The DP-600 blueprint is split across 3 domains. Weights are the official share of the exam; see the official exam guide for the authoritative breakdown.
What you must be able to do. Choose the least-privileged workspace role for an authoring task, put a workspace under branch-based version control and reconcile it with the right Git action, promote content with deployment pipelines, and apply the correct security tier and governance label.
In one sentenceThe operations layer: least-privilege roles, branch-based Git version control reconciled with Commit and Update, deployment pipelines for stage promotion, and the correct security and governance scope.
Recall check: answer these from memory first
- What is the lowest Fabric workspace role that lets a person create, edit, and publish reports, semantic models, and a Lakehouse without managing users or settings, and what does the next role up add?
- A workspace is connected to a Git branch and the Source control pane flags it as behind the branch after a teammate's merge. Which action applies the merged changes to the workspace, and what does the opposite action do?
- When does the exam want deployment pipelines versus Git integration for managing changes to semantic models and reports?
What it tests. Securing and operating the solution once it exists. Workspace-level and item-level access controls, the cumulative Fabric workspace role hierarchy, and least-privilege role choice; row-level, column-level, object-level, and file-level access control across lakehouses, warehouses, and semantic models; sensitivity labels and endorsement for governance; and the development lifecycle, meaning workspace version control with Git integration and Power BI Desktop projects, deployment pipelines with deployment rules across development, test, and production, plus impact analysis and the XMLA endpoint.
How to study it. Memorise the cumulative workspace role hierarchy until least-privilege choice is automatic: Viewer reads only, Contributor adds full create and edit rights for items, Member adds sharing and user management, and Admin adds workspace governance, so Contributor is the lowest authoring role and only Admin can change settings, manage roles, or delete the workspace. Learn the lifecycle split that the exam tests directly: workspace Git integration with an Azure DevOps or hosted repository gives branch-based version control with pull-request review and is bidirectional, where Commit pushes workspace edits up and Update pulls merged branch changes down, while deployment pipelines promote content across development, test, and production stages with deployment rules. Practise applying sensitivity labels and endorsing items, and rehearse the security tiers so you can map row-level, column-level, object-level, and file-level control to lakehouses, warehouses, and semantic models.
Easy to confuse
- Deployment pipelines versus Git integration. Git integration puts a workspace under branch-based source control with pull-request review and is the native version-control mechanism; deployment pipelines promote content across development, test, and production stages with deployment rules. Version control and review means Git; stage promotion means pipelines.
- Contributor versus Member workspace role. Contributor is the lowest role that can create, edit, and publish all item types but cannot manage workspace access or settings; Member adds sharing the workspace and managing users with lower roles. The gap is user and sharing management, not authoring rights, so least-privilege authoring is Contributor.
Worked example from the DP-600 bank
lock_openFree sampleMaintain a Data Analytics Solutionmedium
An analytics team wants to put a "Microsoft Fabric" workspace under source control so that changes to its semantic models and reports are tracked, branched, and reviewed by pull request. They already use a hosted repository for the rest of their code. Which configuration meets this requirement with the least operational overhead?
- AExport each item manually as a file, commit those files to a local folder repository, and re-import them by hand whenever a teammate needs the latest version.
- BUse deployment pipelines to move the items between development, test, and production stages, treating each stage promotion as the team's version history.
- CConnect the workspace to an Azure DevOps Git repository through workspace Git integration so items sync to a branch and are reviewed through pull requests.check_circle Correct
- DApply a sensitivity label and an endorsement to every item so that each change is recorded against the item's governance metadata for later review.
Use workspace Git integration with an Azure DevOps repository to put Fabric items under branch-based source control with pull-request review. Git integration links a Fabric workspace to a Git branch so supported items are serialised, committed, and merged through standard pull requests, which is the native mechanism for version control rather than manual export or stage promotion.
Why A is wrong: Manual export and re-import is exactly the unmanaged process Git integration replaces, so it carries high overhead and gives no branch or pull-request workflow for the workspace.
Why B is wrong: Deployment pipelines promote content across stages but do not provide branching or pull-request review, so they address release flow rather than the source-control requirement stated.
Why C is correct: Workspace Git integration with an Azure DevOps repository syncs supported items to a branch and supports pull requests, meeting the tracking and review requirement with built-in tooling.
Why D is wrong: Labels and endorsements describe governance and trust, not change history, so they record nothing about edits and cannot deliver branching or pull-request review.
What you must be able to do. Choose the right Fabric store and ingestion surface for the workload, transform and cleanse data correctly, build a star schema with the right enrichment, and select, filter, and aggregate with SQL, KQL, or DAX as the scenario demands.
In one sentenceThe preparation core: pick the right store and ingestion path, cleanse and shape correctly, build a star schema with aggregation, and query with the right language.
