TL;DR
Data Factory in Microsoft Fabric is best when your data ends up in Fabric, but SSIS users should keep Azure Data Factory for now. Use Copy job to load data, and use Mirroring to keep a live copy of a database. Clean and shape data with Dataflow Gen2, or with notebooks if your team writes code. Pipelines run these steps on a schedule and send an alert when one fails. Costs come out of your Fabric capacity. Old Azure Data Factory pipelines can move over with a built-in upgrade tool.
Key Takeaways Data Factory in Microsoft Fabric is the data integration workload inside Fabric, and Microsoft calls it the next generation of Azure Data Factory. Copy job is the default way to bring data in, while Mirroring keeps a near real-time replica of supported databases. Dataflow Gen2 suits low-code transformations, whereas notebooks and dbt jobs suit code-first teams. Pipelines are for orchestration: they schedule steps, branch on failure and send alerts. Cost is measured in Fabric capacity units (CUs), and copy time usually drives most of it. Keep Azure Data Factory for now where you depend on the Azure-SSIS runtime or managed virtual networks. Watch on YouTube
Azure to Fabric Migration Made Simple with Kanerika
A Kanerika video on moving Azure data workloads to Microsoft Fabric, the same move the team in the scenario below is weighing.
When One Integration Team Runs Three Tools Picture a five-person data team at a mid-sized distributor. Its Azure Data Factory pipelines copy ERP and CRM tables every night, while analysts refresh their own Power Query steps inside Power BI. A Synapse Link feed also replicates the order database into a separate workspace that only two engineers understand.
Each tool works fine on its own. The trouble starts when a report is wrong, because then nobody can say which step moved, changed or broke the data.
Moving that team to Microsoft Fabric comes down to three decisions. How does data enter Fabric, where is it transformed, and what runs on a schedule?
The Data Factory workload in Fabric is the part of the platform that answers all three. This guide walks through each choice, with the trade-offs, the costs and a first build you can copy.
What Is Data Factory in Microsoft Fabric? Data Factory in Microsoft Fabric is the integration workload of Fabric, where you connect sources, move data, reshape it and schedule the work. Microsoft’s own Data Factory overview describes it as the next generation of Azure Data Factory, rebuilt as a software-as-a-service experience.
In plain terms, it is the plumbing between your business systems and OneLake, the single data lake that every Fabric workload shares. A sales order in SQL Server, a customer record in Salesforce and a spreadsheet in SharePoint can all reach the same Lakehouse tables through it. Our overview of Microsoft Fabric architecture explains how the other workloads read from that shared lake.
Two ideas, however, often get blurred here. Getting data in is ingestion, whereas making data from different systems agree is integration. Our guide to ingestion versus integration covers that difference in depth. Fabric Data Factory, in short, does both jobs.
The Four Jobs It Does Every task you hand to Data Factory falls into one of four jobs, and each job has its own tools.
Connect. Connections and a broad connector library reach databases, files, SaaS apps and cloud stores, with a gateway for systems behind your firewall.Move. Copy job, the Copy activity and Mirroring bring data into OneLake in bulk, in increments or as a continuous replica.Transform. Dataflow Gen2, notebooks and dbt jobs clean, join and model the data once it has landed.Orchestrate. Pipelines and Apache Airflow jobs decide what runs, in which order, and what happens when a step fails.Copilot also sits across all four jobs. Microsoft’s Copilot for Data Factory documentation says it can generate pipelines from a description, explain errors and summarize what a pipeline does. It also helps write expressions and adds Dataflow Gen2 steps from plain-language requests. A new team member therefore gets a quick first draft to review.
Where It Sits in Fabric Think of Fabric as a line that runs from left to right. Sources sit on the left, Data Factory moves and shapes their data in the middle, and OneLake stores the result in a Lakehouse or Warehouse. Power BI and the other Fabric workloads then read it on the right.
