Fabric Deployment Pipelines

Hi,

I have tried to differentiat between what Microsoft docs says from what might actually happen in fabric deployment pipelines, hth!

Semantic models - "only metadata" vs actual data

• What Microsoft documentation means by "only metadata is deployed" is that the deployment pipeline process does not trigger a data refresh in the target stage. It takes the model definition (tables, measures, relationships, etc) from source and applies it to target.
• However, in reality if the target stage already has a dataset with the same ID and there is cached data present, the deployment will overwrite the metadata but leave the existing data in place. That is why after deployment, when you open the semantic model in the target stage, you see all tables already populated, it is showing the previously cached data, not newly fetched data.
• If you delete the target dataset before deployment, or if the model definition changes enough to force a full data clear (example: removing/replacing tables), you will see the “no data” state and have to refresh.
• So in practice:
1. Fresh target dataset (no prior data) > post-deployment, it will be empty until you refresh.
2. Target dataset with prior cache > post-deployment, data appears instantly, but it is stale until refreshed.

This behaviour is often misunderstood as “deployment is copying data,” but technically it is just leaving the old cache intact.

Dataflows Gen1 in Deployment Pipelines

• Official documentation says: "Deployment pipelines support Dataflow Gen2" because Gen2 is the new Fabric-native storage layer.
• Dataflow Gen1 (the Power BI Service version) is not natively supported in Fabric deployment pipelines for proper stage management like Gen2.
• However, when you test with Dataflow Gen1 inside a deployment pipeline, it may appear to work because the pipeline is essentially copying the Dataflow definition (JSON) to the target workspace. This is more of a “content clone” than a true pipeline-managed object.
• The subtle difference:
• Gen2: The deployment pipeline understands lineage, refresh scheduling, env parameters, etc.
• Gen1: It just replicates the object, but there is no Fabric-native governance or parameterisation support. You can end up with both environments pointing to the same backend source unless you manually change parameters.

Please 'Kudos' and 'Accept as Solution' if this answered your query.