Recall check: answer these from memory first
- A SQL Server runs inside the company data centre with no public endpoint and must feed a Lakehouse through Dataflows Gen2. Which component lets the cloud Fabric service reach it, and which gateway types do not fit?
- A daily-grain fact table is read almost always as monthly category totals but detail must stay for drill-through. What do you add to cut repeated scans without losing detail?
- In a Dataflows Gen2 query you run Remove Duplicates with no column selected and nothing is removed. Why, and what one change collapses each repeated key to a single row?
What it tests. The heaviest domain by weight: getting data into Microsoft Fabric and shaping it before any modelling. Creating connections and managing gateways and credentials; discovering data through the OneLake catalog and Real-Time hub and choosing between Fabric stores for a workload; ingesting with shortcuts, Dataflows Gen2, pipelines, and notebooks; transforming with views, functions, and stored procedures in a Warehouse or the SQL analytics endpoint; building a star schema and enriching by denormalising and aggregating; cleansing duplicates, missing values, and types; and querying with the Visual Query Editor, SQL, KQL against an Eventhouse, and DAX against a semantic model.
How to study it. This is nearly half the exam, so it earns the most time, and most of it is store and transformation choice plus correct cleansing. Learn the store decision cold: a Lakehouse for files and Spark with a SQL analytics endpoint for read, a Warehouse for a fully transactional T-SQL store, an Eventhouse and KQL database for real-time event data. Learn the ingestion choice: an on-premises data gateway to reach a private database with no public endpoint, shortcuts to reference data in place without copying, Dataflows Gen2 for low-code Power Query transformation, notebooks for Spark-scale or code-first work. For modelling at the storage layer, drill the star schema and the aggregation pattern, adding a pre-aggregated summary table beside the detailed fact so summary reads hit the aggregate while drill-through still reaches detail. Practise the documented cleansing behaviours, such as selecting the key column before Remove Duplicates in Dataflows Gen2 so deduplication compares only the key. For querying, rehearse the DAX table functions inside EVALUATE, knowing FILTER returns whole rows meeting a condition with all native columns preserved.
Easy to confuse
- Lakehouse versus Warehouse. A Lakehouse stores files and Delta tables for Spark and low-code work and exposes a read-only SQL analytics endpoint; a Warehouse is a fully transactional T-SQL store that supports writes, views, functions, and stored procedures through SQL. Need full T-SQL DML and procedural objects means Warehouse; need files, Spark, and shortcuts means Lakehouse.
- Dataflows Gen2 versus notebooks for preparation. Dataflows Gen2 give a low-code Power Query experience for connecting, shaping, and loading with minimal overhead; notebooks give code-first Spark for large-scale or complex transformations. Choose Dataflows Gen2 when the transformation is standard and low-code fits, notebooks when you need Spark scale or custom code.
- On-premises data gateway versus a virtual network or personal gateway. An on-premises data gateway installed inside the corporate network and bound to the connection is the relay for a private database with no public endpoint in a shared production context; a personal gateway is for individual use and a virtual network gateway targets Azure virtual networks, not an on-premises data-centre server.
Worked example from the DP-600 bank
lock_openFree samplePrepare Datamedium
An analytics engineer must ingest data from a SQL Server instance that runs on a server inside the company data centre, with no public endpoint, into a "Lakehouse" using "Dataflows Gen2". The "Microsoft Fabric" service in the cloud cannot reach the server directly. Which component must be installed and configured so that the connection can reach the on-premises database?
- AAn on-premises data gateway installed on a machine inside the company network, registered to the tenant and bound to the connection.check_circle Correct
- BA cloud connection defined in the "Microsoft Fabric" portal, which routes the query to the private server using the engineer's stored organisational account credentials.
- CA virtual network data gateway provisioned in the tenant, which peers with the on-premises subnet and forwards each query to the local SQL Server instance.
- DA personal gateway in standard mode shared across the workspace, which lets every workspace member reuse one connection to the local SQL Server instance.
Source data from a private on-premises database into Fabric by installing and binding an on-premises data gateway to the connection. The on-premises data gateway is the relay that sits inside the corporate network and tunnels query and result traffic between the cloud Fabric service and a source that has no public endpoint, which is why it is mandatory for on-premises ingestion.
Why A is correct: The on-premises data gateway runs inside the private network and relays queries between the cloud service and the local server, which is exactly the bridge a non-public source requires.
Why B is wrong: A cloud connection only reaches sources that already have a public or service endpoint, so it cannot by itself bridge into a private on-premises network that the service cannot see.
Why C is wrong: A virtual network data gateway reaches sources inside an Azure virtual network, not a physical on-premises data centre, so it does not bridge to a server with no cloud presence.
Why D is wrong: The personal mode gateway is single-user and cannot be shared across a workspace, so it is the wrong gateway type for a team ingestion pipeline.
What you must be able to do. Choose the correct storage mode for a freshness and size requirement, lift a model past its ceiling with the right format, honour chained row-level security correctly, and write advanced DAX with the right performance choices.
