An IoT Kafka stream averages ~5000 msgs/s. Create an alert as soon as the 1-hour moving average drops below 4000 msgs/s. What should you implement?

The correct option uses Dataflow with a sliding window, which is ideal for calculating a moving average over a continuous stream. A 1-hour sliding window that advances every 5 minutes ensures the average is re-evaluated frequently (every 5 minutes) based on the most recent 1-hour data, allowing for timely alerts. The incorrect option using a fixed 1-hour window would only compute the average once every hour when the window closes, leading to significant delays in detecting a drop below the threshold. The other incorrect options involve Kafka Connect, Pub/Sub, and then either Bigtable or BigQuery, followed by Cloud Scheduler to query and alert. This approach introduces unnecessary complexity, latency, and cost for a real-time moving average calculation, as it relies on batch processing (Cloud Scheduler queries) rather than continuous stream processing.