Traditionally, batch processing of files has been the default approach in many on-premise solutions. Files are dropped into a directory, collected on a schedule, and processed in groups. This model works fine for systems where latency is not critical, but once you start asking “How do we make this real-time?” the story becomes more interesting. At first, my answer was polling. After all, if we want near real-time, we can just keep checking the directory at short intervals.