Yes, you can migrate Azure Data Factory pipelines to Fabric Data Factory, and Microsoft gives you three supported routes: mount-and-assess, the built-in upgrade experience, or a scripted PowerShell/CI-CD flow. Complex or low-parity pipelines still need a manual rebuild. Triggers stay disabled by default during migration, and expect linked services, datasets, and global parameters to remap into connections, inline activity properties, and variable libraries.
TL;DR:
- Most Azure Data Factory pipelines can be migrated to Fabric using mount-and-assess, built-in upgrade, or scripted PowerShell methods, with manual rebuild needed for low-parity pipelines.
- Fabric replaces datasets with inline activity properties and linked services with workspace-level Fabric connections that favor managed identity, influencing how assets are remapped during migration.
- Inventory must catalog all assets, run readiness assessments, and plan for remapping connections, especially for linked services and integration runtimes, to prevent mid-cutover failures.
- Cost shifts from pay-per-use in Azure Data Factory to a shared capacity model in Fabric, requiring careful capacity planning to avoid performance issues across workloads like Power BI and pipelines.
- Security transitions to workspace-level governance with centralized credentials, and triggers remain disabled post-migration until explicitly re-enabled for safe validation.
Table of Contents
- Data factory to Fabric: which migration path fits your environment?
- What actually changes: Fabric Data Factory vs Azure Data Factory
- What should you inventory before migrating?
- How do linked services and integration runtimes map to Fabric?
- Which transformation approach replaces mapping data flows?
- Testing and cutover: what’s the safe sequence?
- What separates a smooth migration from a costly rebuild?
- How is Fabric Data Factory architecture put together?
- How do you optimise performance in Fabric Data Factory?
- What does migration change about cost and pricing?
- How do you secure a Fabric Data Factory environment?
- What migration issues come up most often?
- How do you monitor and govern Fabric Data Factory at scale?
- When should you bring in outside help?
- Get migration help from DataTaal
- Sources
- FAQ
Data factory to Fabric: which migration path fits your environment?
Choosing the right route depends on how much parity your pipelines have with Fabric-native features and how much risk you can tolerate during cutover.
Mounting is your discovery step. It attaches a read-only reference to your existing ADF instance inside a Fabric workspace, letting you inventory pipelines and run the readiness assessment without touching production. Mounting alone changes nothing in the source factory, so use it whenever you need an asset catalogue before committing to a plan.
The built-in upgrade experience is the low-effort path for pipelines with high feature parity. It works well for incremental migration, letting you move a handful of pipelines at a time while validating each batch. Microsoft’s own migration planning guidance treats this as the default choice for anything that maps cleanly.
The PowerShell route, driven by the Microsoft.FabricPipelineUpgrade module, suits teams running scripted, bulk, or CI/CD-integrated migrations. It exports Fabric resources automatically, but it cannot create Fabric connections on its own. You must supply a resolutions file mapping old linked service identifiers to new connection IDs, as detailed in the PowerShell upgrade tutorial.

Manual rebuild remains the right call for low-parity pipelines or when you want to modernise deliberately rather than replicate legacy behaviour.
What actually changes: Fabric Data Factory vs Azure Data Factory
The differences that trip up migrations aren’t cosmetic. They change how you write, monitor, and pay for pipelines.
Fabric removes the reusable dataset object entirely. Instead of pointing an activity at a shared dataset definition, you configure connection details inline on the activity itself, which the comparison documentation confirms as a deliberate architectural shift. Fewer reusable objects means less indirection, but it also means your inventory of “datasets” from ADF won’t map one-to-one into Fabric.
Linked services give way to Fabric connections, which live at the workspace level and favour managed identity over stored credentials. That encourages reuse across pipelines rather than duplicating connection definitions.
Key shift to plan around: OneLake integration lets Fabric read data via live shortcuts rather than copying it, which is a genuine architectural change, not a rebrand of Blob storage.
