Why You Might Not Need Yet Another Environment

Multiple environments are a staple of software development projects, but do they actually help us make better software? Do they help us get it into the hands of customers faster? In this talk, I look at the goal of every development team -- delivering quality software. I review how environments have historically helped achieve this goal as well as the problems they have caused along the way. Finally, I look at alternative techniques that can be used in lieu of environments.

Transcript

1. WHY YOU MIGHT NOT NEED
   YET ANOTHER
   ENVIRONMENT
   @bradgignac
   

   View Slide

2. Deliver Quality Software
   The Goal:
   

   View Slide

3. Development Production
   

   View Slide

4. Development Production
   Staging
   

   View Slide

5. What’s the big deal?
   

   View Slide

6. Environments drift.
   

   View Slide

7. Environments lie.
   

   View Slide

8. Environments multiply.
   

   View Slide

9. Development Production
   Staging
   

   View Slide

10. Development DI CI Test Staging Production Stable
    

    View Slide

11. Development DI CI Test Staging Production Stable
    Development DI CI Test Staging Production Stable
    Development DI CI Test Staging Production Stable
    Development DI CI Test Staging Production Stable
    Development DI CI Test Staging Production Stable
    

    View Slide

12. Environments are expensive.
    

    View Slide

13. We have more tools available.
    Good News:
    

    View Slide

14. Feature Flags
    if feature.IsEnabled("newWidget") {
    widget := widget.New()
    } else {
    widget := widget.Old()
    }
    widget.Render()
    

    View Slide

15. Team Shipping
    if user.IsInGroup("my-team") {
    widget := widget.New()
    } else {
    widget := widget.Old()
    }
    widget.Render();
    

    View Slide

16. Phased Rollout
    if user.IsInPercentage(10) {
    widget := widget.New()
    } else {
    widget := widget.Old()
    }
    widget.Render();
    

    View Slide

17. Parallel Code Paths
    Request
    Old Code
    New Code
    Compare
    Results
    Persist
    (Old Result)
    Graphite
    

    View Slide

18. Parallel Code Paths
    0
    25
    50
    75
    100
    2:00 2:30 3:00 3:30 4:00 4:30 5:00 5:30 6:00 6:30 7:00 7:30 8:00 8:30 9:00 9:30 10:00
    Total Mismatch
    

    View Slide

19. Parallel Code Paths
    Request
    Old Code
    New Code
    Compare
    Results
    Persist
    (New Result)
    Graphite
    

    View Slide

20. Canary
    v1 v1
    v1 v1
    v1 v1
    v1 v1
    Standby DC Production DC
    Release
    

    View Slide

21. Canary
    v2 v2
    v2 v2
    v1 v1
    v1 v1
    Standby DC Production DC
    Release
    

    View Slide

22. Canary
    v2 v2
    v2 v2
    v2 v2
    v2 v2
    Standby DC Production DC
    Release
    

    View Slide

23. Blue-Green Deployments
    Load
    Balancer
    v1 v1
    v1 v1
    v2 v2
    v2 v2
    100% 0%
    

    View Slide

24. Blue-Green Deployments
    Load
    Balancer
    v1 v1
    v1 v1
    v2 v2
    v2 v2
    50% 50%
    

    View Slide

25. Blue-Green Deployments
    Load
    Balancer
    v1 v1
    v1 v1
    v2 v2
    v2 v2
    0% 100%
    

    View Slide

26. No Free Lunch
    Increased operation complexity.
    Automated deployment pipeline.
    Centralized logging and metrics.
    

    View Slide

27. Happy Accidents
    Everything is automated.
    Experimentation is built-in.
    Disaster recovery is built-in.
    

    View Slide

28. Lessons Learned
    Test everything.
    Measure everything.
    Feature flags by default.
    Operational viability matters.
    Team culture matters most.
    

    View Slide

29. DO YOU REALLY NEED
    YET ANOTHER
    ENVIRONMENT
    

    View Slide

30. Questions?
    

    View Slide