The Stream Processor is the real-time heart of the platform. It reads the telemetry firehose from Kafka, maintains windowed state in Flink, and serves the resulting metrics and anomalies over a small read-only query API.
Why stateful streaming
Aggregates and anomaly scores depend on recent history per device. Keeping that state in Flink (checkpointed, partitioned by device) lets the pipeline answer “what is normal for this device right now?” without round-tripping to a datastore on every event.
Could not build diagram for model "example-iot".
Reads telemetry topics and feeds the pipeline.
Computes tumbling and sliding aggregations per device.
Scores aggregates and emits alerts on outliers.
The processor exposes a read-only query surface for live metrics, per-device windowed aggregates and detected anomalies. The full contract is embedded on the Overview tab.