Skip to content
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.