One route, however, skips Data Factory entirely. A OneLake shortcut points at data that already sits in Azure Data Lake Storage, Amazon S3 or another lake. Fabric then queries it in place, without a copy.
Use a shortcut when the data is already in an open format and clean enough to read. Use Data Factory when it has to be pulled, converted or merged first. The difference between a data lake and a lakehouse matters here too, because the Lakehouse is where most Data Factory loads land.
The Building Blocks and When to Use Each Fabric Data Factory has several tools for the same job, and that overlap is where new teams lose the most time. There are three ways to move data and three ways to transform it. Usually, though, only one option is the cheapest to run and the easiest to support.
The rule of thumb is short. First, move data with the lightest tool that meets the freshness need. Then transform it with a tool your team can read and test, and save pipelines for orchestration.
Copy Job, Copy Activity or Mirroring? Start with Copy job. The Data Factory overview names it the preferred way to move data. It handles bulk copies, incremental loads and change data capture (CDC) replication from one item, and it has its own schedule.
The Copy activity is the older, more granular option, and it lives inside a pipeline. Reach for it when the copy is one step in a larger workflow with custom retries and parameters. For example, a file must arrive, get copied and then trigger a notebook.
Mirroring works differently. Instead of scheduled copies, it keeps a near real-time replica of a source database in OneLake. According to the Mirroring overview , supported sources include Azure SQL Database, SQL Server, Azure Database for PostgreSQL, Cosmos DB, Snowflake, Oracle, BigQuery and SAP Datasphere.
Mirroring is cheap to try because Microsoft says the background compute that replicates your data does not consume capacity. Each CU you buy also includes 1 TB of free mirroring storage, so an F64 capacity comes with 64 TB.
The table puts the four ways of getting data into OneLake side by side.
What you need Use Why it wins A daily or hourly load of tables or files, full or incremental Copy job Built for bulk, incremental and CDC loads with the least setup A copy that is one step in a multi-step workflow Copy activity in a pipeline Gives control over parameters, retries and dependencies A near real-time replica of a supported operational database Mirroring Continuous replication with no copy schedule to manage Data already in ADLS, S3 or another lake in an open format OneLake shortcut No copy at all, because Fabric reads the data in place
Why Incremental Loads Pay Off Incremental loading is where this choice pays off. For example, picture an orders table with 50 million rows. A full copy moves every old row again each night. An incremental Copy job, by contrast, reads only rows whose timestamp or change log moved since the last run.
The cost section below shows the trade. An incremental Copy job is billed at 3 CU hours per hour of throughput optimization, double the 1.5 CU hours of a full copy. It pays off whenever the delta run takes less than half as long as a full reload.
Assume copy time grows roughly with the rows moved. If about 1% of those 50 million rows change each day, the nightly delta is around 500,000 rows. That small delta clears the bar easily, and it is why Copy job is the default and the other three options are exceptions you choose for a reason.
Dataflow Gen2, Notebooks or dbt Job? Once data has landed, the next question is who will maintain the transformation. The best tool is the one your team can read, test and fix at 2 a.m., even if another tool has more features. Teams that need reusable patterns can also follow our guide to building an ETL framework .
Dataflow Gen2 is the low-code choice. It uses the same Power Query editor that many analysts already know from Power BI and Excel, and Microsoft counts over 300 transformations in it. It suits cleaning, merging and reshaping work of modest size that analysts or BI developers own.
Notebooks suit engineers. They run Spark with Python, SQL or Scala. As a result, they suit heavy joins, complex logic and unit-tested code that would be awkward in a visual editor.
Finally, the dbt job is the SQL-first option for teams that already model with dbt . According to the dbt job documentation , it is in preview and switched on through a tenant setting. It runs dbt projects against a Fabric Warehouse or Lakehouse, plus Azure SQL, PostgreSQL and Snowflake. Our comparison of dbt, Informatica and Azure Data Factory shows where dbt fits against classic ETL tools.
