outcome, not the steps
    

    View Slide

24. Idea: Treat the
    container as code
    

    View Slide

25. Idea: Treat the
    container as code
    WebLogic is just XML
    

    View Slide

26. Idea: Treat the
    container as code
    WebLogic is just XML
    so create templates
    

    View Slide

27. Idea: Treat the
    container as code
    WebLogic is just XML
    so create templates
    and version-control them!
    

    View Slide

28. Idea: Treat
    deployment as code
    

    View Slide

29. Idea: Treat
    deployment as code
    Instrument the UI tools
    

    View Slide

30. Idea: Treat
    deployment as code
    Instrument the UI tools
    Learn the command-line tools
    

    View Slide

31. Idea: Build the app
    as a single entity
    

    View Slide

32. Idea: Build the app
    as a single entity
    Single Deployable Artefact
    

    View Slide

33. Idea: Make all the
    servers look the same
    

    View Slide

34. Idea: Make all the
    servers look the same
    Same OS
    

    View Slide

35. Idea: Make all the
    servers look the same
    Same OS
    Same packages
    

    View Slide

36. Idea: Make all the
    servers look the same
    Same OS
    Same packages
    Same versions
    

    View Slide

37. Idea: Make all the
    servers look the same
    Same OS
    Same packages
    Same versions
    Same(ish) settings
    

    View Slide

38. Idea: Automate the
    entire deployment
    

    View Slide

39. Idea: Automate the
    entire deployment
    Conan the Deployer!
    

    View Slide

40. Conan the Deployer
    

    View Slide

41. Conan the Deployer
    - Create a new container instance
    

    View Slide

42. Conan the Deployer
    - Create a new container instance
    - Generate container config
    

    View Slide

43. Conan the Deployer
    - Create a new container instance
    - Generate container config
    - Deploy container config
    

    View Slide

44. Conan the Deployer
    - Create a new container instance
    - Generate container config
    - Deploy container config
    - Bring up the master node
    

    View Slide

45. Conan the Deployer
    - Create a new container instance
    - Generate container config
    - Deploy container config
    - Bring up the master node
    - Deploy the app into the node
    

    View Slide

46. Conan the Deployer
    - Create a new container instance
    - Generate container config
    - Deploy container config
    - Bring up the master node
    - Deploy the app into the node
    - Bring up the app
    

    View Slide

47. Conan the Deployer
    - Create a new container instance
    - Generate container config
    - Deploy container config
    - Bring up the master node
    - Deploy the app into the node
    - Bring up the app
    - Smoke test the environment
    

    View Slide

48. This was the big
    breakthrough
    

    View Slide

49. This was the big
    breakthrough
    Any build into any environment
    

    View Slide

50. This was the big
    breakthrough
    Any build into any environment
    2½ days down to 25 minutes!
    

    View Slide

51. This was the big
    breakthrough
    Any build into any environment
    2½ days down to 25 minutes!
    Deterministically.
    

    View Slide

52. Then optimise…
    

    View Slide

53. Then optimise…
    Blue-Green Deployments
    

    View Slide

54. Then optimise…
    Blue-Green Deployments
    Create self-serve deployments
    

    View Slide

55. Then optimise…
    Blue-Green Deployments
    Create self-serve deployments
    Sub-second cutover, during office hours!
    

    View Slide

56. What did we learn?
    

    View Slide

57. What did we learn?
    Focus on the outcome
    

    View Slide

58. What did we learn?
    Focus on the outcome
    Focus on the current bottleneck
    

    View Slide

59. What did we learn?
    Focus on the outcome
    Focus on the current bottleneck
    You don't "have an investment" in
    hardware
    

    View Slide

60. What did we learn?
    Focus on the outcome
    Focus on the current bottleneck
    You don't "have an investment" in
    hardware
    DevOps requires collaboration
    

    View Slide

61. Epilogue
    

    View Slide

62. Epilogue
    Sam Newman develops DbDeploy
    

    View Slide

63. Epilogue
    Sam Newman develops DbDeploy
    Jez Humble and Chris Read (and me :)
    describe the Build Production Line
    

    View Slide

64. Epilogue
    Sam Newman develops DbDeploy
    Jez Humble and Chris Read (and me :)
    describe the Build Production Line
    Chris Read co-creates DevOps Days
    

    View Slide

65. Epilogue
    Sam Newman develops DbDeploy
    Jez Humble and Chris Read (and me :)
    describe the Build Production Line
    Chris Read co-creates DevOps Days
    Julian Simpson becomes the Build Doctor
    

    View Slide

66. Epilogue
    Sam Newman develops DbDeploy
    Jez Humble and Chris Read (and me :)
    describe the Build Production Line
    Chris Read co-creates DevOps Days
    Julian Simpson becomes the Build Doctor
    Jez Humble and Dave Farley write “Continuous
    Delivery” and win Jolt Award
    

    View Slide

67. Whither DevOps?
    

    View Slide

68. Whither DevOps?
    Commoditise All the Things!
    

    View Slide

69. Whither DevOps?
    Commoditise All the Things!
    Cloud All the Things!
    

    View Slide

70. Whither DevOps?
    Commoditise All the Things!
    Cloud All the Things!
    Hug a SysAdmin today
    

    View Slide

71. Whither DevOps?
    Commoditise All the Things!
    Cloud All the Things!
    Hug a SysAdmin today
    And remember: A “build team” is still an
    anti-pattern!
    

    View Slide

72. Thanks for listening
    @tastapod
    dan@dannorth.net
    http:/
    /dannorth.net
    !
    Don’t forget to vote!
    

    View Slide