streaming-vs-batch

name	streaming-vs-batch
description	Decide streaming vs batch honestly by the real latency need (sub-minute reaction -> streaming; hourly/daily -> batch via data-platform), then design the topology, platform choice, and delivery semantics if streaming is justified.

Streaming vs Batch

Decide by latency need

Need	Approach
Sub-minute reaction / real-time	Streaming
Hourly / daily analytics	Batch ELT -> data-platform

Streaming is operationally heavy — justify it with latency, not novelty.

If streaming

Topology: sources -> topics (partitioned/keyed) -> processors -> sinks, delivery semantic named at each hop. Platform by fit (Kafka default / Pulsar multi-tenant / Kinesis AWS-managed). Get data in via CDC/outbox, never dual-writes.