milvus-storage Python Bindings

Python bindings for milvus-storage, a high-performance storage engine using Apache Arrow Parquet as the underlying format, optimized for analytical workloads.

Features

• High Performance: Built on Apache Arrow and Parquet for efficient columnar storage
• Packed Storage: Groups narrow columns together to reduce file count and control memory usage
• Cloud Native: Support for AWS S3, Azure Blob Storage, Google Cloud Storage, and more
• Zero-Copy: Efficient data transfer between Python and C++ using Arrow C Data Interface
• Pythonic API: Clean, intuitive interface following Python best practices
• Type Safe: Full PyArrow integration with schema validation

Installation

Prerequisites

• Python 3.8 or later
• C++ compiler (for building from source)
• Conan (for C++ dependencies)

Install from Source

# Clone the repository
git clone https://github.com/milvus-io/milvus-storage.git
cd milvus-storage

# Build the C++ library
cd cpp
make python-lib
cd ..

# Install Python package
cd python
pip install -e .

Examples

See the examples/ directory for complete working examples:

• basic_write.py - Writing data to milvus-storage
• basic_read.py - Reading data with full table scan, filtering, and column projection

API Reference

Writer

Create a writer to store data in milvus-storage format.

Writer(path: str, schema: pa.Schema, properties: Optional[Dict[str, str]] = None)

Methods:

• write(batch: pa.RecordBatch) - Write a record batch
• flush() - Flush buffered data to storage
• close() -> str - Close writer and return manifest JSON

Reader

Read data from milvus-storage datasets.

Reader(
    manifest: str,
    schema: pa.Schema,
    columns: Optional[List[str]] = None,
    properties: Optional[Dict[str, str]] = None
)

Methods:

• scan(predicate: Optional[str] = None) -> pa.RecordBatchReader - Full table scan
• take(indices: Union[List[int], np.ndarray], parallelism: int = 1) -> pa.RecordBatch - Random access (not yet implemented)
• get_chunk_reader(column_group_id: int) -> ChunkReader - Get chunk reader for column group

Properties

Configuration properties for milvus-storage.

Common Properties:

Testing

Run tests with pytest:

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=milvus_storage --cov-report=html

Development

Building from Source

# Build C++ library
cd cpp
make python-lib

# Install in development mode
cd ../python
pip install -e ".[dev]"

# Run tests
pytest tests/

Requirements

• Python >= 3.8
• pyarrow >= 10.0.0
• numpy >= 1.20.0

License

Apache License 2.0. See LICENSE for details.

Contributing

Contributions are welcome! Please see the main repository for contribution guidelines.

Support

• GitHub Issues: milvus-storage issues
• Documentation: GitHub Repository