Ad Click Aggregator

Design an ads click aggregation pipeline. The Apple version emphasizes multiple data sources such as web, mobile, and TV, plus click-count accuracy. Requirements Design a click aggregation system for ads. Functional requirements: Ingest click events from web, mobile, and TV sources. Produce near-real-time counts for dashboards and campaign decisions. Preserve accurate click data for billing / audit use cases. Deduplicate retries and handle late / out-of-order events. Support queries by campaign, creative, source, and time window. Scale / constraints: Assume high write throughput, bursty traffic, and hot campaigns. Accuracy matters more than a best-effort dashboard because the data may affect billing or ranking decisions. Design decisions: Streaming path vs batch correction path. Event-time windows and watermarking vs processing-time simplicity. Idempotent event keys and dedupe storage. OLAP store / cube materialization vs raw-log replay. Notes A strong answer uses a raw immutable event log as the source of truth, a streaming processor for fast aggregates, and a batch reconciliation path for correctness. Hot keys need salting or hierarchical aggregation. Exactly-once is usually achieved practically through at-least-once delivery plus idempotent sinks, not by assuming the whole pipeline is magically exactly-once. Preparation Whiteboard the pipeline in five boxes: edge collector, durable log, stream processor, aggregate store, batch reconciliation. Prepare a dedupe key: (event_id, source, campaign_id) or a server-issued impression / click id. Practice the late-event story: event time, watermark, correction event, and dashboard vs billing semantics. Be ready to explain how you detect drift between real-time aggregates and batch truth.