fix: Remote Online Store Type Inference Error with All-NULL Columns by yuan1j · Pull Request #6063 · feast-dev/feast

yuan1j · 2026-03-04T15:30:04Z

What this PR does / why we need it

📋 Problem Description

When reading features from a Feast Remote Online Store, a TypeError is raised if an entire feature column contains only NULL/NaN values. The error occurs in the python_values_to_proto_values() function when it cannot infer the value type from empty data.

🔍 Root Cause Analysis

Component	Issue
Location	`feast/type_map.py` - `python_values_to_proto_values()` function
Trigger	Entire feature column contains `NULL/NaN` values
Reason	Type inference fails when all values are empty, resulting in `UNKNOWN` value type
Current Behavior	Raises `TypeError: Couldn't infer value type from empty value`
Impact	Remote online feature reading fails completely

📊 Test Data Example

                   event_timestamp          driver_id  conv_rate  acc_rate  avg_daily_trips
0  2026-03-04 16:17:44+00:00       1001        NaN    1.000000           1000.0
1  2026-02-17 16:00:00+00:00       1005        NaN    0.708060            395.0
2  2026-02-17 17:00:00+00:00       1005        NaN    0.451616            145.0

⚠️ Note: The conv_rate column contains all NaN values.

💡 Solution

Modify the type inference logic to handle UNKNOWN value types gracefully:

Key Changes

✅ Do not raise error when value type is UNKNOWN
✅ Preserve schema-defined type from FeatureView definition
✅ Return proper NULL values in the response while maintaining type information

Code Change Summary

# Before
if value_type is UNKNOWN:
    raise TypeError("Couldn't infer value type from empty value")

# After
# Remove this part code, and use schema-defined type or skip type inference for NULL-only columns

✨ Expected Behavior

After this fix:

✅ Remote online store can read columns with all-NULL values
✅ Feature schema type is preserved from FeatureView definition
✅ NULL values are returned correctly to the client
✅ No breaking changes to existing functionality

Which issue(s) this PR fixes

Fixes: Remote Online Store Type Inference Error with All-NULL Columns

Related Error:

 Traceback (most recent call last):
  File "C:\svn\project\feature_store\feast\remote\test_workflow.py", line 130, in <module>
    run_demo()
  File "C:\svn\project\feature_store\feast\remote\test_workflow.py", line 25, in run_demo
    fetch_online_features(store)
  File "C:\svn\project\feature_store\feast\remote\test_workflow.py", line 121, in fetch_online_features
    returned_features = store.get_online_features(
  File "C:\Users\j\AppData\Local\Programs\Python\Python310\lib\site-packages\feast\feature_store.py", line 2445, in get_online_features
    response = provider.get_online_features(
  File "C:\Users\j\AppData\Local\Programs\Python\Python310\lib\site-packages\feast\infra\passthrough_provider.py", line 251, in get_online_features
    return self.online_store.get_online_features(
  File "C:\Users\j\AppData\Local\Programs\Python\Python310\lib\site-packages\feast\infra\online_stores\online_store.py", line 199, in get_online_features
    read_rows = self.online_read(
  File "C:\Users\j\AppData\Local\Programs\Python\Python310\lib\site-packages\feast\infra\online_stores\remote.py", line 169, in online_read
    message = python_values_to_proto_values(
  File "C:\Users\j\AppData\Local\Programs\Python\Python310\lib\site-packages\feast\type_map.py", line 838, in python_values_to_proto_values
    raise TypeError("Couldn't infer value type from empty value")
TypeError: Couldn't infer value type from empty value

Misc

🧪 Testing

Manual Test Steps

# 1. Start Feast services
cd C:\svn\project\feature_store\feast\test\feature_repo
feast ui
feast serve_registry
feast serve_offline
feast serve

# 2. Prepare test data with all-NULL column
# (conv_rate column contains only NaN values)

# 3. Run remote client
python remote/test_workflow.py

Configuration Used

project: test
registry: 
  registry_type: remote
  path: 127.0.0.1:6570
provider: local
online_store:
  type: remote
  path: http://127.0.0.1:6566
offline_store:
  type: remote
  host: 127.0.0.1
  port: 8815
entity_key_serialization_version: 3
auth:
  type: no_auth

📝 Additional Notes

Item	Details
Breaking Changes	None
API Changes	None
Documentation	May need update for NULL handling behavior
Unit Tests	Should add test case for all-NULL column scenario
Affected Components	`feast/type_map.py`, `feast/infra/online_stores/remote.py`

👥 Reviewers

@feast-maintainers
@feast-infra-team

✅ Checklist

Code follows Feast code conventions
Unit tests added/updated
Integration tests pass
Documentation updated (if needed)
Commits signed
PR title follows conventional commits format

yuan1j · 2026-03-04T15:33:36Z

小军已收到

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

devin-ai-integration

Devin Review found 1 new potential issue.

