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.
A watermark is a completeness assertion: the minimum-across-partitions rule, the idle-partition stall, state amplification by design, and the DFS-primary inversion that uncouples checkpoints from state size.
You ship a one-minute windowed aggregation over a Kafka topic. Data is flowing — you can see events landing. The query returns nothing. You wait five minutes: nothing. An hour: nothing. No errors, no lag alerts, healthy consumers — and not one window has ever emitted. A teammate suggests restarting; the restart changes nothing. The data is *there*. The windows *exist*. Some invisible condition for "you may now see results" has not been met — and no amount of waiting on the wall clock will meet it. What is Flink waiting for?
The full week 2 brief is part of LeetData Pro.
