Data Engineering11 min readMar 2025

The Azure Fabric Migration Guide Nobody Wrote (Until Now)

What the Microsoft docs don't tell you

Microsoft's documentation for Fabric is excellent for greenfield deployments. It is not written for teams migrating from existing Azure data estates.

We found that out the hard way.

We had: Azure Synapse Analytics, 200+ pipelines, 15 Power BI workspaces, and a 6-week deadline. Here's the real migration playbook.

What Nobody Tells You About Lakehouse Architecture

Fabric's Lakehouse uses Delta Lake format under the hood. If you're coming from Synapse dedicated SQL pools, your data is in a different format and your column types may not map cleanly.

Decimal precision mismatches caused 3 of our first 10 pipeline failures. Always validate numeric column types on ingest.

Run this check before migrating any table:

SELECT COLUMN_NAME, DATA_TYPE, NUMERIC_PRECISION, NUMERIC_SCALE
FROM INFORMATION_SCHEMA.COLUMNS
WHERE TABLE_NAME = 'your_table'
ORDER BY ORDINAL_POSITION;

The Pipeline Migration Order That Saved Us

Migrate in this exact order or you'll create circular dependency nightmares:

1. Source connections first. Get all linked services and credentials into Fabric Data Factory before touching a single pipeline.
2. Staging/bronze layer. Raw ingestion pipelines only. Validate row counts match source.
3. Transformation/silver layer. Notebook-based transforms. Keep Synapse running in parallel here.
4. Gold/reporting layer last. Only cut over Power BI datasets after upstream data is validated for 5+ days.

The Shortcut That Breaks Everything

Do not use "Import from Synapse" tooling for complex pipelines. It works for simple copy activities. For anything with dynamic expressions, ForEach loops, or nested activities — rewrite manually. The exported JSON rarely translates cleanly and you'll spend more time debugging the import than rewriting.

Power BI Workspace Migration

The most underestimated part. Fabric workspaces have capacity-based licensing. Moving existing Power BI workspaces to Fabric capacity triggers dataset recertification requirements in most enterprise tenants.

Plan 2 weeks of stakeholder comms before the Power BI cutover. The data engineers will finish in 4 weeks. The BI sign-off will take the other 2.

The Go-Live Checklist We Wish We Had

Row count validation on all migrated tables (automated, not manual)
Pipeline run history exported from Synapse for audit trail
Capacity monitoring alerts set before first production run
Rollback plan documented and tested (can you repoint Power BI to Synapse in under 30 minutes?)
On-call roster for the first 72 hours post-cutover