Schema Registry & contracts

The platform publishes an order event. Version 1 of the Avro value schema is:

{"type": "record", "name": "Order", "namespace": "de.commerce",
 "fields": [
   {"name": "id", "type": "long"},
   {"name": "amount", "type": "double"}
 ]}

Finance now wants a currency on every order. Hundreds of consumers already read version 1. They will not all redeploy today. The subject uses the default BACKWARD compatibility mode. Evolve the schema so that the registry accepts it and old consumers keep working.

Your task — implement solution.py

Implement three functions:

• base_schema() -> dict — return the version-1 schema exactly as above.
• evolved_schema() -> dict — return a BACKWARD-compatible version-2 schema that adds a currency string field. (Think about what a new field needs, so that a reader can still read an old record.)
• produce_and_read(bootstrap_servers, topic, orders) -> dict — the input is a list of (id, amount, currency) tuples. Do these steps:
  1. Serialize each order with your evolved schema and produce it to topic.
  2. Consume the records back. Decode each record twice. Decode it once with the evolved (new) reader schema and once with the base (old) reader schema.
  3. Return a dict:

     {
       "new_reader": [{"id": ..., "amount": ..., "currency": ...}, ...],
       "old_reader": [{"id": ..., "amount": ...}, ...],  # no currency key
     }
     

     Both lists must be ordered by id ascending.

Serialize with fastavro.schemaless_writer into bytes. Send those bytes as the Kafka message value. Decode with fastavro.schemaless_reader. Pass the writer schema (evolved) and the chosen reader schema.

Why this proves the point

The old_reader list decodes new-schema bytes with the old schema. A consumer that you have not upgraded yet does exactly this in production. If your evolution is BACKWARD-safe, these decodes succeed and ignore currency. Suppose you added a required field instead. Then the old reader would fail. The registry would reject that change.

Constraints

• Do not drop or rename id or amount.
• currency must be a plain string field with a default (not a union/nullable).
• No Docker. Tests run against a native Kafka broker and the pure-Python Avro compatibility checker.