How not to screw up when scaling processing from 1 country to 33

Talk video

How did we scale data processing with tens of machine learning models, working only for the Ukrainian market, in 33 countries?

• Splitted monoliths, responsible for queue management and intellectual processing, to independent microservices
• Unified codebase for individual data processing algorithms in different countries
• Got rid off queues in mysql
• Automated services deploy
• Set up services discovery
• Integrated load testing
• Made reliable processing of thousands of events per day