Is there any way to map the keyspace attribute on the UDT annotation?

[Adrian Cole]

Hi, there.

We use your driver in Zipkin, which is a distributed tracing system. Particularly in our most recent schema, we take advantage of UDTs for things like service endpoints.

We bake-in a schema with indexing, and by default everything goes to "zipkin2". Users can override this, and by basic string templating, we'll use whatever they want.

There's a glitch with the UDTs, because they are mapped with an annotation attribute:

@UDT(keyspace = "zipkin2_udts", name = "endpoint")
static final class EndpointUDT {

So, as far as we know, we *also* have to install the schema to whatever the annotation says, then do a dance like below to remap it later:

    TypeCodec<EndpointUDT> endpointCodec = mapping.udtCodec(EndpointUDT.class);

    KeyspaceMetadata keyspace =

        session.getCluster().getMetadata().getKeyspace(session.getLoggedKeyspace());

    session.getCluster().getConfiguration().getCodecRegistry()

      .register(

        new TypeCodecImpl<>(keyspace.getUserType("endpoint"), EndpointUDT.class, endpointCodec))

When people look at our schema they are surprised to see the *extra* copy of it "zipkin2_udts", and in

some cases they can't install something like this anyway. For example, their team may have a naming

policy that requires a prefix.

I was hoping that 3.4 would fix this, but it looks like the new mapping functionality doesn't work for UDTs

https://github.com/openzipkin/zipkin/pull/1891

So, anyway we have some users who literally can't use cassandra due to this, so looking for a way out.

Is there some other way to register a java type mapped to a UDT without relying on anything hard-coded?

Perhaps a mapping? I'm getting desperate enough to try and generate the annotation at runtime :)

-A