A topic is split into partitions — ordered, append-only logs. Partitions are the unit of parallelism and ordering. Concept What to know Partition count max parallel consumers in a group; can be increased, never decreased Key hash(key) % partitions picks the partition → same key, same partition, ordered No key records spread across partitions (sticky batching) — no ordering guarantee Offset position of a record in a partition; consumers track it Replication factor copies of each partition (3 in production) min.insync.replicas replicas that must acknowledge with acks=all (usually 2) Retention retention.ms / retention.bytes — time or size based deletion Compaction cleanup.policy=compact keeps the latest value per key (changelogs, state) Choosing a key Use the entity whose events must stay…