Stream processing

Stream processing means computing results continuously over data as it arrives, such as counting clicks per minute or flagging fraud, rather than running a batch job over a finished dataset. The tricky part is time: events carry the moment they actually happened, but they can arrive late or out of order, for instance from a phone that was offline. Processors like Flink group events into windows by that event time, and use a watermark, a running estimate that says “I probably won’t see anything older than this now,” to decide when a window can be closed and its result sent out. Anything arriving after that is late, and you choose whether to drop it, update the result, or route it elsewhere. It’s a trade between waiting for completeness and answering quickly.