Github user dawidwys commented on the issue:
https://github.com/apache/flink/pull/6259
Hi @medcv ,
First of all thanks for the contribution. There is a couple of problems
with this implementation though:
* ConfluentSchemaRegistry follows the idea that schemas are grouped by
topic name. Providing the topic name in ctor of `SerializationSchema` will
break this assumption when used with topic pattern. I think we should somehow
enable passing schema from `KafkaProducer` to `SerializationSchema`, but it
would require far more changes.
* You assume in `RegistryAvroSerializationSchema` that schema is always
represented with a single id(`int`). That is true only for the
ConfluentSchemaRegistry, e.g. HortonworksSchemaRegistry uses
metadata-id(`long`) and version id(`int`). The idea of `SchemaCoder` was to
abstract away all "vendor" specific features.
---
