Nessie is an OSS service and libraries that enable you to maintain multiple versions of your data and leverage Git-like Branches & Tags for your Data Lake. Nessie enhances the following table formats with version control techniques:
Nessie is heavily inspired by Git. The main concepts Nessie exposes map directly to Git concepts. In most cases, you simply need to replace references of files and directories in Git with Tables in Nessie. The primary concepts in Nessie are:
- Commit: Consistent snapshot of all tables at a particular point in time.
- Branch: Human-friendly reference that a user can add commits to.
- Tag: Human-friendly reference that points to a particular commit.
- Hash: Hexadecimal string representation of a particular commit.
Out of the box, Nessie starts with a single branch called
main that points to the beginning of time. A user can immediately start adding tables to that branch. For example (in pseudo code):
$ create t1 ... $ insert 2 records into t1 ... $ create t2 ... $ insert 2 records into t2 ...
A user can then use the Nessie CLI to view the history of the main branch. You’ll see that each operation was automatically recorded as a commit within Nessie:
$ nessie log hash4 t2 data added hash3 t2 created hash2 t1 data added hash1 t1 created
A user can then create a new tag referencing this point in time. After doing so, a user can continue changing the tables but that point in time snapshot will maintain that version of data.
$ nessie tag mytag hash4 $ insert records into t1 $ select count(*) from t1 join t2 .. record 1 .. .. record 2 .. .. record 3 .. .. 3 records .. $ select count(*) from t1@mytag join t2@mytag .. record 1 .. .. record 2 .. .. only 2 records ..
Data and Metadata¶
Nessie does not make copies of your underlying data. Instead, it works to version separate lists of files associated with your dataset. Whether using Spark or some other tool, each mutation operation you do will add or delete one or more files from the definition of your table. Nessie keeps tracks of which files are related to each of your tables at every point in time and then allows you to recall those as needed.
Scale & Performance¶
Nessie is built for very large data warehouses. Nessie supports millions of tables and thousands of commits/second. Because Nessie builds on top of Iceberg and Delta Lake, each table can have millions of files. As such, Nessie can support data warehouses several magnitudes larger than the largest in the world today. This is possible in large part due to the separation of transaction management (Nessie) from table metadata management (Iceberg and Delta Lake).
Nessie can be deployed in multiple ways and is composed primarily of the Nessie service, which exposes a set of REST APIs and a simple browser UI. This service works with multiple libraries to expose Nessie’s version control capabilities to common data management technologies.
Nessie was built as a Cloud native technology and is designed to be highly scalable, performant and resilient. Built on Java and leveraging Quarkus, it is compiled to a GraalVM native image that starts in less than 20ms. This makes Nessie work very well in Docker and FaaS environments. Nessie has a pluggable storage backend and comes pre-packaged with support for DynamoDB and local storage.
License and Governance¶
Nessie is Apache-Licensed and built in an open source, consensus-driven GitHub community. Nessie was originally conceived and built by engineers at Dremio.