Question Dataflow Gen2 Notebook dbt job (preview) Who owns it Analysts and BI developers Data engineers Analytics engineers who use dbt How you build it Power Query, mostly visual Python, Spark SQL or Scala SQL models with dbt tests Typical job Cleaning, merging, reshaping Heavy joins, complex logic, ML preparation Layered SQL models in a Warehouse or Lakehouse Watch out for Run time and cost on very large volumes Needs coding skills and code review Preview status, no build caching, 5,000-file project import limit
When You Actually Need a Pipeline A pipeline is an orchestration tool first and a data mover second. If the idea is new to you, our explainer on how data pipelines work covers the basics. This section looks only at when a Fabric pipeline earns its place.
You need one when steps depend on each other. A typical case: wait for the Copy job to finish, run the Dataflow, refresh the semantic model, and email the team if anything fails. Pipelines give you control flow such as If Condition, ForEach and Until, plus schedules, event triggers and an On-fail path for every activity.
A single Copy job on a schedule does not need one, because Copy job already schedules itself. Teams that write orchestration in Python can also use an Apache Airflow job instead, which runs Airflow DAGs inside Fabric. Our guide to data orchestration tools compares Airflow with the other common options.
Kanerika Service
Data Integration Services
Kanerika connects ERPs, CRMs and SaaS apps to Microsoft Fabric with the right mix of Copy jobs, Mirroring and pipelines, so every source lands in one governed place.
Explore Data Integration How to Build Your First Fabric Data Factory Flow The quickest way to learn the tool is to build one small flow from end to end. The example below loads an orders table from an on-premises SQL Server into a Lakehouse, cleans it, and alerts the team when something fails.
Create a workspace on Fabric capacity. Data Factory needs a Fabric tenant and a Fabric-enabled workspace, and a free trial works for learning. Add a Lakehouse to the workspace as the destination.Set up the connection. For an on-premises SQL Server, install the on-premises data gateway on a machine that can reach the database, then create the connection through it. Cloud sources, by contrast, connect directly.Create a Copy job. Pick the orders table, choose the Lakehouse as the destination, and select incremental copy with a column that changes on every update, such as a last-modified timestamp. The first run loads everything, and later runs move only the changes.Transform the data. Build a Dataflow Gen2 that removes test orders and adds customer details, or use a notebook if the logic is heavy. Then write the result to a clean table in the same Lakehouse.Wrap it in a pipeline. Add a Copy job activity and the transform step, join them with On-success arrows, and put an Outlook or Teams activity on the On-fail path. Then schedule the pipeline rather than the individual items.Check the monitoring hub. After the first scheduled run, open the monitoring hub and confirm the status, duration and row counts for each activity before anyone builds a report on the table.Checklist
Data Integration Checklist
Use Kanerika’s Data Integration Checklist to plan sources, load patterns, data quality checks and monitoring before your first Fabric flow goes live.
Get the Checklist → Two Expressions Worth Knowing Pipelines use the same expression language as Azure Data Factory, so existing skills carry over. Two expressions cover most first flows: a dated folder path for raw files, and the error text for the failure alert.
// Folder path for today's raw files, for example raw/orders/2026/10/25
@concat('raw/orders/', formatDateTime(utcNow(), 'yyyy/MM/dd'))
// Message body for the On-fail email or Teams post
@concat('Copy orders failed: ', activity('Copy orders').error.message)Give every item a readable name from day one, such as pl_sales_orders_daily. The monitoring hub lists runs by item name, so a vague name slows down every investigation.
Data Factory in Microsoft Fabric vs Azure Data Factory Teams that already run Azure Data Factory usually ask whether the Fabric version is the same tool in a new place. The authoring canvas will look familiar, but the operating model underneath is different, and that model is what changes daily work.
The biggest shift is from platform-as-a-service to software-as-a-service. In ADF you provision and manage integration runtimes and publish changes before they run. In Fabric there is no runtime to manage and no publish step, so you simply save and run. All the work then draws on the workspace’s Fabric capacity.
