[jira] [Created] (KAFKA-7371) Finally deprecate org.apache.kafka.common.serialization.ExtendedSerializer

["Jouni (JIRA)" <ji...@apache.org>]

Jouni created KAFKA-7371:
----------------------------

             Summary: Finally deprecate org.apache.kafka.common.serialization.ExtendedSerializer
                 Key: KAFKA-7371
                 URL: https://issues.apache.org/jira/browse/KAFKA-7371
             Project: Kafka
          Issue Type: Wish
          Components: streams
    Affects Versions: 2.0.0
            Reporter: Jouni


As mentioned in the javadocs "Once Kafka drops support for Java 7, the serialize()} method introduced by this interface will be added to Serializer with a default implementation so that backwards compatibility is maintained. This interface may be deprecated once that happens.". Support for Java 7 was already dropped in Kafka 2.0.0 but this hasn't yet happened.

The problem is that some out-of-project external serializers (for example, org.springframework.kafka.support.serializer.JsonSerializer, quite commonly used) already do add message headers when using producer API, one of those being __TypeId__, which contains the java class name. But when using streams DSL, there's no way to either access or modify those headers, and according to KIP-244, they get just copied to the sink. Also, because in RecordCollectorImpl.send there are calls

final byte[] keyBytes = keySerializer.serialize(topic, key);
final byte[] valBytes = valueSerializer.serialize(topic, value);

and not

final byte[] keyBytes = keySerializer.serialize(topic, headers, key);
final byte[] valBytes = valueSerializer.serialize(topic, headers, value);

which would be possible when the plain Serializer gets the default method added. So, currently, there's no way for anyone to write a serializer that modifies the headers if necessary when using streams.

In my case, the problem occurred when transforming an object from input stream to a different type of object in output stream. Took a while to debug where did those (wrong) headers come from, and either disable adding those headers on producer side, or as I happened to be using Processor API, modify myself the headers in ProcessorContext.

An unfortunate side-effect of two different projects making decisions affecting each other. Not exactly a bug in either one, but a really big nuisance to find out what's happening. I'd prefer things working as much as possible mostly out-of-the-box. Ok, API changes must sometimes just be made.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)