aabadie / joblib-s3

AWS S3 storage backend for Joblib

Joblibs3

This package provides an AWS store backend for the joblib Memory object used for fast caching of computation results.

Initially, only local (or network shares) file systems were supported with joblib but now joblib offers the possibility to register extra store backends matching the provided StoreBackendBase API.

If you don't know joblib already, user documentation is located on https://pythonhosted.org/joblib

If you are only interested in computation result caching in joblib, the Memory documentation is available here.

The AWS S3 backend relies on the dask s3fs package.

Joblibs3 supports Python 2.7, 3.4 and 3.5.

Getting the latest code

To get the latest code use git:

git clone git://github.com/aabadie/joblib-s3.git

Installing joblibs3

Simply use pip:

$ cd joblib-s3
$ pip install -r requirements.txt .

Using joblibs3 to cache computation results in AWS S3

See the following example:

# Import the required packages
import numpy as np
from joblib import Memory
from joblibs3 import register_s3fs_store_backend
# Optionally, instantiate a session to use AWS_PROFILE from ~/.aws/config
# from aiobotocore.session import AioSession

if __name__ == '__main__':
    register_s3fs_store_backend()

    # we assume you S3 credentials are stored in ~/.aws/credentials, so no
    # need to pass them to Memory constructor.
    backend_options = dict(
        bucket='joblib-example',
        # Optional
        #session=AioSession(profile='my-profile'),
    )
    mem = Memory('joblib_cache', backend='s3', verbose=100, compress=True,
                 backend_options=backend_options)

    multiply = mem.cache(np.multiply)
    array1 = np.arange(10000)
    array2 = np.arange(10000)

    result = multiply(array1, array2)
    print(result)

This example is available in the examples directory.

joblib-s3's Issues

Create a new release to pypi

The version of joblibs3 available in pypi is out of date with the code in this repo. It would be great to create a release of the latest version of this library.