- Datasets → inline activity properties
- Linked services → Fabric connections
- Global parameters → variable libraries
- Standalone storage → OneLake with shortcuts
Fabric also runs on a SaaS capacity model rather than ADF’s pay-per-pipeline-run PaaS billing, which changes how you plan compute and cost allocation across workspaces.
What should you inventory before migrating?
A clean inventory prevents the most common migration failure: discovering mid-cutover that a connection type or activity has no direct Fabric equivalent.
- Catalogue every asset type. List pipelines, activities, datasets, linked services, triggers, integration runtimes, dataflows, global parameters, SSIS packages, and external dependencies.
- Run the readiness assessment and sort results into Microsoft’s four categories: Ready, Needs review, Coming soon, and Not compatible.
- Act by category. Migrate Ready pipelines immediately using the built-in upgrade. Schedule remediation work for Needs review items before touching them. Treat Not compatible pipelines as candidates for a rebuild rather than a blocked task.
Roughly 90% of ADF activities are already available in Fabric, so most inventories will lean heavily toward Ready and Needs review, with a smaller tail requiring genuine reconstruction.
How do linked services and integration runtimes map to Fabric?
Connection mapping is where migrations stall, so treat it as a distinct workstream rather than a side effect of running the upgrader.
- Create a Fabric connection for each linked service type you use, then consolidate duplicates so pipelines share connections instead of each defining their own.
- Map Self-Hosted Integration Runtimes to the On-premises Data Gateway, and Azure VNet Integration Runtimes to the Virtual Network Data Gateway.
- Validate DNS resolution and firewall rules for each gateway before you rely on it for a production run.
- Build a
resolutions.jsonfile that maps old linked service IDs to new Fabric connection IDs. This is the file the PowerShell upgrader needs and cannot generate for you. - Adopt managed identities wherever the connector supports them, which simplifies both authentication and auditing later.
Pro Tip: Build your resolutions file before you run any bulk upgrade. Missing or mismatched connection mappings are one of the most common reasons a scripted migration aborts partway through.
For teams juggling many connector types, DataTaal’s guide to Fabric data connectors is a useful reference for authentication patterns during this mapping work.
Which transformation approach replaces mapping data flows?
Mapping Data Flows don’t have a single successor in Fabric. The right target depends on what the logic actually does.
- Use Dataflow Gen2 for low-code, rowset-level transforms that non-engineers maintain, especially where the original Mapping Data Flow was already visual and simple.
- Prefer Warehouse SQL for set-based ELT logic, since it gives better lineage tracking and performs well at scale for batch transforms.
- Port genuinely complex logic into Spark notebooks or dbt-style SQL when you need repeatable, version-controlled transformations with fine-grained performance control.
- Where governance allows, use OneLake shortcuts and Mirroring instead of physical copies, cutting duplication between Lakehouse and Warehouse layers, a capability the Fabric Data Factory overview highlights as a core integration benefit.
Document every Mapping Data Flow that needs rework rather than a straight port, and decide upfront whether you refactor during migration or afterwards as a follow-up project.
Testing and cutover: what’s the safe sequence?
Cutover mistakes usually come from skipping validation steps under deadline pressure, not from any single tool failing.
- Validate every Fabric connection inside a non-production workspace, authenticating each connector individually rather than assuming inherited credentials work.
- Keep triggers disabled on migrated pipelines until end-to-end smoke tests and data sampling checks both pass.
- Re-enable schedules and monitor job runs closely for the first 48 to 72 hours, with alerting active from the first run.
- Confirm a rollback plan and decommission timetable, and communicate the cutover schedule to stakeholders before you flip the switch.
Pro Tip: Don’t re-enable all triggers simultaneously. Stagger them by business criticality so a problem in one pipeline doesn’t cascade into an incident across the whole estate.
What separates a smooth migration from a costly rebuild?
Migration succeeds or fails on preparation, not on which tool you pick.
