For Databricks Data Engineer Professional candidates, Databricks renamed Delta Sharing to OpenSharing in June 2026 following the release of the open-source OpenSharing standard. The queue topic “Delta Sharing Governance” still reflects terminology that many teams, integrations, and runbooks use, so the useful way to handle the transition is to connect the names rather than pretend the older one disappeared overnight. Existing Delta Sharing concepts—providers, shares, recipients, governed assets, and cross-platform exchange—remain part of the architectural lineage while current Databricks documentation increasingly uses OpenSharing.
The governance challenge is unchanged: data owners need to expose selected assets beyond their normal workspace boundary without turning object storage credentials into the sharing model. That challenge belongs naturally inside Databricks data engineering and Unity Catalog governance, because OpenSharing relies on cataloged assets, explicit grants, auditable sharing relationships, and clear responsibility for what crosses organizational boundaries.
The rename matters most at the boundary between people and automation
Human-facing documentation should move toward OpenSharing so engineers can find current product guidance and understand the long-term direction. Programmatic surfaces, historical Terraform resources, API paths, or existing recipient configurations may still contain Delta Sharing names. An enterprise migration plan should distinguish terminology cleanup from protocol or configuration changes.
That avoids unnecessary breakage. Renaming a runbook heading is low risk. Renaming internal object identifiers, automation modules, service accounts, or integration contracts can be disruptive if downstream systems depend on them. Teams should update language where it improves clarity while preserving stable technical identifiers until the platform and their own tooling provide a tested migration path.
Sharing begins with the provider’s Unity Catalog security model
A provider should share only assets that are intentionally governed in Unity Catalog. Ownership, USE CATALOG, USE SCHEMA, table permissions, views, and other catalog controls define who is allowed to prepare or administer shared content. OpenSharing then extends selected assets to a recipient relationship; it does not make the internal governance model optional.
The key design question is not “can this table be shared?” but “what is the smallest stable data product the recipient actually needs?” Sharing raw operational tables can expose internal schemas, identifiers, and change patterns that were never designed as external contracts. Curated views or purpose-built tables can create a cleaner boundary, especially when the provider expects multiple recipients with different needs.
Recipient identity determines the trust model
Databricks-to-Databricks sharing can use identities that are native to the platform, while open recipients use bearer-style recipient credentials to access shared data through compatible clients. Those modes create different operational responsibilities. In a Databricks-native relationship, governance can integrate more directly with account and workspace identity. Open recipients require careful credential lifecycle management because the recipient credential itself becomes a sensitive access mechanism.
Treat recipient creation like issuing an external credential. Record the business owner, intended organization, allowed assets, expiration or rotation procedure, incident contact, and offboarding trigger. Avoid generic recipients reused by several organizations because shared identity weakens attribution and makes revocation more disruptive.
Provider grants do not become recipient-side governance
A subtle governance point is that policies protecting an asset inside the provider’s metastore do not automatically govern what the recipient does after receiving the shared data. Databricks documentation specifically distinguishes provider-side controls from recipient-side controls for cases such as attribute-based access control. The provider decides what crosses the boundary; the recipient is responsible for governing the copy or view within its own environment.
That has contractual consequences. If a dataset contains fields subject to residency, purpose limitation, retention, or downstream-use restrictions, technical sharing must be paired with an agreement that expresses those obligations. Catalog permissions can enforce who may publish the asset; they cannot enforce every legal duty once another organization legitimately receives it.
Cross-region and cross-regulatory-domain sharing needs explicit review
Databricks has expanded Databricks-to-Databricks sharing across regulatory domains. That increases flexibility but also makes location and compliance assumptions more important. The fact that the platform can establish a sharing relationship does not mean every dataset is approved to cross the associated jurisdictional or regulatory boundary.
Before enabling such a share, identify the storage regions, governing entities, data classifications, residency requirements, and contractual restrictions involved. Security review should also consider whether recipient-side processing creates derived data that falls under the same restrictions. Architecture diagrams need to show the sharing boundary as a real data-flow edge, not merely as a permission entry.
