Repository navigation
Conversation
…mpty BigQueryOfflineStore.pull_all_from_table_or_query and pull_latest_from_table_or_query built their SELECT list from the join keys, feature columns, and timestamps only. FeatureBuilder signals "read all source columns" for BatchFeatureViews with a pandas/python UDF by passing an empty feature_name_columns list, so the generated query selected only the join keys and timestamps and materialization failed with "Column <feature> not found in table". Handle the empty list the way the other offline stores do: select all columns (SELECT * in pull_all; SELECT * in the inner query and SELECT * EXCEPT(_feast_row) in the outer query of pull_latest, which uses a ROW_NUMBER helper column). Fixes feast-dev#7047 Signed-off-by: Mrun25 <[email protected]>
There was a problem hiding this comment.
Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.
An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #7048 +/- ##
=======================================
Coverage 50.99% 50.99%
=======================================
Files 455 455
Lines 56994 56999 +5
Branches 8321 8323 +2
=======================================
+ Hits 29063 29067 +4
Misses 25946 25946
- Partials 1985 1986 +1
... and 1 file with indirect coverage changes Continue to review full report in Codecov by Harness.
🚀 New features to boost your workflow:
|
Mrun25
changed the title
What this PR does / why we need it:
FeatureBuilder.get_column_infosignals "read all source columns" for BatchFeatureViews with a pandas/python (or ray/flink) UDF by passing an emptyfeature_name_columnslist.BigQueryOfflineStore.pull_all_from_table_or_queryandpull_latest_from_table_or_querydid not handle the empty case: they built the select list from join keys + feature columns + timestamps, so the generated query selected only the join keys and timestamps. Materializing such a view from BigQuery then failed withValueError: Column <feature> not found in tablebecause the UDF never received the source columns.This PR handles the empty list the way the other offline stores do:
pull_all_from_table_or_query:SELECT *pull_latest_from_table_or_query:SELECT *in the inner query andSELECT * EXCEPT(_feast_row)in the outer query, so the ROW_NUMBER helper column is not exposedExplicit (non-empty) feature lists are unchanged. Adds regression tests for both pull paths.
Which issue(s) this PR fixes:
Fixes #7047
Testing
test_pull_all_empty_feature_columns_selects_allandtest_pull_latest_empty_feature_columns_selects_allintests/unit/infra/offline_stores/test_bigquery.pytest_bigquery.py+test_bigquery_non_entity_mode.pysuites pass locally (38 passed)ruff checkandruff format --checkclean