Overview#
Data mesh is an operating model, not a Fabric feature, so this pattern treats mesh terms as inference and grounds only the Fabric mechanics in sources. Fabric domains, workspaces, OneLake, shortcuts, governance, and capacities provide the product constructs used to implement domain ownership, data-product discovery, federated governance, and compute isolation [S1] [S2] [S3] [S4].
Fabric Jumpstart can help a domain team learn by deploying Microsoft-published accelerators and tutorials into a sandbox workspace, but a Jumpstart is not itself a data-product contract or governance model [S8]. The domain must still own the source, quality rules, endorsement, and capacity impact of anything it promotes [S4] [S5] [S8].
Core concepts#
A Fabric domain groups workspaces into a logical business area and provides a level above workspaces for delegated management and governance [S1] [S4]. OneLake remains a single tenant-wide lake, so a domain boundary is an ownership and governance boundary rather than a separate physical data lake [S2]. Shortcuts let a consuming domain read producer-owned data in place without duplicating it [S3].
Capacities give the cost and throttling boundary that makes domain autonomy operational. Because throttling is enforced per capacity, separating domain workspaces across capacities can contain noisy workloads and support chargeback [S5] [S6].
Domain operating model#
| Responsibility | Central platform team | Domain team | Source |
|---|---|---|---|
| Tenant standards | Defines tenant-wide governance, labels, endorsement policy, and baseline controls | Follows standards and requests justified exceptions | [S4] |
| Workspace ownership | Provides naming, lifecycle, and capacity assignment guardrails | Owns domain workspaces, items, and access review | [S1] [S5] |
| Data product publication | Defines certification process and catalog expectations | Publishes endorsed/certified products with clear ownership | [S4] |
| Cross-domain consumption | Requires no-copy patterns where possible | Uses shortcuts to consume producer-owned data in place | [S2] [S3] |
| Cost and reliability | Defines capacity strategy and chargeback model | Operates within assigned capacity envelope | [S5] [S6] |
| Accelerators | Provides sandbox rules for Jumpstart evaluation | Tests Jumpstarts before adapting ideas into governed products | [S8] |
How it works and best practices#
Map each genuine business data owner to a Fabric domain, place its curated workspaces under that domain, and publish trusted products through endorsement and catalog discovery [S1] [S4]. Use shortcuts for cross-domain consumption so consumers read governed producer data rather than creating their own copies [S3]. Use capacity placement to separate production domains, development work, and shared platform services where isolation matters [S5] [S6].
Use Jumpstart in a domain sandbox when a team needs a working example of a scenario before building its own governed product [S8]. Because Jumpstart deploys items into a workspace, the platform team should treat it like any other workspace content: review ownership, item names, data files, security, and lifecycle before promotion [S4] [S8].
domain_blueprint:
name: finance
workspaces:
- finance-dev
- finance-curated-prod
- finance-consumption
capacity:
prod: finance-f64
dev: finance-f8
publication_controls:
endorsement_required: true
shortcut_consumption_preferred: true
jumpstart_sandbox_only: true
Inference: this is a governance blueprint, not a product schema. The constructs are sourced: domains/workspaces, OneLake, shortcuts, governance, capacities, and Jumpstart deployment behavior [S1] [S2] [S3] [S4] [S5] [S8].
What goes wrong#
Creating one physical lake per domain is the opposite of the Fabric model and reintroduces duplication that OneLake removes [S2]. Calling an organization a mesh without real domain ownership, product quality processes, or chargeback discipline also adds governance overhead without operational benefit; that operating-model warning is inference, while the Fabric controls are cited [S1] [S4] [S5].
A subtler failure is giving every consumer workspace broad contributor access to producer workspaces. OneLake security separates control-plane and data-plane permissions, and workspace Admin/Member/Contributor roles can bypass granular OneLake security roles, so producer workspace membership should stay tight [S7]. Use shortcuts and read-oriented permissions for consumption rather than making consumers co-owners of producer items [S3] [S7].
Internals#
Architecture & design#
Domains sit above workspaces, workspaces hold items, OneLake provides the shared namespace, and capacities provide compute resources to assigned workspaces [S1] [S2] [S5]. Governance settings operate across tenant, domain, workspace, and item scopes depending on the control [S4].
How it works internally#
Shortcut reads resolve through identity and permission behavior while leaving the target data in place [S3]. OneLake security separates control-plane permissions from data-plane access, and some workspace roles can bypass granular OneLake security roles [S7].
Jumpstart installation uses the fabric-jumpstart package and can deploy bundled assets into a workspace, including small sample files into a lakehouse when the Jumpstart includes file-upload configuration [S8]. Inference: in a domain model, that makes Jumpstart best suited to domain learning and pattern evaluation, not direct publication of production products [S4] [S8].
Performance characteristics#
The most important performance boundary is capacity isolation: one domain's saturated capacity does not throttle unrelated capacities in the tenant [S6]. Shortcut performance depends on the source, cache behavior, and engine path, so measured cross-domain shortcut latency should be validated for critical products [S3].
Worked example#
Finance can own certified revenue tables in a Finance domain workspace, Sales can consume those tables through a shortcut, and both domains can run on separate capacities so Sales exploration does not throttle Finance publishing workloads [S1] [S3] [S5] [S6].
Producer: Finance curated workspace
- Owns gold Delta tables
- Applies endorsement and domain metadata
- Keeps Contributor membership limited
Consumer: Sales analytics workspace
- Reads Finance product through a shortcut
- Builds Sales-specific semantic model/reporting
- Runs on Sales capacity, not Finance capacity
The same domain can evaluate a Jumpstart in a dev workspace first, then rebuild the useful pattern against approved internal sources and endorsement rules before it becomes a certified data product [S4] [S8].