The parquet-hadoop-bundle is not a real dependency but a mere packaging
of three parquet modules to create an uber jar. The Parquet community
created this artificial module on demand by
HIVE-5783 but the
benefits if any are unclear.
On the contrary using the uber dependency has some drawbacks:
- Parquet souce code cannot be attached easily in IDEs which makes debugging sessions cumbersome.
- Finding concrete dependencies with Parquet is not possible just by inspecting the pom files.
- Extra maintenance cost for the Parquet community adding additional verification steps during a release.
The goal of this JIRA is to replace the uber dependency with concrete dependencies to the respective modules: