scio
scio copied to clipboard
SMB module is pulling all storage implementations
Depending on scio-smb pulls transitively all storage implementation dependencies for:
- parquet
- json
- avro
- tensorflow
TensorFlow dependencies alone are ~200Mb.
Users should only have the desired storage implementation in their classpath
SMB should either:
- be available in the storage implementation themselves
- mark storage impl dependencies as provided
- have specific impl modules like