Storage Options

How to authenticate using Activeloop storage, AWS S3, and Google Cloud Storage.

Deep Lake datasets can be stored locally, or on several cloud storage providers including Deep Lake Storage, AWS S3, Microsoft Azure, and Google Cloud Storage. Datasets are accessed by choosing the correct prefix for the dataset path that is passed to methods such as deeplake.load(path), and deeplake.empty(path). The path prefixes are:

Storage Location	Path	Notes
Local	/local_path
Deep Lake Storage	hub://org_id/dataset_name
Deep Lake Managed DB	hub://org_id/dataset_name	Specify runtime = {"tensor_db": True} when creating the dataset
AWS S3	s3://bucket_name/dataset_name	Dataset can be connected to Deep Lake via Managed Credentials
Microsoft Azure (Gen2 DataLake Only)	azure://account_name/container_name/dataset_name	Dataset can be connected to Deep Lake via Managed Credentials
Google Cloud	gcs://bucket_name/dataset_name	Dataset can be connected to Deep Lake via Managed Credentials

Connecting Deep Lake datasets stored in your own cloud via Deep Lake Managed Credentials is required for accessing enterprise features, and it significantly simplifies dataset access.

Authentication for each cloud storage provider:

Activeloop Storage and Managed Datasets

In order to access datasets stored in Deep Lake, or datasets in other clouds that are managed by Activeloop, users must register and authenticate using the steps in the link below:

User Authentication

AWS S3

Authentication with AWS S3 has 4 options:

1. Use Deep Lake on a machine in the AWS ecosystem that has access to the relevant S3 bucket via AWS IAM, in which case there is no need to pass credentials in order to access datasets in that bucket.

2. Configure AWS through the cli using aws configure. This creates a credentials file on your machine that is automatically access by Deep Lake during authentication.

3. Save the AWS_ACCESS_KEY_ID ,AWS_SECRET_ACCESS_KEY , and AWS_SESSION_TOKEN (optional) in environmental variables of the same name, which are loaded as default credentials if no other credentials are specified.

4. Create a dictionary with the AWS_ACCESS_KEY_ID ,AWS_SECRET_ACCESS_KEY , and AWS_SESSION_TOKEN (optional), and pass it to Deep Lake using:

   Note: the dictionary keys must be lowercase!

deeplake.load('s3://<bucket_name>/<dataset_name>', 
               creds = {
                        'aws_access_key_id': <your_access_key_id>,
                        'aws_secret_access_key': <your_aws_secret_access_key>,
                        'aws_session_token': <'your_aws_session_token'>, # Optional
                        }
)

endpoint_url can be used for connecting to other object storages supporting S3-like API such as MinIO, StorageGrid and others.

Custom Storage with S3 API

In order to connect to other object storages supporting S3-like API such as MinIO, StorageGrid and others, simply add endpoint_url the the creds dictionary.

deeplake.load('s3://...', 
               creds = {
                        'aws_access_key_id': <your_access_key_id>,
                        'aws_secret_access_key': <your_aws_secret_access_key>,
                        'aws_session_token': <'your_aws_session_token'>, # Optional
                        'endpoint_url': 'http://localhost:8888'
                        }
)

Microsoft Azure

Authentication with Microsoft Azure has 4 options:

1. Log in from your machine's CLI using az login.

2. Save the AZURE_STORAGE_ACCOUNT, AZURE_STORAGE_KEY , or other credentials in environmental variables of the same name, which are loaded as default credentials if no other credentials are specified.

3. Create a dictionary with the ACCOUNT_KEY or SAS_TOKEN and pass it to Deep Lake using:

   Note: the dictionary keys must be lowercase!

deeplake.load('azure://<account_name>/<container_name>/<dataset_name>', 
               creds = {
                        'account_key': <your_account_key>, 
                        #OR
                        'sas_token': <your_sas_token>,
                        }
)

Google Cloud Storage

Authentication with Google Cloud Storage has 2 options:

1. Create a service account, download the JSON file containing the keys, and then pass that file to the creds parameter in deeplake.load('gcs://.....', creds = 'path_to_keys.json') . It is also possible to manually pass the information from the JSON file into the creds parameter using:

   deeplake.load('gcs://.....', creds = {information from the JSON file})

2. Authenticate through the browser using deeplake.load('gcs://.....', creds = 'browser'). This requires that the project credentials are stored on your machine, which happens after gcloud is initialized and logged in through the CLI.

   1. After this step, re-authentication through the browser can be skipped using: deeplake.load('gcs://.....', creds = 'cache')