When you materialize a dagster asset, how does dagster decide which materialization result of the upstream asset to use as an input?

When you materialize a Dagster asset, Dagster uses the asset's dependencies to determine which upstream assets need to be materialized first. The materialization result of an upstream asset that is used as input for a downstream asset is typically the latest materialization of that upstream asset. However, if the assets are partitioned, Dagster will use the partition mapping to decide which partitions of the upstream asset to use. By default, if the upstream asset is partitioned and a partition mapping is not provided, Dagster will use the default partition mapping for the partitions definition, which typically maps partition keys to the same partition keys in upstream assets. In the case of more complex scenarios, such as when an asset depends on the latest materialized partition of an upstream asset, the documentation suggests that there isn't a straightforward way to handle this within Dagster's current feature set. One workaround mentioned is to define an

AllPartitionMapping

between the upstream and downstream assets, which would require the downstream asset to load all upstream partitions and then programmatically determine the latest partition with an existing output value within the downstream asset's code. For non-partitioned assets, Dagster will use the latest materialization of the upstream asset by default. If there are specific requirements or custom logic needed to select a particular materialization result, this would need to be implemented within the asset's code or through custom scheduling or sensing logic. It's important to note that the exact behavior can depend on the configuration and definitions set up in your Dagster code, including how you've defined your assets, their dependencies, and any partitioning logic.