Scoring 2000 Products per Request: Performance Pitfalls in Golang

Modern recommendation and ranking systems often need to score hundreds or even thousands of candidates within a single request. While the scoring logic itself may appear simple, performance problems quickly emerge as the number of objects and features grows. In this session, Ivan will walk through real-world challenges encountered when scoring up to 2000 products per request in production Golang services. He will cover memory allocation pitfalls, excessive object creation, garbage collector pressure, and concurrency issues that arise when parallelizing scoring logic. The session will also explore how algorithm choice, data structures, and access patterns directly impact latency and throughput. By the end of the talk, you will understand how to identify performance bottlenecks in Go-based scoring pipelines and apply practical techniques to keep latency predictable under high load.