Right now there are several issues with out parquet support. Specifically, the only way to access parquet files though pure SQL is by including Hive, which has the following issues
- fairly verbose syntax
- requires you to explicitly add partitions
- does not support decimal types.
- querying tables with many partitions results in metadata operations dominating the query time (even worse when reading from S3).
It would be great to have better native support here though the new datasources API. Ideally once that is in place we can deprecate the existing ParquetRelation.