Hudi

Hudi stands for Hadoop Update Delete Incremental. This table management technology combines multiple serialization techniques to allow columnar database performance for analytics queries while also supporting atomic, transactional updates. A typical Hudi application is a table that is updated from a CDC stream from a transactional application database. The stream is captured into a row-oriented serialization format, while the bulk of the table is retained in a columnar format. A query runs over both columnar and row-oriented files to return results for the current state of the table. Periodically, a repacking process runs that combines the row and columnar files into updated columnar files to maximize query efficiency.