Kafka's mental model is small but easy to get wrong, and getting it wrong shows up as data loss or a stalled consumer group at 3am. This sheet covers the model and the configs together.
Two pages:
The core model, with a diagram mapping partitions to consumers in a group
Producer configs, and why the safe settings are already your defaults since 3.0
Replication, ISR, and why min.insync.replicas and acks=all only work as a pair
Consumer groups, offset commits, and max.poll.interval.ms
Delivery semantics, with a straight answer on when exactly-once is worth the complexity
Retention vs compaction, partitioning strategy, schema management
CLI commands, including the consumer lag check you'll run most
Current through Kafka 4.0: ZooKeeper is removed, clusters are KRaft only, and the new KIP-848 rebalance protocol needs consumers to opt in. All flagged on the sheet.