Delta Sharing

Quick Answer: Delta Sharing is an open protocol for sharing live data securely across organizations without copying. Providers create shares containing tables; recipients access shared data using any compatible client (Spark, Pandas, Power BI) — no Databricks account required.

How Delta Sharing Works

Architecture

Data Provider                        Data Recipient
┌──────────────┐                    ┌──────────────────┐
│ Unity Catalog │  ── Share ──>     │ Any Delta Sharing │
│ Delta Tables  │  (no data copy)   │ Compatible Client │
│ + Shares      │                   │ (Spark, Pandas,   │
│ + Recipients  │                   │  Power BI, etc.)  │
└──────────────┘                    └──────────────────┘

Key Concepts

Provider Setup (Sharing Data)

Create a Share

-- Create a share
CREATE SHARE customer_data_share;

-- Add tables to the share
ALTER SHARE customer_data_share ADD TABLE prod.sales.orders;
ALTER SHARE customer_data_share ADD TABLE prod.sales.customers;

-- Add a table with partition filtering
ALTER SHARE customer_data_share ADD TABLE prod.sales.orders
PARTITION (region = 'US');

-- Add a view (for row-level filtering)
ALTER SHARE customer_data_share ADD TABLE prod.sales.filtered_orders;

Create a Recipient

-- Open sharing: any Delta Sharing client (non-Databricks)
CREATE RECIPIENT external_partner;

-- Databricks-to-Databricks sharing
CREATE RECIPIENT partner_workspace
USING ID '<partner-metastore-sharing-id>';

-- Grant share access to recipient
GRANT SELECT ON SHARE customer_data_share TO RECIPIENT external_partner;

Manage Shares

-- List all shares
SHOW SHARES;

-- Show tables in a share
SHOW ALL IN SHARE customer_data_share;

-- Show recipients of a share
SHOW GRANTS ON SHARE customer_data_share;

-- Remove a table from a share
ALTER SHARE customer_data_share REMOVE TABLE prod.sales.orders;

-- Drop a share
DROP SHARE customer_data_share;

Recipient Access (Consuming Shared Data)

Databricks-to-Databricks

When both provider and recipient use Databricks:

-- Recipient sees the shared data as a catalog
-- No additional client setup needed
SELECT * FROM partner_share.sales.orders;

Open Sharing (Non-Databricks Recipients)

Recipients download a credential file and use any Delta Sharing client:

# Python with delta-sharing library
import delta_sharing

# Read shared data
profile = "config.share"  # Credential file from provider
table_url = profile + "#share_name.schema_name.table_name"

# Read as Pandas DataFrame
df = delta_sharing.load_as_pandas(table_url)

# Read as Spark DataFrame
df = delta_sharing.load_as_spark(table_url)

Supported Open Sharing Clients

• Apache Spark (Python, Scala, Java)
• Pandas (Python)
• Power BI
• Tableau
• Rust, Go, Node.js, and other language connectors

Benefits of Delta Sharing

On the Exam: Know that Delta Sharing does not copy data (recipients read from the provider's storage), recipients do not need Databricks, and shares can include whole tables or filtered partitions/views.