Microsoft’s comparison of Fabric Data Factory and Azure Data Factory lists every difference. The table below groups only the ones that change how you operate, because those matter most.
Area Azure Data Factory Data Factory in Fabric Service model PaaS resource in an Azure subscription SaaS workload inside a Fabric workspace Compute Integration runtimes you size and manage No integration runtime to manage On-premises access Self-hosted integration runtime On-premises data gateway Publishing Publish before changes take effect Save and run, with no publish step Change data capture CDC resource Copy job with CDC Near real-time replicas Synapse Link Mirroring Calling other pipelines Execute Pipeline activity Invoke pipeline, which can also call ADF and Synapse pipelines Copying a pipeline definition ARM template export Save as Scheduling Central triggers shared across pipelines A schedule set on each pipeline CI/CD Git plus ARM-based releases Git integration and deployment pipelines Monitoring A monitor per factory Monitoring hub across workspaces Identity Managed identity Workspace identity SSIS Azure-SSIS integration runtime To be determined, with an Invoke SSIS Package activity in preview Network isolation Managed virtual network and private endpoints To be determined, with a virtual network data gateway available
What Changes Day to Day In practice the SaaS model removes chores such as sizing runtimes and remembering to publish. The on-premises gateway, however, still needs a patched and monitored machine, just as the self-hosted runtime did.
The other change is in how teams think about cost and access. Every pipeline now shares capacity with reports, notebooks and warehouses. A slow load is therefore no longer only the data team’s problem. Access, too, follows Fabric workspace roles rather than Azure resource permissions.
When to Keep Azure Data Factory for Now ADF remains a supported Azure service, while new capabilities such as Copy job and Mirroring are landing in Fabric. That makes the move a question of timing rather than urgency. Keep ADF, or run both side by side, when any of these apply:
You run SSIS packages on the Azure-SSIS integration runtime, which Fabric still lists as to be determined. Security rules require a managed virtual network or private endpoints for every data movement. You rely on central triggers shared by many pipelines, since Fabric sets schedules per pipeline. Pipelines read many secrets from Azure Key Vault, where Fabric support is narrower and works through Key Vault references. A mixed estate is also a sound interim design. Fabric’s Invoke pipeline activity can call ADF pipelines. One Fabric pipeline can therefore orchestrate the parts that stay in Azure until they are ready to move.
Datasheet
Automating Azure to Fabric Migration With FLIP
See how Kanerika’s FLIP accelerator converts Azure Data Factory pipelines, datasets and linked services into Fabric pipelines while keeping their logic and dependencies.
View the Datasheet → What Data Factory Costs: Capacity Units, Not Per-Run Bills Fabric does not send you a bill per pipeline run. Instead, you buy a Fabric capacity, an F SKU whose number equals its capacity units, so an F64 provides 64 CUs. Every Data Factory job then consumes some of those units while it runs.
Our guide to Microsoft Fabric pricing covers the capacity side, so this section explains only what Data Factory draws from it. Each tool reports to its own meters, as Microsoft’s Data Factory pricing overview sets out. Pipelines use Data Orchestration and Data Movement, Dataflow Gen2 uses Standard and High Scale Compute, and Copy job uses Data Movement and Data Movement Incremental. OneLake storage, however, is billed separately.
What runs Published rate Copy activity, or a Copy job doing a full copy 1.5 CU hours per hour of intelligent throughput optimization Copy job in incremental mode, including its first full load 3 CU hours per hour of intelligent throughput optimization Any other pipeline activity, such as Lookup or a Teams alert 0.0056 CU hours per activity run SSIS package run from a pipeline 1.35 CU hours per vCore Dataflow Gen2 standard compute (CI/CD dataflows) 12 CU seconds per second of query time for the first 10 minutes, then 1.5 (non-CI/CD: 16 throughout) Dataflow Gen2 High Scale Compute 6 CU seconds per second of staged query time
Rates come from the pipeline pricing , Copy job pricing and Dataflow Gen2 pricing pages. Mapping Data Flow transforms in Dataflow Gen2 are not billed while they are in preview.
