Metadata-Version: 2.1
Name: mpipool
Version: 1.0.0a0
Summary: MPI based task scheduling for both `multiprocessing` and `concurrent.futures` interfaces.
Home-page: https://github.com/mpipool/mpipool
Author: Robin De Schepper
Author-email: robingilbert.deschepper@unipv.it
License: UNKNOWN
Keywords: mpi pool mpipool mpipool
Platform: UNKNOWN
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Science/Research
Classifier: Intended Audience :: Developers
Classifier: Topic :: Scientific/Engineering
Classifier: License :: OSI Approved :: BSD License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.6
Classifier: Programming Language :: Python :: 3.7
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Description-Content-Type: text/markdown
Requires-Dist: mpi4py (>=3.0.3)
Requires-Dist: dill (>=0.3.3)
Requires-Dist: errr (>=1.0.0)

[![Documentation Status](https://readthedocs.org/projects/mpipool/badge/?version=latest)](https://mpipool.readthedocs.io/en/latest/?badge=latest)

# About

`mpipool` offers an `MPIExecutor` class, an implementation of the
`concurrent.futures.Executor` class of the standard library.

# Example usage

```
from mpipool import MPIExecutor
from mpi4py import MPI

def menial_task(x):
  return x ** MPI.COMM_WORLD.Get_rank()

with MPIExecutor() as pool:
  pool.workers_exit()
  print("Only the master executes this code.")

  # Submit some tasks to the pool
  fs = [pool.submit(menial_task, i) for i in range(100)]

  # Wait for all of the results and print them
  print([f.result() for f in fs])

  # A shorter notation to dispatch the same function with different args
  # and to wait for all results is the `.map` method:
  results = pool.map(menial_task, range(100))

print("All processes join again here.")
```

You'll see that some results will have exponentiated either by 1, 2, ..., n
depending on which worker they were sent to. It's also important to prevent your
workers from running the master code using the `pool.workers_exit()` call. As a
fail safe any attribute access on the `pool` object made from workers will
result in them exiting anyway.

The `MPIExecutor` of mpipool is designed to function without `MPI.Spawn()`
for cases where this approach isn't feasible, like supercomputers where
`MPI.Spawn` is deliberatly not implemented (for example CrayMPI).

Therefor the pool can only use MPI processes that are spawned when the MPI world
is initialised and must be run from the command line using an MPI helper such as
`mpirun`, `mpiexec` or SLURM's `srun`:

```
$ mpirun -n 4 python example.py
```


