Serialization

Prompt A. "Explain to a Python/lake engineer why you would put Avro records on a Kafka topic. The alternative is to pickle or JSON-dump objects. Cover schema, cross-language use, evolution, and size. Also explain the difference between GenericRecord and SpecificRecord."

Prompt B. "The Flink state for a job is unexpectedly huge and slow. Explain the Kryo-fallback problem. Name the two things that you would check about the event class, without a cluster, to prove that it would NOT fall back to Kryo."

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

• Names Avro's explicit, evolvable, cross-language schema. Contrasts it with pickle and Java serialization (opaque, JVM-only).
• Distinguishes GenericRecord (map-like, untyped) from SpecificRecord (generated, typed). Notes that the wire bytes are identical.
• States that a same-schema binary round-trip is lossless (decoded.equals(original)).
• Notes that binary encoding does not embed the schema in each record (hence Registry or the container header).
• Explains Kryo fallback. The type is not an Avro or recognized type, and it is not a valid POJO. So Flink uses Kryo (slow, brittle).
• Lists Flink POJO rules: public class, public no-arg ctor, public fields or public getters/setters.
• (Bonus) Connects evolution to KFK-06 (reader/writer schema resolution).