A Worked Example in CU Hours Take a nightly pipeline with one Copy activity that uses 4 units of throughput optimization for 15 minutes. It also runs 10 other activities, such as lookups, a stored procedure and a Teams alert.
The copy uses 4 x 0.25 hours x 1.5 = 1.5 CU hours. The other activities add 10 x 0.0056 = 0.056 CU hours, so one run costs about 1.56 CU hours. Over 30 nightly runs, that comes to about 46.7 CU hours a month, and the copy accounts for roughly 96% of it.
Convert that total at your region’s CU-hour rate on the Microsoft Fabric pricing page , then watch real use in the Capacity Metrics app after go-live. The lesson is plain: shorter, incremental copies cut cost far more than trimming pipeline steps does.
Capacity is also shared, so a heavy dataflow can slow a report that refreshes at the same hour. Schedule large loads in quiet hours, and size the capacity for the busiest window rather than the average. For license strategy across Fabric and Power BI, see our guide to Microsoft licensing optimization .
Migrating ADF and Synapse Pipelines to Fabric A move from ADF or Synapse to Fabric is mostly an inventory problem rather than a rewrite problem. The pipelines that convert cleanly take little effort, so the real work is finding the few that depend on features Fabric handles differently.
Microsoft’s upgrade planning guide describes three paths. First, you can mount an existing data factory in a Fabric workspace and keep running it as it is. You can also use the built-in upgrade experience or rebuild pipelines by hand. Many estates use all three.
Assess and inventory. List every pipeline, linked service, trigger, integration runtime and global parameter, along with its owner and schedule.Mount the factory. Add the Azure Data Factory item to a Fabric workspace, so teams can see and run existing pipelines from Fabric while the move proceeds.Run the built-in upgrade. The assessment sorts each pipeline into Ready, Needs review, Coming soon or Not compatible, which shows where the effort sits before anything is converted. The same upgrade, in preview, converts eligible Mapping Data Flows into Mapping Data Flow transforms in Dataflow Gen2.Rebuild what cannot upgrade. Recreate the Not compatible items by hand. Unsupported Mapping Data Flows, such as those that use flowlets or user-defined functions, get redesigned in Power Query, Warehouse SQL or a notebook.Run in parallel, then cut over. Run old and new pipelines side by side, compare row counts and totals, then switch the schedules and retire the ADF triggers.Gaps That Need a Design Decision Some differences cannot be converted by any tool, because they change how a pipeline is designed rather than where it runs. Global parameters become variable libraries in Fabric, with different patterns and data types. Dynamic linked-service parameters move as well: Fabric connections do not take dynamic properties, so metadata-driven pipelines parameterize the connection object instead.
Record each decision next to its pipeline in the inventory, so the cutover team knows what is settled. Funding can offset part of the effort, and our guide on how to reduce Fabric migration costs with Microsoft funding explains the programs.
SSIS needs its own plan. The Invoke SSIS Package activity is still in preview, according to the activity documentation . SSIS-heavy estates therefore often keep packages in ADF while new loads are built in Fabric.
Case Study
Azure Data Factory to Microsoft Fabric Migration
A global packaging-solutions leader moved scattered ADF and Synapse workflows to Fabric with Kanerika’s migration utility, cutting cloud and data costs by 30% and delivering business insights 80% faster.
Read the Case Study → Running It in Production: Governance, Monitoring and CI/CD A flow that works on one developer’s screen is not yet a production system. Production means changes are tested before they go live, failures reach a person quickly, and anyone can see what ran and when.
First, get change control in place. Connect each workspace to Git and promote changes through deployment pipelines from development to test to production. Values that differ by stage, such as server names and folder paths, also belong in a variable library . That workspace item holds per-stage values for pipelines and notebooks.
