Publish SAP Databricks Views As Custom Data Products In SAP Business Data Cloud

What's New

SAP Business Data Cloud (BDC) Connect SDK now supports publishing in SAP Databricks – Views as custom data products — not just physical tables. This unlocks a clean, flexible pattern for sharing data from Databricks environments that use native Databricks types not directly supported by CAP's CDS type system.

The Solution: A View as the Publication Layer

You now create a Databricks View that handles all type conversions at the SQL layer, then publish that view directly. No duplication, no extra tables.

Step 1 — Create a Type-Compatible View

CREATE VIEW demo.default.customer_view AS

SELECT

    customer_id,

    CAST(loyalty_level AS INT)      AS loyalty_level,   — TINYINT  → cds.Integer

    CAST(age_group AS INT)          AS age_group,        — SMALLINT → cds.Integer

    CAST(credit_score AS DOUBLE)    AS credit_score,     — FLOAT    → cds.Double

    HEX(profile_hash)               AS profile_hash,     — BINARY   → cds.String

    CAST(country_code AS STRING)    AS country_code,     — CHAR(2)  → cds.String

    customer_name,

    CAST(created_ts AS TIMESTAMP)   AS created_ts,       — TIMESTAMP_NTZ → cds.Timestamp

    TO_JSON(favorite_colors)        AS favorite_colors,  — ARRAY    → JSON String

    TO_JSON(preferences)            AS preferences,      — MAP      → JSON String

    TO_JSON(address)                AS address           — STRUCT   → JSON String

FROM demo.default.customer_raw;

Type mapping reference:

Databricks Type CDS-Compatible Type Conversion

`TINYINT` / `SMALLINT`	`cds.Integer`	`CAST(… AS INT)`
`FLOAT`	`cds.Double`	`CAST(… AS DOUBLE)`
`BINARY`	`cds.String`	`HEX(…)`
`CHAR(n)`	`cds.String`	`CAST(… AS STRING)`
`TIMESTAMP_NTZ`	`cds.Timestamp`	`CAST(… AS TIMESTAMP)`
`ARRAY<T>`, `MAP<K,V>`, `STRUCT`	`cds.String` (JSON)	`TO_JSON(…)`

Step 2 — Install the SDK

pip install sap-bdc-connect-sdk

Step 3 — Register the Share with ORD Metadata

you can refer this blog in detail about Data Product Sharing

from bdc_connect_sdk.auth import BdcConnectClient, DataBricksClient
bdc_connect_client = BdcConnectClient(

    DataBricksClient(dbutils, “bdc-partner-connect-tenant-<your-tenant-id>”)

)
share_name = “demo_dbx_view”
bdc_connect_client.create_or_update_share(

    share_name,

    {

        “@openResourceDiscoveryV1”: {

            “title”: “Demo DBX View”,

            “shortDescription”: “Databricks Data Product”,

            “description”: “Databricks data product from a View without a primary key”

        }

    }

)

Step 4 — Generate and Register the CSN Schema

from bdc_connect_sdk.utils import csn_generator
csn_schema = csn_generator.generate_csn_template(share_name)

bdc_connect_client.create_or_update_share_csn(share_name, csn_schema)

The SDK inspects the view schema and auto-generates the CDS Schema Notation (CSN) definition. You can review and adjust it before registering.

Step 5 — Publish

bdc_connect_client.publish_data_product(share_name)

The data product is now available in SAP Business Data Cloud for downstream consumption.

Why This Matters

No primary key required. Views do not need a primary key — a common blocker with raw or analytical tables.

Data governance at the SQL layer. The view becomes your data contract. Consumers see a clean, typed schema; the raw table stays untouched. Masking, filtering, and reshaping happen in SQL.

No data redundancy. Type conversion happens on-the-fly in the view — no intermediate tables needed.

Flexible schema evolution. Decouple the physical table from the published schema. Evolve either side independently.

Before vs. After

	Before	Now
Supported source	Delta tables only	Delta tables and Views
Type compatibility	Must match CDS types in the table	Cast incompatible types in the view
Primary key	Required	Not required for views
Data duplication	Sometimes unavoidable	Not needed

Publishing Databricks Views as SAP BDC custom data products is especially valuable in medallion architectures where bronze/silver layers carry raw SAP Databricks types that are not directly shareable. The view pattern gives data teams a lightweight, maintainable bridge between the Databricks schema world and the SAP data product ecosystem — without modifying or duplicating source data.

Source link