Back to problems

Implement streaming SRM detector with late events

System Design · ByteDance · Hard

Design a streaming monitor that detects sample-ratio mismatch (SRM) across many experiments running concurrently. The system consumes two key-partitioned topics. One topic carries experiment assignment records with fields experiment_id, user_id, variant, and ts; the other carries page-view events with user_id and ts. Your implementation must satisfy the following: Idempotently de-duplicate assignment records for each (experiment_id, user_id) pair, so reprocessing or…

Checking your access…