Mirroring (Microsoft Fabric)
Mirroring in Microsoft Fabric brings source data into its analytics environment through database replication, metadata shortcuts, or an open change-data landing interface. Database and open mirroring produce Delta tables in OneLake; metadata mirroring leaves the data at its source and reads it through shortcuts.
Three kinds of mirroring
| Type | What moves | Example sources in the documentation |
|---|---|---|
| Database mirroring | Data: tables are replicated into OneLake and changes are merged in near real time | Azure SQL Database, SQL Server, Azure Cosmos DB, PostgreSQL, Snowflake, Oracle, Google BigQuery |
| Metadata mirroring | Only catalog, schema, and table metadata; the data is read in place through shortcuts | Azure Databricks Unity Catalog |
| Open mirroring | Change data written by any application to a landing zone through public APIs, then merged into Delta tables | Custom applications and third-party providers |
Each mirrored database also gets a read-only SQL analytics endpoint.
How current, and at what cost
For database mirroring, the documentation says changes can be published as fast as every 15 seconds, and that near real-time latency depends on the source and destination regions, the volume and frequency of changes, network conditions, and gateway resources for on-premises sources. That is a range to measure, not a guarantee.
- Storage: one terabyte of mirroring storage free for each capacity unit purchased (64 TB on an F64); beyond that, or while the capacity is paused, normal OneLake storage charges apply.
- Replication compute: free, and it does not consume the capacity. Querying the mirrored data with SQL, Spark, or Power BI is charged normally.
- A running capacity is required. If the capacity is paused or deleted, no data is replicated.
What it does not remove
Database mirroring is still replication, so the usual responsibilities of a copy remain. The replica lags the source by an amount that should be measured. Old files are vacuumed according to a retention setting — one day by default for mirrored databases created after mid-2025, per the documentation — which also bounds how far back Delta time travel can reach. Changes in the source schema, and the effect of capturing changes on the source database, need to be checked against the documentation for that particular source. Mirroring removes the pipeline to build; it does not remove the need to know how fresh the data is and who owns the source. How it compares with shortcuts and fits into Fabric’s architecture is worked through in Microsoft Fabric Architecture: OneLake, Capacity, and the Move from Synapse.
Further reading
Discover more from Insightful Data Lab
Subscribe to get the latest posts sent to your email.