rugo 0.1.2

Parquet Reader

rugo

A lightning-fast Parquet file reader built with C++ and Cython, optimized for ultra-fast metadata extraction and analysis.

🚀 Features

• 🚀 Lightning-fast metadata reading - 10-50x faster than PyArrow for metadata operations
• 🏗️ C++ core with Cython bindings - Maximum performance with Python convenience
• 📊 Complete schema information - Physical types, logical types, and statistics
• 🔄 Schema conversion - Convert rugo schemas to orso format (optional)
• 🔬 Zero dependencies - No runtime dependencies for core functionality
• ✅ PyArrow compatible - Validated results, drop-in replacement for metadata operations

📦 Installation

# Basic installation (coming soon to PyPI)
pip install rugo

# With orso schema conversion support
pip install rugo[orso]

From Source

# Clone the repository
git clone https://github.com/mabel-dev/rugo.git
cd rugo

# Create virtual environment
python -m venv venv
source venv/bin/activate

# Install build dependencies
pip install setuptools cython

# Build the extension
make compile

# Install in development mode
pip install -e .

Requirements

• Python 3.9+
• C++ compiler with C++17 support
• Cython (for building from source)

🔧 Usage

Reading Parquet Metadata

Rugo provides blazing-fast access to Parquet file metadata without the overhead of loading actual data:

import rugo.parquet as parquet_meta

# Extract complete metadata from a Parquet file
metadata = parquet_meta.read_metadata("example.parquet")

print(f"Number of rows: {metadata['num_rows']}")
print(f"Number of row groups: {len(metadata['row_groups'])}")

# Analyze row groups and column statistics
for i, row_group in enumerate(metadata['row_groups']):
    print(f"Row Group {i}:")
    print(f"  Rows: {row_group['num_rows']}")
    print(f"  Size: {row_group['total_byte_size']} bytes")
    
    for col in row_group['columns']:
        print(f"    Column: {col['name']}")
        print(f"    Physical Type: {col['type']}")
        print(f"    Logical Type: {col.get('logical_type', '(none)')}")
        print(f"    Nulls: {col['null_count']}")
        print(f"    Min: {col['min']}")
        print(f"    Max: {col['max']}")
        
        # Check for bloom filter availability
        if parquet_meta.has_bloom_filter(col):
            print(f"    Has bloom filter: Yes")
        else:
            print(f"    Has bloom filter: No")

Advanced Features

Schema Analysis

Extract detailed schema information including both physical and logical types:

import rugo.parquet as parquet_meta

metadata = parquet_meta.read_metadata("example.parquet")
for col in metadata['row_groups'][0]['columns']:
    print(f"{col['name']}: {col['type']} -> {col.get('logical_type', '(inferred)')}")
    # Example output:
    # name: BYTE_ARRAY -> STRING
    # timestamp: INT64 -> TIMESTAMP_MILLIS
    # price: DOUBLE -> (inferred)

Bloom Filter Testing

Quickly test if values might exist in columns without reading the actual data:

import rugo.parquet as parquet_meta

metadata = parquet_meta.read_metadata("example.parquet")

for col in metadata['row_groups'][0]['columns']:
    if parquet_meta.has_bloom_filter(col):
        # Test if a value might be present
        might_exist = parquet_meta.test_bloom_filter(
            "example.parquet",
            col['bloom_offset'],
            col['bloom_length'], 
            "search_value"
        )
        if might_exist:
            print(f"Value might be in column {col['name']}")
        else:
            print(f"Value definitely not in column {col['name']}")

Schema Conversion to Orso

Convert rugo parquet schemas to orso format:

from rugo.converters.orso import rugo_to_orso_schema, extract_schema_only
import rugo.parquet as parquet_meta

# Read parquet metadata
metadata = parquet_meta.read_metadata("example.parquet")

# Convert to orso RelationSchema
orso_schema = rugo_to_orso_schema(metadata, "my_table")

print(f"Schema: {orso_schema.name}")
print(f"Columns: {len(orso_schema.columns)}")
print(f"Estimated rows: {orso_schema.row_count_estimate}")

# Access individual columns
for column in orso_schema.columns[:3]:
    print(f"{column.name}: {column.type} ({'nullable' if column.nullable else 'not null'})")

# Or get a simplified column mapping
schema_info = extract_schema_only(metadata, "simple_name")
print("Column types:", schema_info['columns'])

Note: Orso conversion requires the optional orso dependency:

pip install rugo[orso]

