ezMPI

A simple MPI-based processing pool for Python that distributes tasks across multiple processes using MPI (Message Passing Interface).

Features

• Simple API similar to Python's built-in map() function
• Automatic worker process management
• Support for complex objects via optional dill integration
• Context manager support for automatic cleanup
• Lightweight and easy to integrate into existing projects

Documentation

Full documentation is available at: https://ezmpi.readthedocs.io/

Requirements

• Python 3.8+
• MPI implementation (OpenMPI, MPICH, or similar)
• mpi4py library

Installation

Install from PyPI

pip install ezmpi

Install from source

git clone https://github.com/minask/ezmpi.git
cd ezmpi
pip install -e .

Install with development dependencies

pip install -e ".[dev]"

Install with dill support

pip install -e ".[dill]"

Quick Start

Basic Usage

from ezmpi import MPIPool

def square(x):
    return x * x

# Create a pool (must be run with mpiexec)
with MPIPool() as pool:
    results = pool.map(square, [1, 2, 3, 4, 5])
    print(results)  # [1, 4, 9, 16, 25]

Run with multiple processes:

mpiexec -n 4 python your_script.py

Using dill for Complex Objects

from ezmpi import MPIPool

def process_data(data):
    # Complex processing logic
    return sum(data) / len(data)

# Enable dill for better pickling support
with MPIPool(use_dill=True) as pool:
    results = pool.map(process_data, [[1, 2, 3], [4, 5, 6], [7, 8, 9]])
    print(results)  # [2.0, 5.0, 8.0]

MPI Execution

ezMPI requires at least 2 MPI processes (1 master + 1+ workers):

# Minimum: 2 processes
mpiexec -n 2 python script.py

# Typical: 4-8 processes
mpiexec -n 4 python script.py

# On HPC systems
srun -n 32 python script.py

Note: The master process (rank 0) coordinates tasks, all other processes are workers.

API Reference

MPIPool

class MPIPool(comm=None, use_dill=True)

Parameters:

• comm: MPI communicator (optional, defaults to MPI.COMM_WORLD)
• use_dill: Use dill for pickling instead of standard pickle (default: True)

Methods:

• map(worker, tasks): Execute worker function on each task in parallel
• close(): Shutdown worker processes
• is_master(): Check if current process is the master
• is_worker(): Check if current process is a worker

Development

Setup Development Environment

git clone https://github.com/minask/ezmpi.git
cd ezmpi
pip install -e ".[dev]"

Running Tests

# Run basic tests pytest -v

# Run MPI tests
mpiexec -n 4 pytest -v

# Run with coverage
pytest --cov=ezmpi --cov-report=html

Code Quality

# Lint code
ruff check .

# Format code
ruff format .

# Type checking
mypy src/ezmpi

# Import sorting
isort src/

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

This implementation is inspired by and adapted from similar MPI pool implementations in the scientific Python community.