Query Apache Hudi datasets

Apache Hudi is an open-source data management framework that simplifies incremental data processing. Record-level insert, update, upsert, and delete actions are processed much more granularly, reducing overhead. Upsert refers to the ability to insert records into an existing dataset if they do not already exist or to update them if they do.

Hudi handles data insertion and update events without creating many small files that can cause performance issues for analytics. Apache Hudi automatically tracks changes and merges files so that they remain optimally sized. This avoids the need to build custom solutions that monitor and re-write many small files into fewer large files.

Hudi datasets are suitable for the following use cases:

Data sets managed by Hudi are stored in Amazon S3 using open storage formats. Currently, Athena can read compacted Hudi datasets but not write Hudi data. Athena supports up to Hudi version 0.8.0 with Athena engine version 2, and Hudi version 0.14.0 with Athena engine version 3. This is subject to change. Athena cannot guarantee read compatibility with tables that are created with later versions of Hudi. For information about Athena engine versioning, see Athena engine versioning. For more information about Hudi features and versioning, see the Hudi documentation on the Apache website.

A Hudi dataset can be one of the following types:

With CoW datasets, each time there is an update to a record, the file that contains the record is rewritten with the updated values. With a MoR dataset, each time there is an update, Hudi writes only the row for the changed record. MoR is better suited for write- or change-heavy workloads with fewer reads. CoW is better suited for read-heavy workloads on data that change less frequently.

Hudi provides three query types for accessing the data:

The following table shows the possible Hudi query types for each table type.

Currently, Athena supports snapshot queries and read optimized queries, but not incremental queries. On MoR tables, all data exposed to read optimized queries are compacted. This provides good performance but does not include the latest delta commits. Snapshot queries contain the freshest data but incur some computational overhead, which makes these queries less performant.

For more information about the tradeoffs between table and query types, see Table & Query Types in the Apache Hudi documentation.

Hudi terminology change: Views are now queries

Starting in Apache Hudi release version 0.5.1, what were formerly called views are now called queries. The following table summarizes the changes between the old and new terms.