feat: Support aggregation in odfv #5666
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 3 commits
October 14, 2025 15:28
Signed-off-by: hao-xu5 <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
HaoXuAI
commented
Oct 20, 2025
| if odfv.mode == "python": | ||
| # Apply aggregations if configured. | ||
| if odfv.aggregations: | ||
| if odfv.mode == "python": |
Collaborator
Author
There was a problem hiding this comment.
Currently if this is a python input, we don't have aggregation function to use OOTD, so it use the pandas Dataframe backend to aggregate. I think in long term we need to create a real time compute engine, where we support python native compute operations such as aggregations, filtering, joining etc.
Collaborator
Author
There was a problem hiding this comment.
Member
There was a problem hiding this comment.
what does OOTD mean ?
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR adds aggregation support to OnDemandFeatureView (ODFV) and refactors the backend dataframe module structure. The implementation allows ODFVs to apply aggregations during online feature serving, either using aggregations or feature transformations (but not both). The backend modules are moved from the local directory to a parent-level backends directory to enable reusability across different compute engines like Spark.
Key Changes:
- Moved backend dataframe modules from
local/backendsto parent-levelbackendsdirectory - Added aggregation support to ODFV with
_apply_aggregations_to_responseutility function - Aggregations and transformations are mutually exclusive in ODFVs to simplify schema inference
Reviewed Changes
Copilot reviewed 8 out of 10 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| sdk/python/tests/unit/test_on_demand_feature_view_aggregation.py | New test file covering aggregation functionality in Python and Pandas modes |
| sdk/python/feast/utils.py | Added aggregation utility functions and integrated aggregation logic into ODFV transformation pipeline |
| sdk/python/feast/infra/compute_engines/local/nodes.py | Updated import path for DataFrameBackend |
| sdk/python/feast/infra/compute_engines/local/feature_builder.py | Updated import path for DataFrameBackend |
| sdk/python/feast/infra/compute_engines/local/compute.py | Updated import paths for DataFrameBackend and BackendFactory |
| sdk/python/feast/infra/compute_engines/backends/polars_backend.py | Updated import path for base DataFrameBackend |
| sdk/python/feast/infra/compute_engines/backends/pandas_backend.py | Updated import path for base DataFrameBackend |
| sdk/python/feast/infra/compute_engines/backends/factory.py | Updated import paths for backend classes |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
Co-authored-by: Copilot <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
Signed-off-by: hao-xu5 <[email protected]>
ntkathole
reviewed
Oct 22, 2025
Signed-off-by: hao-xu5 <[email protected]>
franciscojavierarceo
pushed a commit
that referenced
this pull request
Oct 27, 2025
# [0.56.0](v0.55.0...v0.56.0) (2025-10-27) ### Bug Fixes * Add mode field to Transformation proto for proper serialization ([2390d2e](2390d2e)) * Date wise remote offline store historical data retrieval ([#5686](#5686)) ([949ba3d](949ba3d)) * Fix STRING type handling in on-demand feature views ([#5669](#5669)) ([dfbb743](dfbb743)) * Fixed torch install issue in CI ([366e5a8](366e5a8)) * ODFV not getting counted in resource count ([1d640b6](1d640b6)) * Skip tag updates if user do not have permissions ([#5673](#5673)) ([0a951ce](0a951ce)) ### Features * Add document of Go feature server. ([#5697](#5697)) ([cbd1dde](cbd1dde)) * Add flexible commandArgs support for complete Feast CLI control ([#5678](#5678)) ([6414924](6414924)) * Add HDFS as a feature registry ([#5655](#5655)) ([4c65872](4c65872)) * Add nodeSelector to service config ([#5675](#5675)) ([9728cde](9728cde)) * Add OTEL based observability to the Go Feature Server ([#5685](#5685)) ([f4afdad](f4afdad)) * Added health endpoint for the UI ([#5665](#5665)) ([3aec5d5](3aec5d5)) * Added kuberay support ([e0b698d](e0b698d)) * Added support for filtering multi-projects ([#5688](#5688)) ([eb0a86e](eb0a86e)) * Batch Embedding at scale for RAG with Ray ([cc2a46d](cc2a46d)) * Optimize SQL entity handling without creating temporary tables ([#5695](#5695)) ([aa2c838](aa2c838)) * Support aggregation in odfv ([#5666](#5666)) ([564e965](564e965)) * Support cache_mode for registries ([021e9ea](021e9ea))
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What this PR does / why we need it:
local, as it doesn't have to be local abstraction, and spark can inherit it as well.aggregationsconfig into the OnDemandFeatureView. When transforming the features, it can run eitheraggregationsorfeature_transform, not both. The reason is to make the the schema inference simple. But in future this can be redesigned to support both.Which issue(s) this PR fixes:
Misc