Description

haoxu0

opened on Oct 22, 2025

Issue body actions

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

With PR #5666, OnDemandFeatureView supports aggregations. But it is exclusive with transformation.
the reason is the infer_schema will break if having both defined in feature views.
This can be solved by reworking the infer_schema.

    def infer_features(self) -> None:
        random_input = self._construct_random_input(singleton=self.singleton)
        inferred_features = self.feature_transformation.infer_features(
            random_input=random_input, singleton=self.singleton
        )

Schema inference doesn't run the aggregation step - it only:

1. Looks at the source feature view's schema
2. Creates fake random data matching those columns
3. Passes it directly to the UDF
   But at runtime, the UDF receives aggregated columns, not the original source columns.
   So the UDF would need to be written for aggregated input:

  def my_odfv(inputs):
      return {"trips_doubled": inputs["sum_trips"] * 2}  # expects sum_trips

But schema inference passes it:

  {"trips": [10], "rating": [4.5]}

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Reacted by Francisco Javier Arceo

Activity

1. haoxu0

   added

   kind/featureNew feature or requestNew feature or request

   on Oct 22, 2025
2. franciscojavierarceo commented on Oct 22, 2025

   

   franciscojavierarceo

   on Oct 22, 2025

   Member

   More actions

   @ntkathole @jyejare FYI
3. astronautas commented on Nov 11, 2025

   

   astronautas

   on Nov 11, 2025

   Contributor

   More actions

   Also seems ODFV materialize has a completely different compute engine flow, be careful there. Maybe legacy split or smth.
4. franciscojavierarceo commented on Nov 11, 2025

   

   franciscojavierarceo

   on Nov 11, 2025

   Member

   More actions

   yeah we need to think through how to align the compute engine with ODFVs. ODFVs basically use 2 compute engines: one batch one and the feature server. We have to work out how best to make that cleaner.
5. haoxu0 commented on Nov 12, 2025

   

   haoxu0

   on Nov 12, 2025

   CollaboratorAuthor

   More actions

   yep. So we need to refactor OnDemandFeatureView to:

   1. Inherit from FeatureView
   2. use the current feature resolver/builder

   Reacted by Lukas Valatka and Francisco Javier Arceo
6. haoxu0 commented on Nov 12, 2025

   

   haoxu0

   on Nov 12, 2025

   CollaboratorAuthor

   More actions

   another tech debt found: #5716, forcing to refactor the OnDemandFeatureView

   Reacted by Francisco Javier Arceo and Lukas Valatka
7. nquinn408

   added 2 commits that reference this issue on Apr 24, 2026

   feat: Make udf optional if agg defined (feast-dev#5689)

   8c8d87d

   feat: Make udf optional if agg defined (#5689) (#6328)

   f630056
8. franciscojavierarceo

   added a commit that references this issue on May 4, 2026

   chore(release): release 0.63.0

   50ad181

Issue actions