S3 Access in Spark and Databricks

Audience: Data Owners, Data Users, and System Administrators

Content Summary: You can use a library (like Boto 3 in Python) to access standard Amazon S3 and point it at Immuta to access your data. The integration with Databricks uses a file system (is3a) that retrieves your API key and communicates with Immuta as if it were talking directly to S3, allowing users to access S3, Azure Blob, and persisted data sources through Immuta's s3p endpoint.

Note

This mechanism would never go to S3 directly. To access S3 directly, you will need to expose an S3-backed table or view in the Databricks Metastore as a source or use native workspaces/scratch paths.

Accessing Object-Backed Data Sources in Spark or Databricks

Configure Your Cluster

To use the is3a filesystem, add the following snippet to your cluster configuration:

IMMUTA_SPARK_DATABRICKS_FS_IS3A_PATH_STYLE_ACCESS=true

This configuration is needed to allow any access to is3a on Databricks 7+.

Create a Data Source

Create a Persisted data source.
Select Upload Files from the dropdown menu on the data source Overview page.
Opt to add folders in the Directory field in the Upload modal.
Click Add Files to Upload to upload your files, and then click Save.
In Databricks or Spark, write queries that access this data by referencing the S3 path (shown in the Basic Information section of the Upload Files modal above), but using the URL scheme is3a:

Limitations

This integration is only available for object-backed data sources. Consequently, all the standard limitations that apply to object-backed data sources in Immuta apply here.
Additional configuration is necessary to allow is3a paths to function as scratch paths. Contact your Immuta support professional for guidance.