Warning: This extension currently builds with a feature branch of DuckDB. A PR is being worked on. When the PR is merged, this extension will be updated and usable from (nightly) DuckDB releases.

DuckDB extension for Apache Iceberg

Disclaimer: This extension is currently in an experimental state. Feel free to try it out, but be aware that minimal testing and benchmarking was done.

This repository contains a DuckDB extension that adds support for Apache Iceberg. In its current state, the extension offers some basics features that allow listing snapshots and reading specific snapshots of an iceberg tables.

Acknowledgments

This extension was initially developed as part of a customer project for RelationalAI, who have agreed to open source the extension. We would like to thank RelationalAI for their support and their commitment to open source enabling us to share this extension with the community.

Dependencies

building

This extension has several dependencies. Currently, the main way to install them is through vcpkg. To install vcpkg, check out the docs here. Note that this extension contains a custom vcpkg port that overrides the existing 'avro-cpp' port of vcpkg. The reason for this is that the other versions of avro-cpp have some issue that seems to cause issues with the avro files produced by the spark iceberg extension.

test data generation

To generate test data, the script in 'scripts/test_data_generator' is used to have spark generate some test data. This is based on pyspark 3.4, which you can install through pip.

Building the extension

To build the extension with vcpkg, you can build this extension using:

VCPKG_TOOLCHAIN_PATH='<path_to_your_vcpkg_toolchain>' make

This will build both the separate loadable extension and a duckdb binary with the extension pre-loaded:

./build/release/duckdb
./build/release/extension/iceberg/iceberg.duckdb_extension

Running iceberg queries

The easiest way is to start the duckdb binary produced by the build step: ./build/release/duckdb. Then for example:

> SELECT count(*) FROM ICEBERG_SCAN('data/iceberg/lineitem_iceberg', ALLOW_MOVED_PATHS=TRUE);
51793

Note that for testing, the ALLOW_MOVED_PATHS option is available. This option will ensure some path resolution is performed. This path resolution allows scanning iceberg tables that are moved, which is used during testing.

> SELECT * FROM ICEBERG_SNAPSHOTS('data/iceberg/lineitem_iceberg', ALLOW_MOVED_PATHS=TRUE);
1	3776207205136740581	2023-02-15 15:07:54.504	0	lineitem_iceberg/metadata/snap-3776207205136740581-1-cf3d0be5-cf70-453d-ad8f-48fdc412e608.avro
2	7635660646343998149	2023-02-15 15:08:14.73	0	lineitem_iceberg/metadata/snap-7635660646343998149-1-10eaca8a-1e1c-421e-ad6d-b232e5ee23d3.avro

For more examples check the tests in the test directory

Running tests

Generating test data

To generate the test data, run:

make data

Note that the script requires python3, pyspark and duckdb-python to be installed. Assuming python3 is already installed, running python3 -m pip install duckdb pyspark should do the trick.

Running unittests

make test

Running the local S3 test server

Running the S3 test cases requires the minio test server to be running and populated with scripts/upload_iceberg_to_s3_test_server.sh. Note that this requires to have run make data before and also to have the aws cli and docker compose installed.

samansmink/duckdb_iceberg