Treat the move as modernisation rather than a lift-and-shift. Automate whatever maps cleanly, and use the rest as a deliberate opportunity to adopt Fabric-native capabilities like variable libraries, unified monitoring, and Copilot-assisted development, a framing Microsoft’s own migration blog explicitly encourages. Standardise naming conventions, parameterisation patterns, and variable library structure before you migrate a single pipeline. Retrofitting naming standards after 200 pipelines are live costs far more than doing it first.
Remember that mounting is discovery only. Nothing moves or changes in your source ADF factory until you explicitly run an upgrade.
The single biggest cause of a stalled migration isn’t unsupported activities. It’s under-preparing connection resolutions and expression syntax changes, then discovering both mid-cutover when a pipeline that looked “Ready” fails its first production run.
How is Fabric Data Factory architecture put together?
Fabric Data Factory sits inside a single SaaS platform alongside Lakehouse, Warehouse, and Power BI, rather than existing as a standalone Azure resource you provision separately. Every pipeline, dataflow, and connection lives inside a Fabric workspace, which becomes your unit of governance and capacity allocation.
The core components are pipelines (orchestration), Dataflow Gen2 (low-code transformation), Copy jobs (simplified data movement), and connections (the replacement for linked services). OneLake underpins all of it as the unified storage layer, meaning a Lakehouse table and a Warehouse table can both be referenced without duplicating the underlying files.
Architecturally, this collapses several ADF concepts into fewer, more integrated pieces. There’s no separate integration runtime resource to provision and monitor independently. Gateway connectivity for on-premises sources now routes through the On-premises Data Gateway or Virtual Network Data Gateway, both managed at the tenant or workspace level rather than per-factory.
For architects mapping this out, DataTaal’s breakdown of OneLake’s role in Fabric data architecture covers how shortcuts eliminate the duplication that used to be routine in multi-layered ADF and Synapse setups. The practical effect: fewer moving parts to monitor, but a bigger blast radius if a shared connection or capacity allocation is misconfigured, which is exactly why the inventory and connection-mapping steps earlier in this guide matter more in Fabric than they ever did in ADF.
How do you optimise performance in Fabric Data Factory?
Performance tuning in Fabric starts with capacity awareness, since pipelines, dataflows, and Power BI reports now compete for the same shared compute pool rather than each having dedicated resources.
Monitor capacity consumption at the workspace level and stagger heavy pipeline runs away from peak Power BI refresh windows. A Dataflow Gen2 refresh and a large Copy job scheduled simultaneously can both slow down because they’re drawing from the same capacity unit, something that never happened when ADF and Power BI Premium had separate billing and compute boundaries.
For Copy jobs and pipeline Copy activities, prefer OneLake shortcuts over physical data movement wherever the source supports it. A shortcut avoids the I/O cost of copying entirely, which is the single biggest lever for reducing both runtime and capacity consumption on data movement-heavy pipelines.
For Dataflow Gen2, push filtering and aggregation as early in the transformation as possible rather than loading full tables and filtering downstream. This matters more in Fabric than it did in Power Query Online because compute cost is now tied directly to your capacity SKU.
For notebook-based transforms, right-size the Spark pool rather than defaulting to the largest available. Oversized pools burn capacity units without a proportional performance gain on smaller datasets. Track job duration trends over time using Fabric’s monitoring hub, and treat any pipeline whose runtime creeps upward month over month as a candidate for a design review, not just a bigger compute allocation.
What does migration change about cost and pricing?
Azure Data Factory bills on a pay-per-use PaaS model: you pay for pipeline orchestration, data movement, and Data Flow execution separately, each metered individually. Fabric Data Factory runs on a shared capacity model, where a single Fabric SKU covers Data Factory alongside Lakehouse, Warehouse, and Power BI workloads in the same workspace.
