Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

2) Besides validation of the schema on the producer side and keeping client compatibility with that we also need a way for consumers once subscribed to a topic (from a group perspective) to read in the key/value schema information.  This could just be part of the OffsetResponse A Guide To The Kafka Protocol#OffsetResponse

3) We need to get should lump the client compatibility kit (some thoughts on that https://github.com/stealthly/kafka-clients/wiki/Compatibility-Test-Kit )  working work into this effort too.  

4) The topic needs to also contain encoder/decoder being used and NOT let the API override that, ever.

5) This structure for the topic schema (encoder, decoder, schemas[id]=schema) could just be a new ByteBuffer structure which we can store and transport the same. the schema list could get large and dynamic so this should be plug able too. Out of the box we need to support something functional for development and testing.6) This design also can work into an implement (attributed set of schemas for security) in regards ACL.  The Kafka server can do whatever it is trying to-do with the data.  The expectation here is that it can reach into a message and grab out the fields it knows about without having to know anything about the message.  The plugable schema storage layer should handle this nuance of mapping the calling applications understanding of the map and what the schema has stored.  The relationship between the callers mapped name == the schema mapped name is 1:many. e.g. the server wants to .get("identity") which has a structure it is expecting.  The kafka server should be able to get this ByteBuffer so it can read it regardless of the implementation.  Having this feature allow that to be more seamless.to implement Authorizations for the data in regards ACL (at least the security bits on the data to validate). 

67) I think built in initial support for Avro would be awesome and probably account for the largest percentage of existing Kafka installations.  We could use Camus encoders/decoders for avro https://github.com/linkedin/camus/blob/master/camus-kafka-coders/src/main/java/com/linkedin/camus/etl/kafka/coders/KafkaAvroMessageEncoder.java / https://github.com/linkedin/camus/blob/master/camus-kafka-coders/src/main/java/com/linkedin/camus/etl/kafka/coders/KafkaAvroMessageDecoder.java and json https://github.com/linkedin/camus/blob/master/camus-kafka-coders/src/main/java/com/linkedin/camus/etl/kafka/coders/JsonStringMessageDecoder.java but with a layer of faster xml databind over it e.g. 

...