pyremotedata 0.1.1

A package for low- and high-level high-bandwidth asynchronous data transfer

pyRemoteData

pyRemoteData is a module developed for scientific computation using the remote storage platform ERDA (Electronic Research Data Archive) provided by Aarhus University IT, as part of my PhD at the Department of Ecoscience at Aarhus University.

It can be used with any storage facility that supports SFTP and LFTP, but is only tested on a minimal SFTP server found at atmoz/sftp and on the live AU ERDA service which runs on MiG (Minimum intrusion Grid - SourceForge/GitHub) developed by SCIENCE HPC Centre at Copenhagen University.

Capabilities

In order to facility high-throughput computation in a cross-platform setting, pyRemoteData handles data transfer with multithreading and asynchronous data streaming using thread-safe buffers.

Use-cases

If your storage facility supports SFTP and LFTP, and you need high-bandwidth data streaming for analysis, data migration or other purposes such as model-training, then this module may be of use to you. Experience with SFTP or LFTP is not necessary, but you must be able to setup the required SSH configurations.

See Automated for details on how to avoid having to set up SSH configuration.

Setup

A more user-friendly setup process, which facilitates both automated as well as interactive setup is currently in development. (TODO: Finish and describe the setup process)

Installation

The package is available on PyPI, and can be installed using pip:

pip install pyremotedata

Interactive

Simply follow the popup instructions that appear once you load the package for the first time.

Automated

The automatic configuration setup relies on setting the correct environment variables BEFORE LOADING THE PACKAGE:

• PYREMOTEDATA_REMOTE_USERNAME : Should be set to your username on your remote service.
• PYREMOTEDATA_REMOTE_URI : Should be set to the URI of the endpoint for your remote service (e.g. for ERDA it is "io.erda.au.dk").
• PYREMOTEDATA_REMOTE_DIRECTORY : If you would like to set a default working directory, that is not the root of your remote storage, then set this to that (e.g. "/MY_PROJECT/DATASETS") otherwise simply set this to "/".
• PYREMOTEDATA_AUTO : Should be set to "yes" to disable interactive mode. If this is not set, or set to anything other than "yes" (not case-sensitive), while any of the prior environment variables are unset an error will be thrown.

The recommended way to avoid any SSH or environment variables setup is to use:

from pyremotedata.implicit_mount import IOHandler
with IOHandler(lftp_settings = {'sftp:connect-program' : 'ssh -a -x -i <keyfile>'}, user = <USER>, remote = <REMOTE>) as io:
    ...

Here keyfile is probably something like ~/.ssh/id_rsa.

Example

If you want to test against a mock server simply follow the instructions in tests/README.

If you have a remote storage facility that supports SFTP and LFTP, then you can use the following example to test the functionality of the module:

# Set the environment variables (only necessary in a non-interactive setting)
# If you are simply running this as a Python script, 
# you can omit these lines and you will be prompted to set them interactively
import os
os.environ["PYREMOTEDATA_REMOTE_USERNAME"] = "username"
os.environ["PYREMOTEDATA_REMOTE_URI"] = "storage.example.com"
os.environ["PYREMOTEDATA_REMOTE_DIRECTORY"] = "/MY_PROJECT/DATASETS"
os.environ["PYREMOTEDATA_AUTO"] = "yes"

from pyremotedata.implicit_mount import IOHandler

handler = IOHandler()

with handler as io:
    print(io.ls())

# The configuration is persistent, but can be removed using the following:
from pyremotedata.config import remove_config
remove_config()

Issues

This module is certainly not maximally efficient, and you may run into network- or OS-specific issues. Any and all feedback and contributions is highly appreciated.