All terms

Streaming Data

Continuously generated events processed in near real time — logs, clicks, sensor readings — rather than only in daily batches.

Data Science1 min read

Definition

Streaming systems (Kafka, Pub/Sub, Kinesis) move events as they happen. Consumers update dashboards, triggers, or online features with low latency.

Exactly-once processing, late events, and windowing are core streaming challenges.

In simple terms

Batch is checking your mailbox once a day. Streaming is standing at the door as letters arrive.

Where you see it

  • Fraud alerts within seconds of a transaction.
  • Live recommendation features from recent clicks.

How it works

  1. 1.Produce events

    Apps write to a log or topic.

  2. 2.Consume

    Workers process windows of events.

  3. 3.Update state

    Sinks write to stores, features, or alerts.

Why it matters

  • Many ML products need fresh signals — streaming keeps features from going stale.

Often confused

  • Streaming always replaces batch.

    Batch remains simpler and cheaper for many analytics and training jobs; hybrid is common.