Pattern 32 / FLOW

Stream Processing

Use this when reacting to data continuously instead of waiting for batch jobs.

Pressure
Reacting to data continuously instead of waiting for batch jobs
Mechanism
Consume events as they arrive and update windows, joins, alerts, or derived state
Toll
Ordering, replay, watermarks, and exactly-once semantics are hard
Architecture plate32
Diagram for Stream Processing
Executive brief

Stream Processing fits when reacting to data continuously instead of waiting for batch jobs. Mechanism: consume events as they arrive and update windows, joins, alerts, or derived state. Use it for fraud detection, live dashboards, recommendations, telemetry, and alerting. The toll: ordering, replay, watermarks, and exactly-once semantics are hard.

Use when

Fraud detection, live dashboards, recommendations, telemetry, and alerting.

Example

Flag suspicious card transactions within seconds.

Review framing

Describe the pressure first, then the mechanism, then the cost. That keeps the design grounded.

Same pressure family

Data Processing Patterns

31MapReduce33Lambda Architecture34Change Data Capture (CDC)
← 3133 →