...

Assumptions

• An alternative approach is to include this information into the handshake, which eliminates an extra request. However, a single partition-aware client performs a handshake with multiple server nodes. Binary configuration is guaranteed to be the same on all nodes in the cluster, so the client only needs to request it once. As a result, it is more efficient to perform one extra request than extending the handshake message.
• When "custom" mapper type is returned by the server, client is supposed to check that local BinaryConfiguration has a custom name mapper defined, and log/throw and error otherwise. We could return the type name of the custom mapper from the server, but it is of little use - we can't validate the client-side type name in non-Java clients, and even with Java client the class name can be different.

Discussion Links

// TODO Links to discussions on the devlist, if applicable.

...