Ad Click Aggregator
Difficulty: Hard
What It Tests
High-throughput event ingestion, windowed aggregation, deduplication, billing accuracy.
Topics Covered
- Stream Processing
- Distributed Counting / Top K
- Big Data Architectures
- Message Queues & Async Processing
- ZooKeeper / Distributed Coordination
- Scaling Fundamentals
Hello Interview Breakdown
Read the full Hello Interview breakdown →
Approach Hints
- Ingest click events to Kafka partitioned by ad_id so all clicks for the same ad land on the same partition (ordered processing)
- Flink consumes and aggregates in 1-minute tumbling windows; deduplicate within each window using a bloom filter per ad
- Write windowed aggregates to a time-series DB; serve real-time dashboards from the stream layer and billing reports from the batch layer (Lambda architecture)
- Nightly batch reconciliation job reprocesses raw events from Kafka for billing-accurate totals, correcting any approximation from the stream layer