This shift changes how you plan spend. Instead of a per-pipeline cost you can attribute cleanly, Fabric capacity is a shared pool that every workload draws from, which means a spike in Power BI report usage can affect the capacity available to your data pipelines the same afternoon. Capacity planning becomes a cross-team conversation rather than a data engineering line item.
The upside is consolidation. Where an organisation previously paid separately for ADF, Synapse compute, and Power BI Premium capacity, Fabric folds all of that into one SKU tier, which can simplify procurement even where the raw compute cost is comparable. Whether that nets out cheaper depends heavily on your existing footprint and how efficiently you use the shared capacity, so it’s worth modelling before committing, not after. DataTaal’s guide to Fabric’s cost model walks through how to estimate capacity needs against a typical ADF workload before you commit to a SKU tier.
Budget for a transition period where both ADF and Fabric capacity run in parallel during validation. That overlap is temporary but real, and skipping it to save cost is usually where rollback plans get tested for real.
How do you secure a Fabric Data Factory environment?
Security in Fabric shifts from resource-level access control to workspace-level governance, which means your permission model needs rethinking rather than a straight copy from ADF’s role assignments.
Fabric connections centralise credentials at the workspace level, so managed identity adoption matters more here than it did in ADF, where credentials were often scattered across individual linked services. Consolidating onto managed identities reduces the number of secrets an auditor needs to review and removes the risk of a stale connection string surviving in an old linked service nobody remembers.
Workspace roles (Admin, Member, Contributor, Viewer) govern who can create, edit, or run pipelines, and these roles apply across every item in the workspace, not just Data Factory artefacts. That means a workspace granting pipeline edit rights also grants adjacent rights to Lakehouse and Warehouse objects in the same workspace, so workspace boundaries become your primary security control, not item-level permissions.
Sensitivity labels and OneLake’s centralised data access model extend your existing Microsoft Purview governance into pipeline outputs automatically, provided your organisation has already configured Purview integration.
Gateway security deserves specific attention during migration. On-premises Data Gateway and Virtual Network Data Gateway configurations both need firewall rules and DNS resolution validated in the target environment, not assumed to carry over from the ADF-side Integration Runtime setup, since the network path is genuinely different even when the destination system is the same.

What migration issues come up most often?
Most post-migration incidents trace back to one of a handful of recurring causes, and recognising the pattern quickly saves hours of investigation.
Connection authentication failures top the list, usually because a resolutions file mapped an old linked service to the wrong Fabric connection ID, or a managed identity wasn’t granted access to the target resource before the first run.
Silent trigger gaps happen when a team assumes the upgrade re-enables schedules automatically. It doesn’t. Triggers stay disabled by default, and forgetting to re-enable them after validation means pipelines simply don’t run, often for days before anyone notices.
Expression syntax mismatches appear in pipelines that used ADF-specific dynamic content expressions that don’t translate cleanly. These typically surface as an activity failing on its first live run despite passing validation, because the parameter resolves differently at runtime.
Capacity throttling shows up as pipelines that ran fine in isolation but slow down or fail once real production load from Power BI and other workloads shares the same capacity SKU.
Unresolved global parameters cause failures when a global parameter that should have converted to a variable library entry gets missed, leaving an activity referencing a value that no longer exists in the new environment.
For each of these, the fix is almost always the same: re-run the non-production validation step you may have skipped, checking connections, triggers, and parameter resolution individually rather than assuming a passing pipeline test covers all three.
How do you monitor and govern Fabric Data Factory at scale?
Fabric’s monitoring hub gives you a single view across pipelines, dataflows, and notebooks, replacing the separate Azure Monitor and ADF monitoring pane you’d have used previously. Run history, duration trends, and failure alerts all live in one place per workspace.
Governance in Fabric leans on workspace structure more than item-level tagging. Because permissions and capacity both scope to the workspace, your governance model should define workspace boundaries around business domains or environments (development, test, production) rather than around individual pipeline ownership, which was more common in ADF’s flatter resource hierarchy.
