Storage tools

The goal of this project is to make it easy to work with local or cloud storage as part of a data science workflow.

Key Features

You can access a dataset with a single command ...

e.g. download_dataset azure_demo mnist/hand_drawn_digits

... or a few lines of python

Just import the storage tools core module, create a client and download your dataset.

You don't put secret keys in your code

Secret keys live in their own files and storage tools knows how to find them.

Datasets are version controlled

Storage tools makes it easy to

• manage multiple versions of a dataset and
• know which version of the dataset you are working with locally.

Install

pip install storage_tools

Storage tools conventions

We recommend that you ...

Use forward slashes on all platforms

Use forward slashes when specifying files, paths and dataset names.

Avoid whitespace and special characters in file and path names

Use folders for secrets and data

project_root
  ∟ data
  ∟ secrets
    ∟ settings.ini

Add the following to .gitignore

secrets/
data/

Add the following to settings.ini

[DEFAULT]
local_path=data

Running storage_client from your project root will read project_root/secrets/settings.ini and save all local data to project_root/data

How to use azure storage

If we follow the above conventions and have a project folder containing

project_root
  ∟ data
    ∟ mnist
      ∟ hand_drawn_digits
        ∟ digit0.png
        ∟ digit1.png
        ∟ ...
  ∟ secrets
    ∟ settings.ini
∟ main.py

where settings.ini contains

[DEFAULT]
local_path=data

[azure_demo]
storage_client=storage_tools.core.AzureStorageClient
conn_str=<A connection string to an Azure Storage account without credential>
credential=<The credentials with which to authenticate>
container=<The name of a storage container>

We can use main.py to

Create a new storage_client

storage_client=new_storage_client('azure_demo')

List files in the azure container (remote)

storage_client.ls()

List files in data (local)

storage_client.ls('local_path')

Create a new version of a dataset

storage_client.upload_dataset('mnist/hand_drawn_digits','major')

Note: If you run storage_client.ls() again, you'll see the new file in the azure container.

Download the latest version of a dataset

Feel free to delete your local copy of this dataset (from data) to download from azure storage.

storage_client.download_dataset('mnist/hand_drawn_digits')

Note: If you run storage_client.ls('local_path') again, you'll see the dataset in data.

See BlobServiceClient docs for more details on the settings used in settings.ini

• from_connection_string (conn_str and credential)
• get_container_client (container)

How to use AWS storage

It's the same as Azure except settings.ini contains

[DEFAULT]
local_path=data

[aws_demo]
storage_client=storage_tools.core.AwsStorageClient
service_name=s3
aws_access_key_id=<An AWS access key ID>
aws_secret_access_key=<An AWS access key>
bucket=<The name of an AWS bucket that the access key is allowed to read from and write to>

Developers

git config --global core.autocrlf input

conda create -n storage_tools python==3.8 -y
conda activate storage_tools
pip install fastcore nbdev jupyter
pip install boto3 azure-storage-blob

Type checking with mypy

!pip install mypy

Then from the storage_tools project folder

nbdev_build_lib
mypy storage_tools/core.py --ignore-missing-imports

For now, I'm ignoring the "Skipping analyzing 'azure': found module but no type hints or library stubs" error