#21Ad Click Aggregator
Stream processing with watermarks and exactly-once.
Build the system that powers "clicks per ad per minute, per hour, per day" reports for advertisers. Inputs: billions of click events flowing in from edge servers, late-arriving by seconds to days. Output: dashboards advertisers refresh constantly, plus billing rollups they will sue you over if they're wrong. Every choice — windowing, watermarks, idempotency, lambda vs kappa — is about the trilemma: low latency, exact accuracy, late-event tolerance.
Reading: Kafka paper · Streaming 101 & 102 · Flink exactly-once
stream processing
exactly-once
watermarks
lambda vs kappa