Skip to content

Apache Iceberg

Apache Iceberg is a Apache Software Foundation project that provides a rich, relatively new table format. It provides:

  • Single table ACID transactions
  • Scalable metadata
  • Appends via file addition
  • Updates, deletes and merges via single record operations

Iceberg Extension Points

Iceberg exposes two primary classes for working with datasets. These are Catalog and TableOperations. Nessie implements each. These classes are available in the the Iceberg source code and are available directly in Iceberg releases (eg spark-runtime, spark3-runtime, flink-runtime).

Iceberg Snapshots

Iceberg supports the concept of snapshots. Snapshots are point in time versions of a table and are managed as part of each commit operation. Snapshots are limited to single table versioning. Nessie versions and commits provide a broader set of snapshot capabilities as they support multiple tables. Nessie is happy to coexist with Iceberg Snapshots. When working with Nessie, Iceberg snapshots will also be versioned along the rest of Iceberg metadata within the Nessie commit model.

Automatic Snapshot Import

We are exploring the creation of a tool where a user can import table snapshots across multiple Iceberg tables into a single Nessie repository to capture historical data snapshots (interleaved across time).

Hive Compatibility

SerDe Compatibility

There is currently work in progress that provides updates to the existing HiveCatalog so that it can recognize a Nessie Catalog pointer and reroute the metadata lookup to a Nessie server. This allows existing workflows to continue to work while also moving versioning responsibilities to Nessie.

Hive Table Cloning

In Nessie, we plan to add a capability to automatically update one or more Hive Metastore servers (including AWS Glue) every time a Iceberg table is updated in Nessie so that legacy systems can still be exposed to Nessie updates, even if the HMS Bridge service we offer is not used.

HMS Bridge Compatibility

There is a plan to expose Nessie native tables automatically to Hive users when they are interacting through the HMS bridge.