High-Impact refactors keeping the lights on

First presented at PyConES 2017

Transcript

1. High-impact
   refactors
   keeping the
   lights on
   @Kartones
   Diego Muñoz
   

   View Slide

2. 01.
   What
   02.
   Why
   03.
   How
   04.
   When
   Agenda
   

   View Slide

3. What
   

   View Slide

4. What
   

   View Slide

5. What
   

   View Slide

6. Why
   

   View Slide

7. Why
   

   View Slide

8. Why
   

   View Slide

9. How
   

   View Slide

10. Mega
    structures
    How
    

    View Slide

11. Mega
    structures
    How
    

    View Slide

12. DB
    High-frequency
    selling
    How
    CMS
    API
    CRUD
    Current (simplified)
    flow
    

    View Slide

13. How
    Extend,
    not modify*
    API
    DB
    High-frequency
    selling
    CRUD
    CMS
    REDIS
    EVENT
    BUS
    EVENT
    BUS
    APU
    *do modify if needed to allow
    extension
    

    View Slide

14. Patterns &
    Practices
    - Parallel change
    - Strangler
    How
    API
    DB
    High-frequency
    selling
    CRUD
    CMS
    REDIS
    EVENT
    BUS
    EVENT
    BUS
    APU
    

    View Slide

15. How
    API
    DB
    High-frequency
    selling
    CRUD
    CMS
    REDIS
    EVENT
    BUS
    EVENT
    BUS
    APU
    Patterns &
    Practices
    - Event Bus
    - Cursor-based
    - N writers/readers
    

    View Slide

16. How
    API
    DB
    High-frequency
    selling
    CRUD
    CMS
    REDIS
    EVENT
    BUS
    EVENT
    BUS
    APU
    Patterns &
    Practices
    - Event Sourcing
    

    View Slide

17. How
    API
    DB
    High-frequency
    selling
    CRUD
    CMS
    REDIS
    EVENT
    BUS
    EVENT
    BUS
    APU
    “Build exciting
    things with boring
    nice technologies”
    - Python 3.6
    · Strongly typed (Mypy)
    · Dockerized
    - Redis with LUA
    - AWS Kinesis
    github.com/ticketea/pynesis
    

    View Slide

18. When
    

    View Slide

19. When
    Continous Iteration
    - No hard deadline
    - Desired: soon :-)
    

    View Slide

20. Latest benchmark:
    ~4100 ops/sec
    - Single CPU core, 20k
    tickets, 50k concurrent
    users
    - Attacking internal API
    - Room for optimization
    When
    

    View Slide

21. Current status:
    - Free events
    - Paid events on their way
    · Play safe
    (money involved!)
    · Lots of payment methods
    When
    

    View Slide

22. Release plan:
    - Now: Hack
    (Special payment method)
    - 1.0: Feature Toggling
    When
    

    View Slide

23. Rollout plan
    01. Canary: big free event
    02. Dark launch: on paid events
    03. B/G: Gradual switch
    When
    

    View Slide

24. THE END
    Thanks!
    @Kartones
    

    View Slide