In one sentenceThe modelling core: the right storage mode for freshness and size, the large semantic model storage format to lift the ceiling, chained row-level security honoured per identity, and correct advanced DAX.
Recall check: answer these from memory first
- An enterprise semantic model must report over billions of Delta rows in a Lakehouse with near real-time freshness, in-memory speed, and no copy or scheduled refresh. Which storage mode fits, and why do import, DirectQuery, and dual not?
- An import model on Fabric capacity has grown past the default 1 GB single-model ceiling and refreshes now fail. Which single change lets it grow toward the capacity maximum without re-architecting?
- A composite model takes a DirectQuery connection to a certified semantic model that already enforces row-level security. How does that source RLS behave through the composite model for each consumer?
What it tests. Turning prepared data into a semantic model that is correct, fast, and fresh. Choosing a storage mode and designing composite models; implementing a star schema with relationships, bridge tables, and many-to-many; writing DAX with variables and functions, calculation groups, dynamic format strings, and field parameters; identifying use cases for the large semantic model storage format; improving DAX and query performance for enterprise scale; and configuring Direct Lake behaviour including fallback and refresh, choosing between Direct Lake on OneLake and on SQL endpoints, and implementing incremental refresh.
How to study it. Drill the storage-mode decision until it is automatic, because the exam leans on it. Direct Lake reads Delta tables straight from OneLake with in-memory speed and near real-time freshness without copying data or running a scheduled refresh, which is the answer when an enterprise model needs import-grade speed over billions of OneLake rows with live freshness. Import caches for speed but needs a scheduled refresh; DirectQuery queries live but loses in-memory speed; dual blends import and DirectQuery. Learn the large semantic model storage format as the single change that lifts an import model past its small default per-model ceiling toward the capacity maximum, whether the size comes from raw growth or accumulating incremental-refresh partitions, since incremental refresh controls which partitions reload but accumulating partitions still count against the ceiling. Understand composite models chained by DirectQuery to a published semantic model, which honour the source model's row-level security in each consumer's identity without recreating roles. Practise the advanced DAX surface, calculation groups, dynamic format strings, and field parameters, and rehearse incremental refresh policy design.
Easy to confuse
- Direct Lake versus import versus DirectQuery. Direct Lake reads OneLake Delta tables in memory for import-grade speed with near real-time freshness and no scheduled refresh or copy; import caches data and needs a scheduled refresh; DirectQuery queries the source live but answers from the source, not memory. Live freshness plus in-memory speed over OneLake without copy is Direct Lake's job alone.
- Large semantic model storage format versus changing storage mode. The large semantic model storage format raises an import model's in-memory ceiling from the small default toward the capacity maximum without re-architecting, so an oversized import model or accumulating incremental-refresh partitions keep working; switching to DirectQuery removes the cache but loses in-memory speed and is not what a size-ceiling requirement asks for.
Worked example from the DP-600 bank
lock_openFree sampleImplement and Manage Semantic Modelshard
An enterprise "semantic model" must report over billions of Delta rows held in a "Lakehouse" in "OneLake". The team needs near real-time freshness without scheduled refreshes, in-memory query speed for reports, and no duplication of the data into the model. Which storage mode should they choose for these tables?
- AImport mode, because it loads the Delta tables into the model for fast in-memory queries and refreshes them on a schedule so reports stay close to the source data.
- BDirectQuery mode, because it queries the Delta tables in place for near real-time freshness and answers report queries from memory without copying any data into the model.
- CDual storage mode, because it both imports the Delta tables for in-memory speed and queries them live for freshness while avoiding any copy of the data into the model.
- D"Direct Lake" mode, because it reads the Delta tables directly from "OneLake" with in-memory speed and reflects source changes without copying data or running scheduled refreshes.check_circle Correct
Choose Direct Lake when an enterprise model needs in-memory speed and near real-time freshness over OneLake Delta data without copying or scheduled refresh. Direct Lake reads Parquet-backed Delta tables straight from OneLake and pages the columns into memory at query time, so it gives import-grade performance while reflecting source updates without a scheduled refresh and without duplicating the data into the model.
Why A is wrong: Import mode copies the Delta tables into the model and depends on scheduled refresh, which both duplicates the data and breaks the near real-time requirement, so it does not fit.
Why B is wrong: DirectQuery avoids duplication and stays fresh, but it sends each query to the source rather than serving results from memory, so it cannot deliver the in-memory query speed required.
Why C is wrong: Dual mode lets a table behave as import or DirectQuery per query, but its import side still caches a copy of the data, so it does not avoid duplication the way the requirement demands.
Why D is correct: "Direct Lake" loads Delta data from OneLake into memory on demand and tracks source changes, delivering import-like speed with near real-time freshness and no data duplication, which meets every stated need.