Enter your search text in the box above

Select a result to preview

The Ultimate "NO ETL" Show Down: Shortcuts vs Mirroring vs Copy Job

#fabcon2026 #session #learning #microsoft #fabcon2026-session

Description

No ETL - no problem! Fabric introduced several innovative features to access external data without spending time on building complex ETL.

But, which one is right for YOU? We'll examine various data movement strategies, their pros and cons, so you'll feel more confident picking the right one

Key Takeaways

My Notes

Action Items

Slides

📥 Download Slides

The Ultimate “NO ETL” Showdown
Shortcuts vs. Mirroring vs. Copy Job
Nikola Ilic | Data Mozart | Microsoft Data Platform MVP
Your “No ETL” Showdown Host
Principal Architect
Nikola Ilic
➢ I'm making music from the data!
➢ Power BI and Fabric addict, blogger, speaker...
➢ Father of 2, Barca & Leo Messi fan...
data-mozart.com
You are paying an ETL tax…Let’s end that today!
Every hour spent wiring pipelines, debugging flows, and managing schedules is an
hour NOT spent on actual insights!
The “NO ETL” promise
Fabric provides three distinct ways to access external data without building pipelines.
Three approaches -> One goal
Stop buying the DVD – just press PLAY!
What
Does"No
“NoETL"
ETL”
Really
Mean?
What Does
Really
Mean?
Traditional ETL
Extract
"No ETL" in Fabric
Transform
Hours of pipeline building. Rinse & repeat.
Load
External Data
Fabric
Access external data WITHOUT building pipelines
Meet the Contestants
Three fighters. One shared goal. Only YOUR use case decides the winner.
Contestant 1: Shortcuts
Bookmark
✓ Zero-copy symbolic links. Data never moves
✓ Like streaming a movie: zero download, just press play
✓ Sources
✓ ADLS Gen2
✓ Amazon S3
✓ GCS
✓ Dataverse
✓ SharePoint
✓ OneDrive
✓ Cost: Egress fees only
Contestant 1: Shortcuts
Bookmark
✓ Zero-copy symbolic links. Data never moves
✓ Like streaming a movie: zero download, just press play
✓ Sources
✓ ADLS Gen2
✓ Amazon S3
✓ GCS
✓ Dataverse
✓ SharePoint
✓ OneDrive
✓ Cost: Egress fees only
How Shortcuts work?
The data never moves. Fabric looks through a window to see your data on the other side
External Shortcuts in a nutshell
Lakehouse
External storage
Tables
Delta Table
shortcut
Amazon S3
Delta Table
Iceberg Table
shortcut
Iceberg Table
ADLS Gen2
Files
Google Cloud
Folder shortcut
Folders/Files
Internal Shortcuts
Workspace A
Workspace B
Table 3
Lakehouse 1
Lakehouse 3
Shortcut path
Lakehouse 2
Target path
Table 2
Shortcuts:
TheGood,
Good,
and
the Gotchas
Shortcuts: The
the the
Bad, Bad,
and the
Gotchas
THE GOOD
THE BAD
GOTCHAS
• Zero data movement
• Performance depends on external
source
• Files >1GB not cached (cross-cloud)
• Instant setup in seconds
• Cannot shortcut to databases
• No non-Latin chars in names
• No storage cost in Fabric
• No transformations possible
• Max 5 shortcut-to-shortcut depth
• Works with Direct Lake
• Egress fees from cloud providers
• Caching only for S3/GCS
• Up to 100K per item
Demo
Shortcuts in action
Creating a shortcut to ADLS Gen2 and querying from the lakehouse
Contestant 2: Mirroring
The live photocopier
✓ Continuous – Change Data Capture (CDC) only copies what changed
✓ Three flavors – Database, Metadata (Databricks), Open
✓ Cost – Replication compute in Fabric is FREE (and some of the storage☺)
Mirroring:
Three
Distinct
Flavors
Mirroring: Three Unique Flavors
Database Mirroring
Metadata Mirroring
Open Mirroring
Full replication of tables
into OneLake as Delta
Syncs catalog structure
ONLY, NOT actual data
Any app writes CDC data
into mirrored database
Azure SQL DB, SQL MI,
SQL Server 2016-2025,
Snowflake, Cosmos DB,
PostgreSQL
Azure Databricks
Unity Catalog
(hybrid: metadata +
shortcuts)
Oracle GoldenGate,
Qlik, Striim, SAP,
CData + 10 more
partners
The Magic of Change Data Capture (CDC)
Mirroring doesn’t copy everything every time.
It only captures what changed, converting it instantly into open formats
Mirroring:
TheGood,
Good,
and
the Gotchas
Mirroring: The
the the
Bad,Bad,
and the
Gotchas
THE GOOD
THE BAD
GOTCHAS
• Near real-time sync
• Destination is read-only
• No custom scheduling
• FREE replication compute
• Limited to 1,000 tables
• Always-on replication
• Delta format -> Direct Lake
• No transformations
• Stop + restart = full reseed!
• Predictable local performance
• Can't shortcut to databases
• CDC conflict with SQL Server
• Auto converts to open format
Demo
Mirroring in action
Setting up database mirroring
Contestant 3: Copy Job
The moving company
✓ You dictate what to move, when, and where
✓ Scheduled or on-demand physical data copies
✓ Over 80+ supported connectors
Complete Control
✓ Granular table and column mapping -> rename on the fly
✓ Native incremental loads using watermark columns and upserts
Demo
Copy Job in action
Creating a Copy Job with incremental load
It’s show time…
Round #1: Architecture and movement
Shortcuts: STREAMING
Mirroring: DVR
Copy Job: DOWNLOADS
The movie stays on their servers.
Zero data movement
Auto-records new episodes.
Continuous CDC into OneLake
You pick what and when.
Scheduled batch movement
Round #2: Performance and Cost
Variable
Performance
Depends on external source speed
Egress only
Fast
Predictable
Local data in OneLake
Direct Lake optimized
Local after copy
Schedule-dependent
FREE Compute!
CU Based
Cost
No Fabric storage
Cloud provider fees
Storage based on capacity SKU
Pay for compute during copy runs
For raw query speed: Local data always wins (Mirroring & Copy Job)
The Comparison Matrix

Screenshot this!
Feature
Shortcuts
Mirroring
Copy Job
Data movement
None (pointer)
Continuous CDC
Scheduled
Latency
Real-time*
Near real-time
Schedule-based
Sources
Storage only
Databases
80+ connectors
Destination
OneLake (virtual)
OneLake (read-only)
Multiple targets
Cost
Egress fees only
FREE compute!
CU consumption
Scheduling
N/A
Always-on
Custom
Flexibility
Low
Low
High
Direct Lake
Yes (external perf.)
Yes (optimized!)
Yes (after copy)