Home / Databricks on Azure

Databricks on AzureCheat sheetIntermediate

Advisory: sharing data from Databricks - Delta Sharing, Clean Rooms, Marketplace or copying

Teams often export files to share data, creating uncontrolled copies. Choose Databricks-to-Databricks or open Delta Sharing for live read access, Clean Rooms when neither side may see the other's raw data, Marketplace for wider distribution, and copying only when the recipient truly needs it, with governance and audit for each.

Deepak Bhardwaj2 min read

Applies toUnity Catalog Delta Sharing (Databricks-to-Databricks and open sharing), Databricks Clean Rooms, Databricks Marketplace; recipients such as Power BI, pandas, Spark and other platforms

Options

Option Recipient Data movement Fits
Delta Sharing D2D Another Unity Catalog metastore (other workspace, region, cloud, company) None; reads live tables Internal cross-region sharing, partners on Databricks
Delta Sharing open Any client with the open protocol (Power BI, pandas, Spark, other tools) None; short-lived access to files Partners not on Databricks
Clean Rooms Two or more parties running approved code on combined data No raw data exposed to the other party Joint analysis with privacy constraints
Marketplace Many consumers, public or private listings Via sharing Distributing data products
Export/copy Systems that can't use sharing Full copy Last resort; regulated exports

Decision questions

  1. Is the recipient on Databricks? → D2D sharing.
  2. Must neither party see raw data? → Clean Rooms.
  3. Does the recipient need to write or keep their own history? → They copy from the share into their own tables.
  4. How fresh must the data be? Sharing is live; exports go stale.
  5. What egress costs apply for cross-region/cloud reads? (Check current pricing; consider replicating to a nearer region.)

Governance for shares

Control How
Scope Share specific tables, partitions or views, not whole catalogs
Sensitive data Share views with masking/filtering; check what each recipient may see
Recipients Named recipients; tokens for open sharing rotated and expiring
History Choose whether to share table history (time travel / CDF)
Audit System audit logs for share access
Approval Data owner approves each new share or recipient 
CREATE SHARE IF NOT EXISTS sales_partner_share;
ALTER SHARE sales_partner_share ADD TABLE prod.sales.v_partner_orders;  -- a filtered view
CREATE RECIPIENT IF NOT EXISTS partner_contoso;   -- open sharing recipient
GRANT SELECT ON SHARE sales_partner_share TO RECIPIENT partner_contoso;

Check current support for sharing views and materialized views to open recipients.

Tip

Replace scheduled CSV exports to partners with shares where you can. You get one governed copy, live data, revocable access and an audit trail.

Checklist

  • Option chosen per recipient
  • Shares scoped to needed tables/views
  • Sensitive data filtered or masked
  • Recipient tokens managed and expiring
  • Owner approval and audit in place
  • Egress cost for cross-region access checked

Related fixes