Refacoring vs Rewriting WixStores
0 likes848 views
This document discusses Wix's transition from a monolithic eCommerce system to a microservices architecture using event sourcing and CQRS patterns. The old system had scaling issues as a single monolith. The new system decomposes functionality into independent microservices that are easier to develop and scale. It addresses problems like single points of failure, deployment challenges, and improving developer velocity with event sourcing and CQRS. Lessons learned include the complexity of these patterns, importance of frameworks, and handling data growth. The transition solved scalability and development speed at the cost of additional architectural complexity.
Refacoring vs Rewriting WixStores
- 1. Wix eCommerce Case Study Refactoring vs Rewriting Doron Rosenstock doron@wix.com
- 2. What is Wix? What is WixStores?
- 3. Agenda Part I Overview of old system Problems with the old system Refactoring vs rewriting Part II Problems and solutions Part III Lessons learned 3
- 4. PART I Wix eCommerce aka the “old system”
- 5. Architecture One monolithic server Magento as a base Written in PHP MySql DB Redis Cache
- 6. Some Numbers ~1M stores Over 5M http requests per day Average response time of ~300 ms 2 data centers + 1 cloud (AWS) Couple of seconds for store provision
- 7. The Good Stable system Mature product Lots of satisfied customers Domain knowledge
- 13. Pros Will continue to deliver Only critical sections fixed Cons Stuck with dead tech Still different stack than Wix Rewriting? Refactoring?
- 14. Pros If done correctly, solves old system’s problems Cons Product freeze for 6 months Might fail Pros Will continue to deliver Only critical sections fixed Cons Stuck with dead tech Still different stack than Wix Rewriting?Refactoring?
- 17. PART II aka the “new system”
- 19. What’s the problem with Mono-server? One server to rule them all - single point of failure Dependency between unrelated features Need to deploy the entire system each time
- 20. What’s the problem with Mono-server? One server to rule them all - single point of failure Dependency between unrelated features Need to deploy the entire system each time
- 22. Micro-services Each microservice is relatively small Each service can be deployed independently of other services Easier to scale development Improved fault isolation Each service can be developed and deployed independently Eliminates any long-term commitment to a technology stack
- 23. Did we solve our problems?
- 24. Did we solve our problems? One server to rule them all - single point of failure Dependency between unrelated features Need to deploy the entire system each time
- 25. Did we solve our problems? One server to rule them all - single point of failure Dependency between unrelated features Need to deploy the entire system each time
- 26. Lessons Learned about Micro-Services Error handling and log tracing in the cluster Synced micro-services (WIP) Configuration of micro-services (WIP) Backward and forward compatibility design
- 27. Backward and forward compatibility Service1 V1 Service2 V1
- 28. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT
- 29. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2
- 30. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2 PRODUCT
- 31. Backward and forward compatibility Service2 V1 Service1 V2 PRODUCT
- 32. Backward and forward compatibility Service2 V1 Service1 V2 PRODUCT Service2 V2
- 33. PRODUCT’ Backward and forward compatibility Service2 V1 Service1 V2 PRODUCT Service2 V2
- 34. PRODUCT’ Backward and forward compatibility Service1 V2 Service2 V2
- 35. Backward and forward compatibility Service1 V1 Service2 V1
- 36. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT
- 37. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2
- 38. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2 PRODUCT
- 39. Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2 PRODUCT Service2 V2
- 40. PRODUCT’ Backward and forward compatibility Service1 V1 Service2 V1PRODUCT Service1 V2 PRODUCT Service2 V2
- 42. Web App DBClient Traditional CRUD
- 43. Web App DBClient Traditional CRUD upadteProduct PRODUCT
- 44. Web App DBClient Traditional CRUD upadteProduct PRODUCT
- 45. Web App DBClient Traditional CRUD getProduct PRODUCT
- 46. Web App DBClient Traditional CRUD getProduct PRODUCT
- 47. someCommand Event Sourcing Web App DBClient COMMAND
- 48. someCommand Event Sourcing Command = “Change product X price to 3$“ Web App DBClient COMMAND
- 49. someCommand Event Sourcing Command = “Change product X price to 3$“ Web App DBClient COMMAND
- 50. someCommand Event Sourcing Command = “Change product X price to 3$“ Web App DBClient COMMANDEVENT event = “Product X price was changed to 3$“
- 51. someCommand Event Sourcing Command = “Change product X price to 3$“ Web App DBClient EVENT event = “Product X price was changed to 3$“
- 52. Event Sourcing Web App DBClient EVENT EVENT EVENT
- 53. getProduct Event Sourcing Web App DBClient EVENT EVENT EVENT
- 54. getProduct Event Sourcing Web App DBClient EVENT EVENT EVENT
- 55. getProduct Event Sourcing Web App DBClient VIEW
- 56. getProduct Event Sourcing Web App DBClient VIEW
- 57. Event Sourcing @ WixStores Flexible Immutable Stateless
- 58. DB view
- 60. Lessons Learned about Event Sourcing Paradigm shifts - convincing the team… Code complexity Invest time in your framework - it will pay up Solves backward and forward compatibly problem DB size - space problem?
- 67. getProduct Web App Events DB Client VIEW SnapShots VIEW Command Query Responsibility Segregation
- 69. Lessons Learned about CQRS Again - Code complexity Self-healing DevOps - two DB’s (Mysql + Mongo) Performance
- 71. So… When should we stop and think?
- 72. CQRS CUTTING EDGE TECHNOLOGY ! MICRO-SERVICES EVENT-SOURCED LOW DEV VELOCITY DEAD TECHNOLOGY MONOLITH SERVER CAN’T SCALE ?
- 73. CQRS CUTTING EDGE TECHNOLOGY ! MICRO-SERVICES EVENT-SOURCED LOW DEV VELOCITY DEAD TECHNOLOGY MONOLITH SERVER CAN’T SCALE ?