The log — events, state, retention and replay

Prompt A. "Explain why a Kafka partition is a log and not a queue. Explain how 'the current state' relates to 'the events'. Use the Iceberg-snapshot analogy. What plays the role of a snapshot id? What plays the role of time travel?"

Prompt B. "A new analytics team wants to build a table from a topic. The topic is one year old. Explain exactly how they rebuild the table from scratch. Include how they start 'from 09:00 last Monday' and not from the very beginning. Also explain what can stop them from going back that far."

Rubric — a strong explanation hits these (5–7):

• Defines a partition as an append-only, immutable, ordered log with monotonic offsets.
• Distinguishes events (immutable facts) from state (a fold/projection over events).
• Maps offset ⇄ Iceberg snapshot id and replay ⇄ time travel.
• Explains replay from an offset (assign + read forward).
• Explains replay from a timestamp via offsets_for_times (first offset ≥ ts).
• Notes retention (time/size) and that replay only reaches the earliest retained offset.
• (Bonus) Notes per-key ordering within a partition and compaction as "latest per key".