The root cause for DRILL-6216 is due to a recent change made as part of
https://issues.apache.org/jira/browse/DRILL-6049. With this PR a default field
name for values ValueVector inside any NullableValueVector was introduced which
is $values$ . Before this PR the values ValueVector field name was same as
the field name of actual NullableValueVector holding it . In the load method
of certain ValueVectors like BigIntVector there is an equality check for the
ValueVector field name and metadata.name_part name .
In setup where Drillbit and DrillClient are running in different version
(between 1.12 and 1.13) the equality check in load method will fail. For
example: When server is running 1.13 and client is on 1.12, in that case the
record batch from server side will come with NullableValueVector (NV1 with
field name as num_val) but with it's values ValueVector field name as $values$.
When on client side corresponding NullableValueVector (NV2) is created it will
use the actual field name (num_val) for values ValueVector. After calling load
on received NullableValueVector NV2 with metadata information from server that
internally alls load on values ValueVector and the check fails since ($values$
Since the change is in template of ValueVector, to fix this issue both client
and server needs to identify their respective version on which they are running
and choose the field name for values ValueVector correspondingly. Given
DRILL-6049 touches huge sets of files I am also not sure what are the other
impacts with it. It would be great to discuss on how we should proceed with
this issue before making any further change.