Microservices - Portland Oregon 2018-09-25

Microservices

Portland Java User Group
Portland Oregon
September 25, 2018

#hbctech #pdxtech #java #scala #microservices

Transcript

1. Sean Sullivan
   September 25, 2018
   Portland Java User Group
   MICROSERVICES
   AT HBC
   

   View Slide

2. software engineer
   Portland Oregon
   back office systems
   Hudson’s Bay Company
   About me
   

   View Slide

3. Monoliths
   Microservices
   

   View Slide

4. MICROSERVICES:
   A FUTURE
   TO BELIEVE IN
   

   View Slide

5. saksfifthavenue.com
   saksoff5th.com
   lordandtaylor.com
   thebay.com
   

   View Slide

6. saksfifthavenue.com
   

   View Slide

7. January 2016
   

   View Slide

8. June 2018
   

   View Slide

9. 2009 large monolith
   2010 a few services
   2011 microservices
   gilt.com
   

   View Slide

10. Ruby
    on Rails
    Postgres
    Gilt monolith (2009)
    

    View Slide

11. Gilt monolith (2009)
    customer facing UI
    customer service UI
    purchase order
    management
    shopping cart logic
    checkout logic
    order processing logic
    fraud scanning
    sales management
    inventory management
    email
    return logic
    daily reports
    

    View Slide

12. Gilt monolith (2009)
    drawbacks
    • unintended coupling of feature releases
    • single repo for customer-facing UI and non-
    customer-facing UI
    • rollbacks were painful
    

    View Slide

13. Gilt monolith (2009)
    drawbacks
    • central coordination of each release
    • difficult to scale to higher levels of traffic
    • difficult to upgrade to new version of Ruby
    on Rails
    

    View Slide

14. Gilt services (2011)
    product
    service
    inventory
    reservation
    service
    shopping
    cart
    service
    user
    service
    checkout
    service
    credit card
    service
    kvstore
    service
    discount
    service
    order
    history
    service
    

    View Slide

15. 2012 L.O.S.A.
    2013 binary coupling hell
    2014 ApiBuilder
    gilt.com
    

    View Slide

16. L.O.S.A. ?
    

    View Slide

17. Lots
    Of
    Small
    Applications
    

    View Slide

18. Customer facing UI
    was a monolithic app
    (hundreds of JSP’s in a
    single Git repo)
    

    View Slide

19. large UI monolith
    smaller UI apps
    

    View Slide

20. Gilt L.O.S.A.
    web-
    product-
    detail
    web-
    homepage
    web-mosaic
    web-search
    web-
    account
    web-
    checkout
    web-login
    web-
    navigation
    web-seo-
    page
    

    View Slide

21. Lots
    Of
    Small
    Bugs
    Lots
    Of
    Small
    Applications
    

    View Slide

22. Benefits of L.O.S.A.
    easier to troubleshoot UI bugs
    easier to implement bug fixes
    easier to test fixes
    easier to release fixes to production
    

    View Slide

23. binary
    coupling
    gilt.com (2013)
    

    View Slide

24. binary coupling
    can emerge in large
    microservice systems
    

    View Slide

25. binary coupling was a
    significant problem at
    Gilt.com and Netflix
    

    View Slide

26. View Slide

27. how to mitigate
    binary coupling?
    

    View Slide

28. API descriptors
    +
    code generation
    

    View Slide

29. ApiBuilder
    gilt.com (2014)
    

    View Slide

30. describe REST API
    in a JSON file
    

    View Slide

31. www.apibuilder.io
    

    View Slide

32. View Slide

33. code generation
    

    View Slide

34. Web
    Checkout
    Android
    Checkout
    iPhone
    Checkout
    Checkout
    service
    

    View Slide

35. ApiBuilder CLI
    

    View Slide

36. $ brew install apibuilder-cli
    $ cd myproj
    $ vi .apibuilder/config
    $ apibuilder update
    

    View Slide

37. 2015 migrate to AWS
    2016 AWS cloud native
    gilt.com
    

    View Slide

38. 2017 embrace serverless
    2018 payment platform v3
    2018 shutdown
    gilt.com
    

    View Slide

39. lessons learned
    from
    gilt.com
    

    View Slide

40. decentralized
    decision making
    

    View Slide

41. minimize
    undifferentiated
    heavy lifting
    

    View Slide

42. one path to production
    

    View Slide

43. small, autonomous
    teams
    

    View Slide

44. end-to-end
    ownership
    

    View Slide

45. deploy to production
    everyday
    

    View Slide

46. small incremental
    releases
    

    View Slide

47. ApiBuilder is awesome
    

    View Slide

48. developer experience
    matters
    

    View Slide

49. continuous
    improvement
    

    View Slide

50. testing in production
    

    View Slide

51. feature flags
    

    View Slide

52. engineering velocity
    is a
    competitive advantage
    

    View Slide

53. View Slide

54. HBC 2018
    multi-tenant services
    ApiBuilder
    safe deployments
    

    View Slide

55. Mark Zuckerberg
    

    View Slide

56. Move Fast and Break Things
    Move Slow and Don’t Break Things
    

    View Slide

57. https://twitter.com/jaykreps/status/312338174108106754
    

    View Slide

58. Move Fast and Don’t Break Things
    Move Purposefully and Fix Things
    

    View Slide

59. continuous delivery
    

    View Slide

60. https://twitter.com/evanderkoogh/status/1036787227697664000
    

    View Slide

61. View Slide

62. HBC email types
    • order acknowledgement
    • order shipment partial
    • order shipment full
    • order cancel
    • ready for pickup in store
    

    View Slide

63. Queue
    consumer
    Event
    queue
    Email
    Dashboard UI
    Email
    Sender
    Email
    Render
    External
    Email
    Provider
    Feature
    Flag
    service
    

    View Slide

64. Email
    Render
    Product
    Detail
    service
    Delivery
    Date
    Estimation
    service
    request
    HTML
    Feature
    Flag
    service
    

    View Slide

65. HBC email system
    • multi-tenant design
    • Cloudwatch Metrics
    • Cloudwatch Alerts
    

    View Slide

66. HBC email system
    • code coverage threshold
    • components deployed independently
    • safe deployments
    

    View Slide

67. safe
    AWS Lambda
    deployments?
    

    View Slide

68. traffic shifting
    

    View Slide

69. You can now shift
    incoming traffic between
    two AWS Lambda
    function versions based on
    pre-assigned weights.
    You can now also
    use AWS
    CodeDeploy to
    automatically manage
    the rollout of new
    function versions.
    This allows you to
    gradually shift traffic
    between two versions source: aws.amazon.com
    

    View Slide

70. AWS traffic hooks
    • PreTraffic hook
    • PostTraffic hook
    

    View Slide

71. Final thoughts
    

    View Slide

72. autonomous teams
    

    View Slide

73. ApiBuilder
    

    View Slide

74. engineering velocity is a
    competitive advantage
    

    View Slide

75. https://www.infoq.com/presentations/microservices-arch-infrastructure-cd
    

    View Slide

76. The end
    

    View Slide

77. View Slide