Data & Analytics
Fabric mirroring can be free while its change feed consumes capacity
Budget Fabric Mirroring Delta change data feed separately from core replication before scaling an F capacity.
November 18, 2026
A Fabric team turns on Azure SQL mirroring, enables Delta change data feed for downstream bronze processing, and sees “Mirror Replication Premium” in Capacity Metrics. In a recent Fabric practitioner discussion, the team asks whether enabling more databases will consume too much of an F16. The useful distinction is in Microsoft’s billing guide: core mirroring compute is free, while extended capabilities such as Delta change data feed are optional and billed for replication work.
Price the capability you actually need
Document why each consumer needs every intermediate change rather than the latest mirrored table state. If a downstream job only needs current values, do not make a paid change feed the default. If audit or state-transition analysis truly requires CDF, record the tables and consumers that depend on it before changing the mirror configuration.
Run a bounded capacity pilot
Enable one representative mirror first and observe the operation name and DataMovementIncrementalCopy meter over normal and peak source-change periods. Microsoft says the billing scope is the mirror-level workload and that idle or empty iterations are not charged. A duration counter that matches wall-clock time is therefore not, by itself, a bill. Compare actual CU consumption with source change rate and downstream compute in Capacity Metrics before extrapolating to four databases.
Keep the total decision visible
Measure the alternative pipeline as well: direct queries, scheduled copy, or CDF each trades cost against freshness and access to intermediate states. Set a capacity budget and an alert threshold for the combined mirror, notebooks, pipelines, and reports. Revisit the design if the feed uses the headroom reserved for business reporting.
Adovo’s view
“Free mirroring” is true for the core service and incomplete for the architecture around it. A platform choice is economical only when its required capabilities fit the capacity left for everything else.
