We found that resolvingDecoder could not provide bidirectional compatibility between different version of schemas.
Especially for records, for example:
Say, node A knows only the first schema and node B knows the second schema, and the second schema has more fields.
Any data generated by node B can be resolved by first schema 'cause the additional field is marked as skipped.
But data generated by node A can not be resolved by second schema and throws an exception "Don't know how to handle excess fields for reader."
This is because data is resolved exactly according to the auto-generated codec_traits which trying to read the excess field.
The problem is we just can not only ignore the excess field in record, since the data after the troublesome record also needs to be resolved.
Actually this problem stucked us for a very long time.