There seems to be some kind of incompatibility with parquet’s text datatype and the JSON converter.
I’ve tried the same thing with dremio’s native text datatype and a CSV PDS, it works fine.
So, the workaround would be to convert the text datatype to dremio’s native (either by binary or whatever)
and then convert it to JSON.
(Oh, and the JSON conversion also doesn’t work for VDS, because when querying a subquery, it cannot know the original SCHEMA).