View 6 additional findings in Devin Review.

devin-ai-integration · 2026-03-11T06:21:07Z

sdk/python/feast/type_map.py


    if value_type == ValueType.UNKNOWN:
-        raise TypeError("Couldn't infer value type from empty value")
+        return [ProtoValue() for _ in values]


🔴 Silent data loss for empty-collection values (e.g., []) now treated as null instead of erroring

Replacing the TypeError with return [ProtoValue() for _ in values] silently converts legitimate empty-collection values (like []) into null ProtoValues. The _non_empty_value function at sdk/python/feast/type_map.py:1351-1360 treats empty lists/sets as "empty" (returns False), so when the only values are empty collections, sample is None, value_type stays UNKNOWN, and the new code returns null ProtoValues.

This causes silent data corruption in at least two call paths:

_python_dict_to_map_proto (line 983): A struct like {"scores": []} will have its empty-list value silently converted to a null ProtoValue, losing the semantic distinction between "empty list" and "null".

remote.py:205-213: When the remote online store returns a feature with status "PRESENT" and value [] (a valid empty list), the function returns ProtoValue() (null). The caller then stores this as the feature value (feature_values_dict[feature_name] = message[0]), silently dropping the actual empty-list value.

A safer fix would be to only return empty ProtoValues when all values are actually None, not when they are empty collections.

Prompt for agents

In sdk/python/feast/type_map.py at line 1017-1018, the fix for all-NULL columns is too broad — it also silently converts empty-collection values (like []) to null ProtoValues. Instead of unconditionally returning empty ProtoValues when value_type is UNKNOWN, check whether ALL values are actually None. If so, return empty ProtoValues (the intended fix for all-NULL columns). If some values are non-None but just empty collections (like []), either raise the original TypeError or attempt more aggressive type inference. For example: if value_type == ValueType.UNKNOWN: if all(v is None for v in values): return [ProtoValue() for _ in values] raise TypeError("Couldn't infer value type from empty value")

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

yuan1j

Thank you for your helpful advice. This solution works for my case as well. Could you please review and approve the merge?

fix: Remote Online Store Type Inference Error with All-NULL Columns

4dce5bf

yuan1j requested a review from a team as a code owner March 4, 2026 15:30

This comment was marked as resolved.

Sign in to view

Update sdk/python/feast/type_map.py

6d04918

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

devin-ai-integration bot reviewed Mar 11, 2026

View reviewed changes

yuan1j commented Mar 11, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: Remote Online Store Type Inference Error with All-NULL Columns#6063

fix: Remote Online Store Type Inference Error with All-NULL Columns#6063
yuan1j wants to merge 2 commits intofeast-dev:masterfrom
yuan1j:fix_remote_online_type_error_with_all_null_columns

yuan1j commented Mar 4, 2026 •

edited by devin-ai-integration bot

Loading

Uh oh!

This comment was marked as resolved.

Uh oh!

yuan1j commented Mar 4, 2026 via email

Uh oh!

devin-ai-integration bot left a comment

Uh oh!

devin-ai-integration bot Mar 11, 2026

Uh oh!

yuan1j left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

yuan1j commented Mar 4, 2026 • edited by devin-ai-integration bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this PR does / why we need it

📋 Problem Description

🔍 Root Cause Analysis

📊 Test Data Example

💡 Solution

Key Changes

Code Change Summary

✨ Expected Behavior

Which issue(s) this PR fixes

Misc

🧪 Testing

Manual Test Steps

Configuration Used

📝 Additional Notes

👥 Reviewers

✅ Checklist

Uh oh!

This comment was marked as resolved.

Uh oh!

yuan1j commented Mar 4, 2026 via email

Uh oh!

devin-ai-integration bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration bot Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

yuan1j left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

yuan1j commented Mar 4, 2026 •

edited by devin-ai-integration bot

Loading