OpenSharing should expose governed products, not internal implementation details
A strong sharing interface behaves like a data product. Stable names, documented semantics, quality expectations, schema-change rules, and ownership all matter. If a shared table changes columns every sprint, recipients will build brittle integrations. Delta schema evolution becomes even more important when the consumer is outside the provider’s normal release process.
Backward-compatible additions are usually easier to absorb than destructive renames or type changes. For major changes, providers may need versioned assets or a deprecation period. The sharing technology makes distribution efficient; it does not remove the need for API-like contract management around the data itself.
AI assets make sharing governance broader than tables
The Databricks governance model increasingly includes AI assets alongside traditional data. That expands the questions a sharing program must ask: is the asset a table, a model, a feature set, or another governed object; what intellectual property does it contain; what training or inference rights apply; and can a recipient combine it with other datasets? The intersection with Unity Catalog means sharing governance should involve AI platform owners as well as data engineers when model-related assets are in scope.
Model and AI sharing can introduce additional concerns around license terms, model provenance, evaluation evidence, and downstream safety. Those issues are not solved by the transport protocol. OpenSharing supplies a controlled exchange mechanism; organizational policy still defines which AI assets are appropriate to exchange and under what terms.
Auditing needs both technical events and business ownership
Audit logs can show actions such as recipient management and share changes, but logs are only useful when teams can map identifiers back to accountable owners. Every production share should therefore have a human or team owner, a business purpose, an expected recipient, and a review date. Without that context, a technically valid share can persist long after the project that created it has ended.
Periodic access review should ask whether the recipient still needs the asset, whether the asset still contains the intended fields, whether new sensitive columns have been added, and whether the recipient credential or relationship is healthy. This is especially important for shared views built on evolving source tables, where upstream schema changes can expand the exposed surface unexpectedly.
The safest migration is a governance refresh, not just a vocabulary update
Use the Delta Sharing-to-OpenSharing rename as a reason to review the sharing estate. Inventory providers, shares, recipients, credentials, external organizations, asset classifications, and owners. Update human-facing terminology, but also look for stale recipients, overly broad assets, undocumented contracts, and weak schema-change practices. That turns a naming transition into a meaningful control improvement.
OpenSharing remains valuable because it separates governed exchange from direct object-storage credential distribution. Combined with Databricks catalog controls and clear recipient management, it can support cross-team and cross-organization data products without giving away the underlying storage account. The governance standard should focus on that boundary: what leaves, who receives it, how the contract changes, and who is accountable throughout the relationship.
OpenSharing and clean rooms solve different collaboration problems
A direct share is appropriate when the recipient is allowed to receive the governed asset and use it under the agreed terms. A clean room is more appropriate when parties need to collaborate or compute across sensitive datasets without broadly exposing the underlying records. Conflating the two can lead to oversharing: the fact that two organizations need a joint analysis does not necessarily mean either should receive the other’s raw table.
Choose the collaboration primitive from the trust model. If the business relationship permits persistent recipient access to the data product, OpenSharing may be the simpler design. If policy requires controlled queries, approved notebooks, or privacy-preserving collaboration, evaluate clean-room patterns. Governance starts by deciding what the other party is allowed to learn, then choosing the technical mechanism that enforces that boundary.
Recipient offboarding deserves the same rigor as onboarding
Sharing programs often spend considerable effort approving a new external recipient and much less effort removing one. Offboarding should revoke credentials or recipient relationships, confirm that no replacement integration still depends on them, and document the effective end date. If contractual requirements demand deletion of previously received data, that obligation must be handled with the recipient because revoking future access does not erase copies already delivered.
Automate review dates where possible. A quarterly certification of active recipients, owners, shared assets, and business purposes can expose stale relationships before they become permanent. The technical share list then becomes an accurate representation of current business relationships rather than an archive of every collaboration the organization has ever had.
Naming should be explicit during the transition
For internal standards, a temporary formulation such as “OpenSharing (formerly Delta Sharing)” is clearer than using either term without context. It helps operators connect current Databricks documentation with existing recipient configurations, historical audit records, and older automation. Once the organization has completed its documentation and tooling transition, the legacy name can be retired from human-facing material while preserved only where an interface still requires it.