Batch vs. Streaming Workflow Processing

Decide separately from topology whether a workflow processes a known finite set of work items together or an ongoing stream as items arrive.

Independent of workflow topology (how tasks connect), decide whether a workflow processes work items in batch — a known, finite set, processed together — or streaming — an arbitrary or ongoing number of items created or retrieved as the workflow runs. The same workflow can often go either way: a marketing-email workflow over a list of storefronts could run batch (collect a storefront list, then process it) or streaming (a web crawler continuously discovers storefronts and processes each as it arrives).

Batch is typically simpler to set up and maintain, and efficient for large known volumes. Streaming suits real-time, low-latency needs but is more complex to build and operate. Choose based on whether the workload is naturally bounded-and-known ahead of time or open-ended and arrival-driven, not based on topology.

---

Sources

• Prompt Engineering for LLMs (John Berryman), ch. 9