For monitoring, the monitoring hub shows every run across workspaces, and Microsoft’s SLA promises that 99.9% of pipeline requests succeed. Under the same SLA, activity runs start within four minutes of schedule 99.9% of the time.
That service level covers the platform, not your data. A run can succeed and still load half the rows. So add your own checks on freshness and row counts, as our guide to data observability describes. Before go-live, confirm each point below:
Every pipeline has an On-fail path that alerts a named owner. Connections use workspace identity or a service principal instead of a personal account. Pipelines, dataflows and Lakehouse tables follow one naming convention. Every workspace is linked to Git, and production changes arrive only through deployment pipelines. Someone reviews capacity use in the Capacity Metrics app every week. Rolling this out across many teams is as much a people task as a technical one. Lineage and cataloging also sit with your data governance tools , so connect Fabric to them early. Our guide to Microsoft Fabric adoption covers training, ownership and the order in which to move workloads.
Checklist
Microsoft Fabric Checklist
Download Kanerika’s Microsoft Fabric Checklist to cover capacity, workspaces, governance, CI/CD and monitoring before you put Data Factory into production.
Get the Checklist → Where Teams Use Fabric Data Factory The same tool looks quite different depending on the problem it solves. The four scenarios below show which part of Data Factory each one relies on, and whether Fabric is the right home for it. For the architecture behind such programs, see our guide to enterprise data integration .
ERP and CRM Data in One Lakehouse A manufacturer wants SAP orders and Salesforce opportunities in one sales model. Copy job loads both incrementally into a Lakehouse, a notebook joins them, and a pipeline refreshes the Power BI model at 6 a.m. Fabric fits well here because the data ends up in Power BI anyway.
Near Real-Time Operations on a Mirrored Database A logistics team needs shipment status within minutes, not the next morning. Mirroring replicates the Azure SQL operational database into OneLake, so reports query the replica without touching production. No pipeline is needed to move the data at all, and our roundup of data replication tools covers the options outside Fabric.
Consolidating an Old ADF Estate A company runs 200 ADF pipelines across five factories, with schedules nobody has reviewed in years. Mounting the factories in Fabric gives one view, and the upgrade assessment sorts the work into waves.
Fabric pipelines then take over one wave at a time. The SSIS-heavy factory stays in ADF until Fabric’s SSIS support leaves preview.
Analyst Self-Service With Dataflow Gen2 A finance team rebuilds the same spreadsheet cleanup every month. Dataflow Gen2 turns those Power Query steps into a shared, scheduled item that writes to a Lakehouse table. Engineers review it once, and analysts own it after that.
Each scenario points to the same test. Fabric Data Factory fits best when the data lands in OneLake and Fabric or Power BI users consume it. For platforms outside Microsoft, see our roundup of data integration tools . For where integration ends and ETL begins, see integration approaches compared with ETL .
Common Mistakes That Raise Cost or Break Pipelines The same few design habits cause most Fabric Data Factory problems, and they usually arrive with teams moving from older tools. Each one has a simple fix.
Full reloads where incremental would do. Copying a whole table every night burns capacity on rows that never changed. Use incremental Copy job or Mirroring instead.A pipeline around everything. Wrapping a single scheduled copy in a pipeline adds parts to maintain without adding control. Let Copy job run on its own schedule.Heavy transforms in the wrong tool. Very large joins in Dataflow Gen2 can run long and use a lot of capacity. Move them to a notebook or the Warehouse.No failure path. A pipeline without an On-fail alert fails silently until a report looks wrong. Add an alert to every pipeline on day one.Hard-coded connections. Server names typed into each pipeline break the moment you promote to test. Keep them in a variable library.Big loads at report time. Large jobs that run while users refresh reports compete for the same capacity. Move them to quiet hours.The first three mistakes raise the bill, while the last three break pipelines or slow them down. Fix the cost mistakes before a capacity upgrade looks like the only answer.