### Metadata Structure

The `read_metadata()` function returns a dictionary with the following structure:

```python
{
    "num_rows": int,           # Total number of rows in the file
    "row_groups": [            # List of row groups
        {
            "num_rows": int,           # Rows in this row group
            "total_byte_size": int,    # Size in bytes
            "columns": [               # Column metadata
                {
                    "name": str,           # Column name/path
                    "type": str,           # Physical type (INT64, BYTE_ARRAY, etc.)
                    "logical_type": str,   # Logical type (STRING, TIMESTAMP_MILLIS, etc.)
                    "min": any,            # Minimum value (decoded)
                    "max": any,            # Maximum value (decoded)
                    "null_count": int,     # Number of null values
                    "bloom_offset": int,   # Bloom filter offset (-1 if none)
                    "bloom_length": int,   # Bloom filter length (-1 if none)
                }
            ]
        }
    ]
}

⚡ Performance

Rugo is specifically designed for blazing-fast Parquet metadata operations:

• ⚡ 10-50x faster than PyArrow for metadata extraction
• 🧠 Minimal memory footprint - Direct binary parsing without intermediate objects
• 🚀 Lightning startup - Fast imports with optimized compiled extensions
• 📊 Efficient statistics - Decode min/max values without loading columns

Benchmarks

Run performance comparisons yourself:

make test  # Includes comprehensive PyArrow vs Rugo benchmarks

Why is Rugo so fast?

• Direct C++ implementation of Parquet metadata parsing
• Zero-copy binary protocol parsing
• Optimized Thrift deserialization
• No Python object overhead during parsing

🛠️ Development

Building from Source

# Install development dependencies
make update

# Build Cython extensions
make compile

# Run tests
make test

# Run linting
make lint

# Check type hints
make mypy

# Generate coverage report
make coverage

Project Structure

rugo/
├── rugo/
│   ├── __init__.py          # Main package
│   └── parquet/             # Parquet decoder implementation
│       ├── metadata.cpp     # C++ metadata parser
│       ├── metadata.hpp     # C++ headers
│       ├── thrift.hpp       # Thrift protocol implementation
│       └── metadata_reader.pyx  # Cython bindings
├── tests/
│   ├── data/                # Test Parquet files
│   └── tests
├── Makefile                 # Build automation
├── setup.py                 # Build configuration
└── pyproject.toml           # Project metadata

Testing

The test suite includes:

• Validation tests - Compare output with PyArrow
• Performance benchmarks - Speed comparisons
• Edge case handling - Various Parquet file formats

# Run all tests
make test

# Run specific test
python -m pytest tests/test_compare_arrow_rugo.py -v

Code Quality

We maintain high code quality with:

• Linting: ruff, isort, pycln
• Type checking: mypy
• Formatting: ruff format
• Cython linting: cython-lint

🤝 Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature-name
3. Make your changes and add tests
4. Run the test suite: make test
5. Run linting: make lint
6. Submit a pull request

Development Setup

# Clone your fork
git clone https://github.com/yourusername/rugo.git
cd rugo

# Set up development environment
python -m venv venv
source venv/bin/activate
make update
make compile
make test

📊 What Rugo Does

✅ Currently Supported:

• Fast Parquet metadata extraction - Schema, statistics, row group information
• Logical type detection - STRING, TIMESTAMP, DECIMAL, etc.
• Bloom filter testing - Value presence checks without data scanning
• Statistics decoding - Min/max values properly typed and decoded
• Cross-platform support - Linux, macOS

🎯 Focus Areas: Rugo is laser-focused on being the fastest Parquet metadata reader available. It doesn't try to be everything to everyone - it does one thing exceptionally well.

🐛 Known Limitations

• Metadata-only: Rugo focuses on metadata extraction, not data reading
• C++ compiler required: Building from source requires C++17 compiler
• Parquet-specific: Designed specifically for Parquet format

📄 License

Licensed under the Apache License 2.0. See LICENSE for details.

👨‍💻 Authors

• Justin Joyce - Initial work - joocer

🙏 Acknowledgments

• Built on top of the Apache Parquet format specification
• Inspired by PyArrow's parquet module design
• Uses optimized Thrift binary protocol for metadata parsing
• Performance insights from the Apache Arrow community

📈 Roadmap

Core Focus: Fastest Parquet Metadata Reader

• Lightning-fast metadata extraction
• Complete schema information with logical types
• Bloom filter support
• Advanced statistics (histograms, sketches)
• Parquet format validation

---

For more information, visit the GitHub repository or open an issue.