Flink platforms, state internals, watermarks, exactly-once — the compute layer above the broker. The Kafka alumni ladder.
Deliberately compute-layer: where the Kafka cohort taught the broker, this one teaches the processing — LinkedIn's 4-trillion-event pipelines and −94% unification, Alibaba's 4B records/sec Double-11 spine, and the state/time semantics that decide whether streaming results are actually correct.
Exactly-once is a chain — transactions, deterministic identity, idempotent sinks, LSN audits — and streaming SQL is a tower with three known leaks: hot keys, destroyed watermarks, retractions.
These events are money. Every impression your pipeline counts becomes a line on an advertiser's invoice; every click feeds a budget pacer deciding what to spend next minute. Lose events and you lose revenue. Count one twice and you overcharge a customer — and the failure modes stack: the job restarts mid-window, the message bus redelivers, a whole region fails over and replays. Your CEO asks the question in four words: *"can we double-bill?"* "The framework has exactly-once" is not an answer — exactly-once *where*, against *which* failure? What would a real answer look like?
The full week 3 brief is part of LeetData Pro.