Stream Processing with Kafka Streams — Apache Kafka
Kafka Streams is a Java library (no separate cluster) for stateful stream processing. Scale by running more instances — partitions are shared like a consumer group. Abstraction Meaning KStream stream of events (inserts) KTable changelog → latest value per key GlobalKTable full copy of a small table on every instance Stateless ops filter, mapValues, flatMapValues, branch/split Stateful ops groupByKey, count, aggregate, reduce, joins Windows tumbling, hopping, sliding, session State lives in local RocksDB stores backed by compacted changelog topics, so it survives restarts. Interactive queries expose that state over your own API. Alternatives: Apache Flink, ksqlDB (SQL over streams).
Key Takeaways
Kafka Streams is a library — scale it by running more app instances.
KStream models events; KTable models the latest state per key.
Re-keying triggers a repartition topic so related records meet on one instance.
Windowed aggregations need a grace period for late events.