Skip to content
Migration · Track 03

Migration toFabric.

SQL Server, SSIS, ADF and Synapse estates move to OneLake as items your team owns: mirrored where a source stays put, shortcut where a copy is not needed, and rebuilt as Fabric pipelines and Warehouse T-SQL where the workload moves. Power BI reads the result over Direct Lake.

01 / The Shift

From Estate to OneLake

Today

An estate running out of road

Hardcoded SSIS packages, ADF and Synapse pipelines, T-SQL nobody wants to touch, a dedicated pool with a retirement date, and every Power BI report refreshed from all of it.

Target

OneLake, as items in Git

One copy of the data in OneLake, Warehouse and Lakehouse items, Fabric pipelines and Direct Lake models beneath the same reports. All of it is in Git, with deployment pipelines.

02 / OneLake

One lake, every workload

01

OneLake

One logical lake for the whole tenant. A table is stored once, in the open Delta Parquet format, and every workload reads that same copy. There are no extracts between Warehouse, Lakehouse and Power BI, and no second bill for the same data.

  • One copy
  • Delta Parquet
  • Every workload
02

Lakehouse

Delta tables and files under one item, with Spark notebooks on top for the heavy transformations. Every table gets a SQL analytics endpoint as soon as it lands, so T-SQL readers are never waiting on anyone.

  • Tables and files
  • Spark notebooks
  • SQL analytics endpoint
03

Warehouse

A full T-SQL warehouse on OneLake, with transactions, stored procedures and views. Dedicated-pool schemas and the procedures written against them arrive largely as they are: the dialect differences are reviewed and the logic is not rewritten.

  • T-SQL
  • Transactions
  • Procedures kept
04

Mirroring

A snapshot of a database, followed by every change, replicated into OneLake in near real time as Delta tables. When the source is on-premises this runs through the on-premises data gateway. SQL Server is mirrored, not extracted on a schedule.

  • SQL Server
  • Azure SQL
  • Near real time
05

Shortcuts

A reference to data that stays where it is and is read by OneLake as if it were local: ADLS Gen2 folders, S3 buckets and Unity Catalog tables. There is one copy of the files, with OneLake’s security on top and no second pipeline.

  • ADLS Gen2
  • Unity Catalog
  • Zero copy
03 / On Fabric

The Workloads on Top

06

Fabric pipelines

Data Factory in Fabric: activities, loops, triggers and the copy activity, built and scheduled in the same workspace as the data. ADF pipelines, Synapse pipelines and SSIS packages arrive here as pipelines your team can open and change.

  • Data Factory in Fabric
  • Orchestration
  • Copy activity
07

Dataflow Gen2

Power Query transformations that write straight into a Lakehouse or a Warehouse, for loads that were originally shaped in Power Query or in SSIS data flows. Low code where the original was low code.

  • Power Query
  • Low code
  • Writes to OneLake
08

Direct Lake under Power BI

The semantic models read the Delta tables directly. There is no import refresh window and no DirectQuery latency, only a metadata-only framing step. The models keep their measures and the reports keep their users.

  • Direct Lake
  • Semantic models
  • Reports unchanged
09

OneLake security and Purview

Permissions are defined on the data once and enforced by every engine that reads it, down to rows and columns. Sensitivity labels travel with the data, lineage spans items, and the OneLake catalog shows what exists and what is endorsed.

  • Row and column level
  • Sensitivity labels
  • Lineage
10

Git, deployment pipelines and capacity

Every item is source-controlled in Git and promoted from development to test to production through deployment pipelines, reviewed like any other code. One capacity runs every workload and is measured in the Capacity Metrics app.

  • Git
  • Dev · Test · Prod
  • F SKUs
04 / In five lines

What changes, and what stays

ADF · Synapse · SSIS

Pipelines become Fabric pipelines.

ADF and Synapse pipelines map activity for activity. SSIS packages are rebuilt as pipelines and Dataflow Gen2 with their schedules kept. All of it lives in the workspace, in Git.

Dedicated pool

The pool becomes the Warehouse.

Schemas, tables and stored procedures move to the Warehouse with their T-SQL kept. The history is backfilled and reconciled before the pool is frozen and retired.

SQL Server

Mirrored, not extracted.

The operational databases replicate into OneLake through the on-premises gateway: a snapshot, then every change, readable in Fabric from the first day.

ADLS Gen2 · Unity Catalog

Shortcuts, not copies.

Files in ADLS Gen2 and tables in Unity Catalog stay where they are and are reached from OneLake as shortcuts, governed once.

Power BI

Direct Lake; the reports stay.

The semantic models switch storage mode from import to Direct Lake. Measures, reports and users are untouched, and the nightly refresh window goes away.

05 / Private assessment

Tell us what the estate is made of

What we need
  1. 01An inventory of the estate: package, pipeline, procedure and table names, and where the reports read from.
  2. 02Who owns what, and which objects are already thought to be unused.
  3. 03Your Fabric position: capacity, workspaces, tenant settings, and who administers them.
  4. 04The constraints that shape the cutover: maintenance windows, regulators, and the reports that cannot be interrupted.

No files or data at this stage