How Kanerika Implements Data Factory in Microsoft Fabric Kanerika is a Microsoft Solutions Partner for Data and AI , and we run Fabric Data Factory projects from the first assessment through to production support.
Every engagement starts with an assessment of the ADF and Synapse estate against the built-in upgrade categories. The plan then shows which pipelines upgrade as they are, which need review and which must be rebuilt. Our migration utility converts the pipelines, and we set governance conventions for naming, version control and documentation before anything goes live. Old and new pipelines run in parallel until the numbers match.
One example is a global packaging-solutions leader with scattered ADF and Synapse workflows. An intermediate Parquet conversion step was causing latency and failures.
Kanerika moved the workflows to Fabric with its migration utility, connected SAP C4C directly to Fabric, and put one governance framework in place. The ADF to Fabric case study reports a 30% reduction in cloud and data costs, a 50% improvement in the data pipeline and 80% faster business insights and reporting.
To plan a similar move, explore our Azure to Fabric migration services or our wider Microsoft Fabric services .
Kanerika Service
Azure to Fabric Migration Services
Kanerika assesses your ADF and Synapse estate, converts pipelines with its migration utility, and runs old and new side by side until the numbers match.
Plan Your Migration Wrapping Up Data Factory in Microsoft Fabric is less a single tool than a set of choices. Copy job brings data in, Mirroring keeps live replicas, and Dataflow Gen2, notebooks and dbt jobs transform it. Pipelines then tie the steps together with schedules and alerts.
Start small with one incremental flow, measure its capacity use, and grow from there. So is Fabric Data Factory the right integration layer for you? If your data is headed for OneLake and Power BI, it is, and Copy job is the place to start. If you depend on SSIS runtimes or managed virtual networks, run it alongside Azure Data Factory until those gaps close.
Frequently Asked Questions
What does Data Factory in Microsoft Fabric actually do? Data Factory in Microsoft Fabric connects to your sources, moves data into OneLake, transforms it and schedules the work. Copy job and Mirroring bring data in, while Dataflow Gen2, notebooks and dbt jobs reshape it. Pipelines and Airflow jobs run the steps in order, retry them and send alerts when something fails.
Does Microsoft Fabric include Data Factory? Yes. Data Factory is one of the built-in workloads of Microsoft Fabric, alongside Data Engineering, Data Warehouse, Real-Time Intelligence and Power BI. There is nothing separate to install. Any Fabric-enabled workspace on a Fabric capacity or trial can create pipelines, Copy jobs and dataflows, and every workload shares the same OneLake storage.
Can Fabric Data Factory connect to on-premises SQL Server? Yes. Install the on-premises data gateway on a machine that can reach the SQL Server, then create a connection through that gateway in Fabric. Copy jobs, pipelines and Dataflow Gen2 items can all use that connection. The gateway replaces the self-hosted integration runtime that Azure Data Factory uses for the same purpose.
How is Data Factory in Microsoft Fabric billed? It draws capacity units from your Fabric capacity instead of billing per run. Copy work uses 1.5 CU hours per throughput hour for full copies and 3 for incremental Copy job. Each other pipeline activity uses 0.0056 CU hours. Dataflow Gen2 is billed by query duration, while OneLake storage is billed separately.
Can I migrate existing Azure Data Factory pipelines to Microsoft Fabric? Yes. You can mount an existing data factory in a Fabric workspace, use the built-in upgrade, or rebuild pipelines by hand. The upgrade assessment labels each pipeline Ready, Needs review, Coming soon or Not compatible. Plan for gaps such as workspace identity, per-pipeline schedules and variable libraries in place of global parameters.
Can Fabric run SSIS packages, and what replaces SSIS? Fabric pipelines offer an Invoke SSIS Package activity, which is in preview. The Azure-SSIS integration runtime is still listed as to be determined in Fabric. For new work, Copy job, Dataflow Gen2 and notebooks replace most SSIS tasks. Many teams keep existing packages running in ADF until they can rebuild them.
