GeezerVison

The World Wide Web, restored to 800 pixels of glory.

GeezerVison is showing github.com — fix: ODFV output projection in offline retrieval (#6099) by jyejare · Pull Request #6140 · feast-dev/feast

fix: ODFV output projection in offline retrieval (#6099) #6140

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged

ntkathole merged 1 commit into feast-dev:master from jyejare:fix/odfv-output-projection-6099

Apr 2, 2026

Uh oh!

There was an error while loading. Please reload this page.

Merged

fix: ODFV output projection in offline retrieval (#6099) #6140

Show file tree

Hide file tree

Changes from all commits

Commits

Show all changes

1 commit

Select commit

f9a2751

Fix ODFV output projection in offline retrieval (#6099)

Mar 23, 2026

File filter

Filter by extension .py (4) All 1 file type selected

Viewed files

Clear filters

Conversations

Failed to load comments.

Uh oh!

There was an error while loading. Please reload this page.

Jump to

Jump to file

Failed to load files.

Uh oh!

There was an error while loading. Please reload this page.

Diff view

Unified

Split

Hide whitespace

Diff view

Unified

Split

Hide whitespace

Some comments aren't visible on the classic Files Changed page.

There are no files selected for viewing

15 changes: 14 additions & 1 deletion 15 sdk/python/feast/feature_store.py

Show comments

View file

Open in desktop

Comment thread

devin-ai-integration Bot Apr 2, 2026

Copy link

Copy Markdown

Contributor

There was a problem hiding this comment.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Feature name collision validation bypassed for implicitly-added ODFV dependency refs

_validate_feature_refs at sdk/python/feast/feature_store.py:1476 is called with _feature_refs (the user's original refs), but the provider receives _feature_refs_for_provider which includes additional dependency FV refs added at lines 1454-1461. If full_feature_names=False and an ODFV dependency feature has the same name as a user-requested feature from a different FV (e.g., user requests fv1:conv_rate and the ODFV depends on fv2:conv_rate), the collision is not detected by validation, and both features produce identically-named output columns, leading to data corruption or errors in the offline store's point-in-time join.

(Refers to line 1476)

Was this helpful? React with 👍 or 👎 to provide feedback.

Sorry, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

All reactions

Original file line number	Diff line number	Diff line change
Expand Up		@@ -1447,6 +1447,19 @@ def get_historical_features(
		feature_views = list(view for view, _ in fvs)
		on_demand_feature_views = list(view for view, _ in odfvs)

		# ODFV source FV dependencies (e.g. driver_stats:conv_rate) are resolved
		# by _group_feature_refs and included in `fvs`, but not in _feature_refs.
		# Offline stores use feature_refs to map which features to fetch from each
		# FV, so we must include these implicit dependency refs.
		_feature_refs_for_provider = list(_feature_refs)
		existing_refs = set(_feature_refs)
		for view, feats in fvs:
		for feat in feats:
		ref = f"{view.projection.name_to_use()}:{feat}"
		if ref not in existing_refs:
		_feature_refs_for_provider.append(ref)
		existing_refs.add(ref)

		# Check that the right request data is present in the entity_df
		if type(entity_df) == pd.DataFrame:
		if self.config.coerce_tz_aware:
Expand All		@@ -1473,7 +1486,7 @@ def get_historical_features(
		job = provider.get_historical_features(
		self.config,
		feature_views,
		_feature_refs,
		_feature_refs_for_provider,
		entity_df,
		self.registry,
		self.project,
Expand Down

48 changes: 45 additions & 3 deletions 48 sdk/python/feast/infra/offline_stores/offline_store.py

Show comments

View file

Open in desktop

Original file line number	Diff line number	Diff line change
Expand Up		@@ -154,17 +154,59 @@ def to_arrow(
		"""
		features_table = self._to_arrow_internal(timeout=timeout)
		if self.on_demand_feature_views:
		# Build a mapping of ODFV name to requested feature names
		# This ensures we only return the features that were explicitly requested
		odfv_feature_refs: Dict[str, set[str]] = {}
		try:
		metadata = self.metadata
		except NotImplementedError:
		metadata = None

		if metadata and metadata.features:
		for feature_ref in metadata.features:
		if ":" in feature_ref:
		view_name, feature_name = feature_ref.split(":", 1)
		# Check if this view_name matches any of the ODFVs
		for odfv in self.on_demand_feature_views:
		if (
		odfv.name == view_name
		or odfv.projection.name_to_use() == view_name
		):
		if view_name not in odfv_feature_refs:
		odfv_feature_refs[view_name] = set()
		# Store the feature name in the format that will appear in transformed_arrow
		expected_col_name = (
		f"{odfv.projection.name_to_use()}__{feature_name}"
		if self.full_feature_names
		else feature_name
		)
		odfv_feature_refs[view_name].add(expected_col_name)

		for odfv in self.on_demand_feature_views:
		transformed_arrow = odfv.transform_arrow(
		features_table, self.full_feature_names
		)

