Kafka Streams의 윈도잉과 상태 저장 집계

Operations like map and filter are stateless. Aggregations such as counting or summing require state that persists across records.

Kafka Streams manages this state for you in local state stores.

Before aggregating you group records by key with groupByKey or groupBy.

A simple aggregation: count orders per customer. The result is a KTable backed by a state store.

Use aggregate for custom accumulation. You provide an initializer and an adder function.

Often you want aggregations over a time window, e.g. orders in the last 5 minutes. Kafka Streams supports tumbling, hopping, and session windows.

A tumbling window is fixed-size and non-overlapping. Each record belongs to exactly one window.

Hopping windows have a size and a smaller advance, so they overlap. A record can fall into multiple windows.

Session windows group records separated by gaps of inactivity — ideal for user sessions where activity bursts then pauses.

State stores are backed by compacted changelog topics in Kafka. If an instance fails, its state is rebuilt from the changelog on another instance.

Interactive Queries let you read state store values directly from the application, exposing aggregations via a REST endpoint without an external database.

Windowed, stateful aggregations turn raw event streams into live metrics. Group by key, choose a window type, aggregate, and optionally expose the state via interactive queries.

You learned windowing and stateful aggregations.

• Group with groupByKey before aggregating.
• Use count or aggregate to build a KTable.
• Tumbling, hopping, and session windows handle time differently.
• State stores are fault-tolerant via changelog topics.