I use AvroConfulent data format with schema registry to consume Kafka events to clickhouse.

• сurrently, Avro schemas are cached once resolved
• after evolving Avro schema(add fields) clickhouse doesn't fetch new schema and failed with an error on SELECT raws from Kafka table
• recreate Kafka table, change format_avro_schema_registry_url and restart clickhouse server doesn't refresh the cache
• if I create the same Table on another Kafka topic with the same format everything works correctly, and I guess that schema cache works by schema name

As I understand these lines of code and comments in PR, fetch a new Avro schema should work automatically:
https://github.com/ClickHouse/ClickHouse/pull/8953/files/927e572d39432d22ae96e087674a5124c6e2931b#diff-4c0061c616a674630b9e2e74706e6255R61
#8571

AvroConfluent only caches schemas per an instance of an InputFormat. This means SchemaRegistry will be queried each time a batch from Kafka is processed.