Changing Tires at 100mph

The massive growth of Lesara in the last three years lead to a high traffic level of the online shop.
With several million page views, API calls and events every day, it is hard to migrate to new, more modern systems. The migration can lead to bugs, a drop in SEO visibility and unavailability of the whole service, which is feared by marketers, engineers and executives.

The talk walks you through the reasons for a new stack, which technologies we selected and what the new architecture looks like.

Transcript

1. @_RobM
   Changing tires at 100mph
   

   View Slide

2. @_RobM 2
   Gaming
   Legal Tech
   E-Commerce
   Robin Müller
   CTO & Co-Founder at Lesara
   What I do at Lesara
   What I did before
   Scaling
   Tracking
   Tech Strategy
   BS Computer Science
   Security Product
   

   View Slide

3. @_RobM 3
   Our vision
   

   View Slide

4. @_RobM
   Baseline
   20x
   Peaks up to 120x
   

   View Slide

5. Our current (Magento) store is legacy
   

   View Slide

6. @_RobM 6
   

   View Slide

7. @_RobM
   We use caching to scale
   7
   • Different cache types
   • MySQL query cache
   • Opcode cache
   • Application cache
   • Frontend cache
   • Browser cache
   

   View Slide

8. @_RobM
   The hard parts about varnish
   8
   • Cache invalidation
   • Hole punching using AJAX or ESI
   Cached
   Uncached
   

   View Slide

9. Varnish makes our life harder
   A/B Testing Personalization
   Real time updates
   

   View Slide

10. How can we get rid of caching?
    

    View Slide

11. Easy. Make the application lighter!
    

    View Slide

12. Let’s build a new shop frontend!
    

    View Slide

13. @_RobM
    What are we looking for?
    13
    Product
    Loading times < 100ms
    A/B Testing in the core
    Live updates
    Completely personalizable
    Step-by-step migration
    Developers
    PHP 7
    Modern framework
    No more FPC
    Community
    Tech
    Fast & light
    Easy to update
    Open Source
    New CDN
    Fully tested
    

    View Slide

14. Lesara Open Source
    

    View Slide

15. @_RobM 15
    Based on
    PHP 7
    Symfony 3
    elasticsearch 5
    Redis 3
    Bundles
    Elasticsearch bundle
    Filter manager bundle
    Settings bundle
    Router bundle
    Translation bundle
    …
    Built for
    Massive scale
    Step-by-step migration
    Frequent data updates
    Rendering in 100ms
    

    View Slide

16. @_RobM 17
    The new architecture
    

    View Slide

17. @_RobM
    The router can make your application smart
    18
    Requirements
    Response time(<5ms)
    Scalability
    Opportunities
    Save a second lookup via header
    Circuit-breaker-pattern
    A/B Testing
    Staged Feature Rollouts
    

    View Slide

18. @_RobM
    Router implementations
    19
    PHP + fastcgi
    Silex
    > 2000 tr/s
    > 25k – 30k tr/s (fastcgi)
    Stable implementation
    LUA-script w/ OpenResty
    ES 5 only hacked
    10k – 15k tr/s
    Go
    Seg-faults under load
    > 30k tr/s
    

    View Slide

19. @_RobM
    Circuit breaker pattern
    Like a real circuit breaker in a fuse box in your home
    Have a pause when a backend service fails
    MONITOR IT!
    20
    

    View Slide

20. @_RobM 21
    

    View Slide

21. @_RobM 22
    

    View Slide

22. @_RobM
    Indexing data
    23
    Documents (in all languages)
    Products
    Categories
    Settings / Configuration
    Translations
    Other content pages
    

    View Slide

23. @_RobM
    How to prevent duplication
    24
    Header and footer are
    present in all pages.
    Use Server side
    includes
    

    View Slide

24. @_RobM
    The migration plan
    25
    High traffic to low traffic
    Header & Footer via SSI
    Home
    Categories
    Search results
    Product Detail pages
    REST-API
    …
    

    View Slide

25. @_RobM
    How fast is it? (Home page)
    26
    

    View Slide

26. @_RobM
    How does it scale? (Home page)
    27
    

    View Slide

27. @_RobM
    How fast is it? (Categories)
    28
    

    View Slide

28. @_RobM
    How does it scale? (Categories)
    29
    

    View Slide

29. Questions?
    

    View Slide