I think the main reason is DataSourceV2 API doesn't support all V1 functionalities (for example, file scan metrics and partition pruning in subquery). Here's the commit to disable V2 datasources for those file formats by default and you can find more descriptions in the commit message: https://github.com/apache/spark/commit/ed44926117d81aa5fa8bd823d401efd235260872

Another thing is, though DataSourceV2 sounds fancy, but in theory, when reading simple files, use V1 and V2 API won't have much performance gaps.