Some steps naturally produce several tables at once: one API call returns customers and orders, one SQL script writes three tables. @dg.multi_asset(specs=[dg.AssetSpec("customers"), dg.AssetSpec("orders")]) declares all of them on one function. Return or yield one dg.MaterializeResult(asset_key=...) per asset.
Each key is a full asset: its own lineage, metadata and history. Downstream assets depend on customers or orders individually.
By default a multi-asset runs as a unit. With can_subset=True, Dagster can select just one of its assets, and the function reads context.selected_asset_keys to produce only those. Every selected key must get a result, or the step fails naming the missing output.
Two assets from one sync.