Data Modelling in Microsoft Fabric: implementation blueprint

Recommended architecture#

Data Modelling in Microsoft Fabric treats architecture as an explicit set of Fabric workload, storage, governance, and operating decisions. Fabric brings multiple analytical workloads together over OneLake, while each workload retains its own execution responsibilities ^{[S1] [S2]}.

Architecture and data flow#

The recommended flow is Conceptual business model → Lakehouse or Warehouse physical model → Semantic model and consumption contract. This is an authored Fabric pattern: validate it against workload-specific constraints, and use the cited product behavior as the factual boundary ^{[S1] [S2]}.

Decision framework#

Use the decision path below to make the boundary visible rather than hiding it in implementation code. Decisions that are not established by a Fabric source are labelled as pattern guidance ^{[S1] [S3]}.

Data flow and implementation#

Start with one bounded data product. Record its owners, source contract, target Fabric items, quality rules, security boundary, recovery objective, and observable completion signal. Implement the thinnest end-to-end slice before generalising the framework; this sequencing is pattern guidance, not a Fabric product guarantee ^{[S1] [S4]}.

Governance, security, and operations#

Apply workspace and item access deliberately, retain source-to-output lineage, and separate configuration from secrets. Monitor run state and capacity behavior at the same boundary used for ownership; exact controls depend on the selected Fabric workloads ^{[S3] [S4]}.

Failure modes and anti-patterns#

Avoid framework-first delivery, hidden grain or history rules, non-idempotent retries, and diagrams that imply guarantees absent from cited sources. Treat preview behavior, limits, and performance figures as source-sensitive and revalidate them during drift review ^{[S1] [S4]}.

Internals#

Architecture & design#

The design composes existing Fabric capabilities rather than introducing a separate execution service. Topic-to-capability mappings keep retrieval and validation anchored to the registry ^{[S1] [S2]}.

How it works internally#

The implementation passes versioned data and metadata contracts between the selected Fabric items. Product-specific execution details remain in the capability articles; this topic explains their architectural composition ^{[S1] [S2]}.

Performance characteristics#

Workload-specific. Performance must be established from workload-specific L4 evidence and measured against the actual capacity, data shape, concurrency, and freshness objective. No universal throughput or latency number is inferred here ^[S4].

Source legend#