Does Microsoft Fabric Data Factory support Apache Airflow? Yes. Fabric includes an Apache Airflow job that runs Airflow DAGs written in Python inside Fabric, so there is no separate Airflow cluster to host. It suits engineering teams that already manage orchestration as code. Teams that prefer a visual canvas can use Fabric pipelines for the same scheduling and dependency work.
How does Copilot work inside Fabric Data Factory? Copilot in Data Factory can generate a pipeline from a plain-language description, explain pipeline errors, summarize what a pipeline does and help write expressions. In Dataflow Gen2 it adds transformation steps from natural-language requests. Treat its output as a first draft, then review and test it before scheduling anything for production.
Can I run dbt inside Fabric Data Factory? Yes, through the dbt job item, which is in preview. An admin first enables the dbt jobs tenant setting. The job then runs your dbt project against a Fabric Data Warehouse, so SQL models and tests execute inside Fabric. Pipelines can schedule it after the Copy job that loads the raw data.
How do you monitor Data Factory pipelines in Microsoft Fabric? Use the monitoring hub, which lists pipeline, Copy job and dataflow runs across workspaces with status, duration and errors. Add an On-fail alert to every pipeline so a named owner hears about failures. Then check freshness and row counts yourself, and review capacity use in the Capacity Metrics app every week.
What is the difference between Azure Data Factory and Fabric Data Factory? Azure Data Factory is a platform service you run in an Azure subscription, with integration runtimes to manage and a publish step. Fabric Data Factory is a software-as-a-service workload inside Fabric. It has no runtime to manage, saves and runs without publishing, uses the on-premises data gateway and bills against shared Fabric capacity.
Is Azure Data Factory being replaced by Microsoft Fabric? Azure Data Factory remains a supported Azure service. Microsoft describes Fabric Data Factory as its next generation, and new capabilities such as Copy job and Mirroring are landing in Fabric. Many teams therefore build new work in Fabric and keep SSIS or network-isolated pipelines in ADF until Fabric covers those needs.
Is Data Factory an ETL tool or an ELT tool? It supports both patterns. You can transform data on the way in with Dataflow Gen2, which is classic ETL. More often, teams land raw data in a Lakehouse first and transform it afterwards with notebooks, dbt jobs or SQL, which is ELT. The right pattern depends on data volume and who maintains the logic.
How do I get to Data Factory in Fabric? Sign in to Microsoft Fabric and open a Fabric-enabled workspace. From the workload switcher or the New item menu, choose Data Factory items such as a pipeline, a Copy job or a Dataflow Gen2. You need a Fabric tenant and capacity, although a free trial is enough to learn the tool.
What is the difference between Copy job and Copy activity in Fabric? Copy job is a standalone item for moving data, with its own schedule and built-in bulk, incremental and change data capture modes. The Copy activity runs inside a pipeline as one step of a larger workflow. Choose Copy job by default, and the Copy activity when the copy needs custom parameters, retries or dependencies.
What is Dataflow Gen2 and when should you use it instead of a pipeline? Dataflow Gen2 is the low-code transformation tool in Fabric, built on Power Query with over 300 transformations. Use it to clean, merge and reshape data. A pipeline does not transform data itself; it orchestrates steps. So use Dataflow Gen2 for the transformation and a pipeline to schedule it alongside other steps.
What is Mirroring in Microsoft Fabric and when should you use it? Mirroring keeps a near real-time replica of a source database in OneLake without scheduled copies. Sources include Azure SQL Database, SQL Server, PostgreSQL, Cosmos DB, Snowflake, Oracle and BigQuery. Use it when reports need data within minutes and you want to query a replica instead of loading the production database.
Does Fabric Data Factory support incremental loads and change data capture? Yes. Copy job supports incremental copy, which reads only rows whose watermark column changed since the last run. It also supports change data capture replication for supported sources, carrying inserts, updates and deletes. The first run in incremental mode loads everything, and later runs move only the changes, which saves capacity.