		# Determine which columns to include from this ODFV
		# If we have metadata with requested features, filter to only those
		# Otherwise, include all columns (backward compatibility)
		requested_features_for_odfv = (
		odfv_feature_refs.get(odfv.name)
		if odfv.name in odfv_feature_refs
		else odfv_feature_refs.get(odfv.projection.name_to_use())
		)

		for col in transformed_arrow.column_names:
		if col.startswith("__index"):
		continue
		features_table = features_table.append_column(
		col, transformed_arrow[col]
		)
		# Only append the column if it was requested, or if we don't have feature metadata
		if (
		requested_features_for_odfv is None
		or col in requested_features_for_odfv
		):
		features_table = features_table.append_column(
		col, transformed_arrow[col]
		)

		if validation_reference:
		if not flags_helper.is_test():
Expand Down

1 change: 0 additions & 1 deletion 1 sdk/python/feast/utils.py

Show comments

View file

Open in desktop

Original file line number	Diff line number	Diff line change
Expand Up		@@ -127,7 +127,6 @@ def compute_non_entity_date_range(
		end_date: Optional[datetime] = None,
		default_window_days: int = 30,
		) -> Tuple[datetime, datetime]:

		if end_date is None:
		end_date = datetime.now(tz=timezone.utc)
		else:
Expand Down

99 changes: 99 additions & 0 deletions 99 sdk/python/tests/integration/offline_store/test_universal_historical_retrieval.py

Show comments

View file

Open in desktop

Original file line number	Diff line number	Diff line change
Expand Up		@@ -840,3 +840,102 @@ def test_historical_features_non_entity_retrieval(environment):
		assert 300 in actual_trips, (
		"Latest trip value 300 for driver 1002 should be present"
		)


		@pytest.mark.integration
		@pytest.mark.universal_offline_stores
		@pytest.mark.parametrize("full_feature_names", [True, False], ids=lambda v: f"full:{v}")
		def test_odfv_projection(environment, universal_data_sources, full_feature_names):
		"""
		Test that requesting a subset of ODFV features only returns those features.

		Regression test for issue #6099: OnDemandFeatureViews should honor output
		projection in offline retrieval, matching the behavior of online retrieval.

		Before the fix, offline retrieval would return ALL ODFV output features even
		when only a subset was requested, while online retrieval correctly returned
		only the requested features.
		"""
		store = environment.feature_store

		(entities, datasets, data_sources) = universal_data_sources

		feature_views = construct_universal_feature_views(data_sources)

		# Add request data needed for ODFV
		entity_df_with_request_data = datasets.entity_df.copy(deep=True)
		entity_df_with_request_data["val_to_add"] = [
		i for i in range(len(entity_df_with_request_data))
		]

		store.apply([driver(), *feature_views.values()])

		# The conv_rate_plus_100 ODFV has 3 output features:
		# - conv_rate_plus_100
		# - conv_rate_plus_val_to_add
		# - conv_rate_plus_100_rounded

		# Test 1: Request only ONE ODFV feature
		job = store.get_historical_features(
		entity_df=entity_df_with_request_data,
		features=[
		"conv_rate_plus_100:conv_rate_plus_100", # Request only this one
		],
		full_feature_names=full_feature_names,
		)

		actual_df = job.to_df()

		# Determine expected column names based on full_feature_names setting
		expected_feature = (
		"conv_rate_plus_100__conv_rate_plus_100"
		if full_feature_names
		else "conv_rate_plus_100"
		)
		unrequested_feature_1 = (
		"conv_rate_plus_100__conv_rate_plus_val_to_add"
		if full_feature_names
		else "conv_rate_plus_val_to_add"
		)
		unrequested_feature_2 = (
		"conv_rate_plus_100__conv_rate_plus_100_rounded"
		if full_feature_names
		else "conv_rate_plus_100_rounded"
		)

		# Verify the requested feature is present
		assert expected_feature in actual_df.columns, (
		f"Requested feature '{expected_feature}' should be in the result"
		)

		# Verify unrequested ODFV features are NOT present (this is the key fix)
		assert unrequested_feature_1 not in actual_df.columns, (
		f"Unrequested ODFV feature '{unrequested_feature_1}' should NOT be in the result. "
		f"This indicates the bug from issue #6099 still exists."
		)
		assert unrequested_feature_2 not in actual_df.columns, (
		f"Unrequested ODFV feature '{unrequested_feature_2}' should NOT be in the result. "
		f"This indicates the bug from issue #6099 still exists."
		)

		# Test 2: Request TWO out of THREE ODFV features
		job2 = store.get_historical_features(
		entity_df=entity_df_with_request_data,
		features=[
		"conv_rate_plus_100:conv_rate_plus_100",
		"conv_rate_plus_100:conv_rate_plus_val_to_add",
		# Deliberately NOT requesting conv_rate_plus_100_rounded
		],
		full_feature_names=full_feature_names,
		)

		actual_df2 = job2.to_df()

		# Verify the two requested features are present
		assert expected_feature in actual_df2.columns
		assert unrequested_feature_1 in actual_df2.columns

		# Verify the unrequested feature is NOT present
		assert unrequested_feature_2 not in actual_df2.columns, (
		f"Unrequested ODFV feature '{unrequested_feature_2}' should NOT be in the result"
		)

Oops, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

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.

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.