...
Often, when reading an input topic, the key is expected to be null, but there's actually no way to represent this fact in Consumed, leading to confusing type signatures down the topology. For example, you might use the BytesSerde, but then you have a KStream<Bytes,...>. When maintaining a large application, this becomes a hazard, since you'd need to "be really careful" not to try and dereference the key at any point, since you actually know it's always null. One might use the BytesSerde, but then you have a KStream<Bytes,...>. When maintaining a large application, this becomes a hazard, since you'd need to "be really careful" not to try and dereference the key at any point, since you actually know it's always null.
Public Interfaces
I propose to add the following new SerDe:
...
Code Block | ||||
---|---|---|---|---|
| ||||
static public final class VoidSerde extends WrapperSerde<Void> { public VoidSerde() { super(new VoidSerializer(), new VoidDeserializer()); } } public class VoidSerializer implements Serializer<Void> { @Override public byte[] serialize(String topic, Void data) { return new byte[0]; } } public class VoidDeserializer implements Deserializer<Void> { @Override public Void deserialize(String topic, byte[] data) { return null; } } |
Proposed Changes
I want to add VoidSerde
to main SerDe collection Serdes
class.
Compatibility, Deprecation, and Migration Plan
There is no compatibility issues here.
Rejected Alternatives
One might use the BytesSerde, but then you have a KStream<Bytes,...>. When maintaining a large application, this becomes a hazard, since you'd need to "be really careful" not to try and dereference the key at any point, since you actually know it's always nullNullSerdes
, NothingSerdes
naming was rejected during KIP discussion.