Flink SQL & dynamic tables

Prompt A. "Explain the stream–table duality and the four RowKinds (+I, -U, +U, -D). Walk through the exact changelog that an unbounded SELECT k, COUNT(*) FROM t GROUP BY k emits when the rows (k='x'), (k='y'), (k='x') arrive."

Prompt B. "Explain the difference between an append-only query and an updating query. Give an example of each. Say which Kafka sink each one needs and why."

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

• A Flink SQL query produces a dynamic table (a result that updates continuously). The changes of this table are a changelog stream.
• Names the RowKinds: +I insert, -U retract old, +U new value, -D delete.
• Traces the GROUP BY changelog: +I(x,1), +I(y,1), then -U(x,1) +U(x,2) when the second x arrives.
• Append-only = only +I, and past output never changes (filter/project, window TVF). Updating = -U/+U/-D, and past output is revised (unbounded GROUP BY, latest-per-key).
• Append-only → plain kafka sink. Updating → upsert-kafka keyed by the grain (null = delete).
• The final materialized state after all input equals the batch answer over the same events.
• (Bonus) A window TVF aggregation is append-only (one row per closed window), unlike an unbounded GROUP BY.