Streaming Joins and Deduplication

A vending machine sends one event to your pipeline: card payment P-5 for 200 cents was approved at 10:04 on machine VM-3. Before that event can appear in a refund report, can you answer three questions about it? Where is VM-3 installed, did an item actually drop out of the machine, and is this the only copy of P-5 that the pipeline will ever receive?

Each question needs a different streaming operation. The site comes from a table of machines, so the stream is joined to a table. The item comes from a second stream of dispense events, so two streams are joined