Deployment pipelines in Fabric support promoting Data Factory items between environments in a structured sequence, which pairs naturally with the CI/CD flows teams already use for the PowerShell upgrade path. Combine this with variable libraries so environment-specific values (connection strings, folder paths, thresholds) change automatically on promotion rather than requiring manual edits per environment.
Set alerting on pipeline failures and unusual duration spikes from day one, not after the first incident. Fabric’s native alerting integrates with the same monitoring hub, so there’s no separate service to configure. Combined with Purview sensitivity labels flowing through OneLake, this gives most organisations tighter oversight than a comparable ADF and Synapse setup ever achieved, largely because everything reports through one surface rather than three.
When should you bring in outside help?
Consolidating pipelines into Fabric pays off over time: less infrastructure to babysit, one monitoring surface instead of three, and access to features like Copilot and unified governance that ADF never had. Most teams can handle Ready and Needs review pipelines themselves.
Bring in a specialist when gateway or network topology gets complicated, when you’re automating migration through enterprise CI/CD, or when early performance numbers under real capacity load don’t match expectations. A short readiness review from someone who’s mapped connections before often catches the resolutions file errors that stall self-managed migrations.
— Michael
Get migration help from DataTaal
DataTaal is the alternative to guessing your way through connection resolutions and rebuilding pipelines twice. Instead of a generic agency retainer, you work directly with a certified Fabric specialist for the whole engagement, no junior staff, no management layers passing your project between teams.

A typical engagement starts with an assessment to catalogue your ADF estate against Fabric’s readiness categories, so you know exactly what’s Ready, what needs review, and what genuinely warrants a rebuild before you touch a single trigger. From there, DataTaal handles connection and resolutions mapping, sets up PowerShell or CI/CD automation for the pipelines that qualify, and runs test cycles alongside your team rather than handing back a black box. For estates with heavy customisation or legacy SSIS dependencies, phased migration engagements reduce risk by moving in controlled batches instead of forcing a single high-stakes cutover weekend.
If your organisation also needs the transformed data feeding into dashboards, Power BI implementation and consulting can be scoped alongside the migration itself. Start with an assessment conversation through consulting services, priced from $20 per hour, and get a concrete readiness picture before you commit to a migration timetable.
FAQ
What is Data Factory in Fabric?
Data Factory in Fabric is Microsoft’s SaaS-native orchestration and transformation service inside the Fabric platform, combining pipelines, Dataflow Gen2, and Copy jobs with direct OneLake integration. Unlike Azure Data Factory, it defines connection details inline on activities rather than through separate dataset and linked service objects, as the official overview explains.
Is Fabric an ETL tool?
Fabric Data Factory supports full ETL and ELT workflows through pipelines, Dataflow Gen2, and Warehouse SQL, but Fabric itself is broader than an ETL tool. It’s an integrated analytics platform where Data Factory handles ingestion and transformation alongside Lakehouse, Warehouse, and Power BI in the same workspace.
Why use Fabric instead of Azure Data Factory?
Fabric consolidates orchestration, storage, and reporting into one SaaS platform with a shared capacity model, reducing the number of separate services you provision and monitor. It also unlocks OneLake shortcuts, Mirroring, and Copilot-assisted development that don’t exist in standalone Azure Data Factory, though the right choice depends on your existing Azure footprint and whether DataTaal’s migration assessment shows strong parity for your current pipelines.
Is Fabric a data warehouse?
Fabric includes a Warehouse item as one of several storage and compute options, alongside Lakehouse, but Fabric as a whole is a broader analytics platform rather than a single data warehouse product. Data Factory pipelines can write to either the Warehouse or Lakehouse, depending on whether you need SQL-based analytics or open file-format storage.
Do triggers migrate automatically with my pipelines?
No. Triggers are disabled by default on migrated pipelines regardless of which migration path you use, so scheduled runs won’t start until you explicitly re-enable them after validation. This is intentional, giving you a safe window to run end-to-end tests before production traffic resumes.