The GraphAr Python SDK provides Python bindings for the GraphAr C++ library. It lets Python applications read GraphAr metadata, use the high-level graph APIs, and run the bundled graphar command-line tool.

This package is separate from the PySpark package in ../pyspark.

• Python >= 3.9
• pip
• CMake >= 3.15, Apache Arrow >= 12.0, and a C++ toolchain when building from source

pip install -U graphar

Verify the installation:

python -c "import graphar; print(graphar.GraphInfo)"
graphar --help

Clone the repository, then from its root:

pip install ./python

# for local development
pip install -e ./python

The project also publishes a development image:

docker run -it ghcr.io/apache/graphar-dev

Load graph metadata from a GraphAr YAML file:

import graphar

graph_info = graphar.GraphInfo.load("path/to/graph.graph.yml")

print(graph_info.get_name())
print(graph_info.get_vertex_info("person").get_type())
print(graph_info.get_edge_info("person", "knows", "person").get_edge_type())

Replace path/to/graph.graph.yml with the path to a GraphAr graph metadata file.

The Python SDK exposes the core GraphAr functionality through these modules:

• graphar.graph_info: graph, vertex, edge, property, and metadata APIs.
• graphar.high_level: high-level vertex and edge collection APIs.
• graphar.types: GraphAr enum types used by metadata and high-level APIs.

Example scripts are available in python/example:

• graph_info_example.py shows how to load graph metadata and inspect vertex and edge information.
• high_level_example.py shows how to use the high-level vertex and edge collection APIs.

The examples expect GAR_TEST_DATA to point to a directory that contains the ldbc_sample/parquet/ldbc_sample.graph.yml test graph:

bash dev/download_test_data.sh
export GAR_TEST_DATA=/tmp/graphar-testing
python python/example/graph_info_example.py
python python/example/high_level_example.py

The package installs a graphar command-line tool:

graphar --help
graphar show --path path/to/graph.graph.yml
graphar check --path path/to/graph.graph.yml

See python/src/cli/README.md for more CLI examples.

Build the Python API documentation from the python directory:

make install_docs
make docs

The generated documentation is written to python/docs.

Install test dependencies from the repository root:

pip install -e "./python[test]"

Run the Python tests from the python directory:

pytest

Some tests require GAR_TEST_DATA; use dev/download_test_data.sh if the test data is not available locally.

For general contribution guidelines, see ../CONTRIBUTING.md.

GraphAr is distributed under the Apache License 2.0. See ../LICENSE and ../NOTICE for details.