feat: Valkey Online Write Batch Vector Search Support by Manisha4 · Pull Request #351 · ExpediaGroup/feast

Manisha4 · 2026-03-31T17:31:20Z

What this PR does / why we need it:

                                                                                                                                                                                                                                                                                                                                                                                                                                 Adds vector embedding storage and indexing support to the Valkey online store, enabling vector similarity search capabilities in Feast (Phase 1 - Write Support).

Changes

Implementation (eg_valkey.py):

Add vector field detection via field.vector_index property
Store vector fields with original field names + raw numpy bytes (required for Valkey Search FT.SEARCH)
Store non-vector fields with mmh3 hash + protobuf (existing behavior preserved)
Automatic vector index creation (FT.CREATE) on first write when vector fields are present
Support for FLAT and HNSW index algorithms with configurable parameters
Support for Float32 and Float64 vector types
Vector dimension validation against field.vector_length

Configuration (EGValkeyOnlineStoreConfig):

vector_index_algorithm: FLAT or HNSW (default: HNSW)
vector_index_hnsw_m: Max outgoing edges per node (default: 16)
vector_index_hnsw_ef_construction: Index build quality (default: 200)
vector_index_hnsw_ef_runtime: Search quality (default: 10)

Design Decisions

Vector support for regular FeatureViews only (not SortedFeatureView, consistent with Milvus/ES)
Single vector per FeatureView (matches current Feast API constraint)
Backward compatible: non-vector feature views work unchanged

…batch functionality

vanitabhagwat · 2026-03-31T19:12:26Z

sdk/python/feast/infra/online_stores/eg_valkey.py

+        feast_dtype: Feast data type (e.g., Array(Float32))
+
+    Returns:
+        Valkey vector type string: "FLOAT32" or "FLOAT64"


Is float64 a supported vector type in valkey?

yes it is supported

This document says, only float32 is supported. A'm I missing something?
https://valkey.io/topics/search-data-formats/

You're right, I looked at data formats not relating to Search by mistake. Thanks!!

vanitabhagwat · 2026-03-31T20:39:18Z

sdk/python/feast/infra/online_stores/eg_valkey.py

+        index_name = _get_vector_index_name(config.project, table.name)
+
+        # Check if index exists
+        try:


If two processes call online_write_batch concurrently for the same feature view, both could see the index as non-existent and both attempt FT.CREATE, with one failing.

Will catch the exception like the other DBs

vanitabhagwat · 2026-03-31T20:58:31Z

sdk/python/feast/infra/online_stores/eg_valkey.py

+
+            if feature_name in vector_field_names:
+                # Vector field: deserialize from raw bytes
+                field = next(f for f in feature_view.features if f.name == feature_name)


This is O(n) field lookup per vector feature per entity. Can we try a dict lookup?

vanitabhagwat · 2026-03-31T21:15:01Z

sdk/python/feast/infra/online_stores/eg_valkey.py

+            schema_fields.append(VectorField(field_name, algorithm, attributes))
+
+        # Define index on HASH keys with specific prefix
+        key_prefix = _redis_key_prefix(table.join_keys)


_redis_key_prefix(table.join_keys) produces a prefix based only on the join key names (e.g., ["item_id"]), without including the project. But _redis_key() builds the actual HASH keys as
<serialized_entity_key>, with the project appended as a suffix.

This means the IndexDefinition prefix will match HASH keys from all projects that share the same join key names. If two projects (e.g., prod and staging) both have a feature view with join_keys=["item_id"], the vector index created for prod will also index staging's keys, and vice versa.
Im not sure how the solution would look like for this. Maybe use a filter expression or something when building the search to scope the results to correct project?

Yes, I added a TODO comment for now but will address this while implementing the Search part.

vanitabhagwat

I think we need to clean up the vector indexes somewhere. Maybe in delete_table and call it in teardown like we do for hashes.

vanitabhagwat · 2026-04-02T19:22:25Z

sdk/python/feast/infra/online_stores/eg_valkey.py

                if len(_tables) == 1:
                    pipe.delete(_k)
                else:
                    pipe.hdel(_k, *valkey_hash_keys)


Since we are using the original field names for vector fields, *pipe.hdel(_k, valkey_hash_keys) wont delete the vector fields. I
n case of vector fields, you need to have the original vector names added to valkey_hash_keys.

Thanks, made the fix!

piket · 2026-04-02T20:13:38Z

sdk/python/feast/infra/online_stores/eg_valkey.py

+def _get_vector_index_name(project: str, feature_view_name: str) -> str:
+    """Generate Valkey Search index name for vector fields."""
+    return f"{project}_{feature_view_name}_vidx"
+


Can we add the feature name into the index name? Looking ahead at multi-vector we are going to run into issues with just project + fv name.

piket

LGTM

* Adding support for Valkey Search, adding changes to the online_write_batch functionality * Addressing PR comments * addressing linting error * fix tests * addressing PR comments * addressing PR comments * fixing linting --------- Co-authored-by: Manisha4 <Manisha4@github.com>

* feat: Valkey Online Write Batch Vector Search Support (#351) * Adding support for Valkey Search, adding changes to the online_write_batch functionality * Addressing PR comments * addressing linting error * fix tests * addressing PR comments * addressing PR comments * fixing linting --------- Co-authored-by: Manisha4 <Manisha4@github.com> * feat: Support Vector Search in Valkey (#354) * Adding support for Valkey Search, adding changes to the online_write_batch functionality * Addressing PR comments * addressing linting error * Adding changes to support search in valkey * fix tests * adding unit tests * reformatting files and adding checks and more tests * reformatting files and adding checks and more tests * reformatting files and adding checks and more tests * Fix linter errors: type annotations and code formatting - Add explicit type annotation for schema_fields to support both TagField and VectorField - Encode project string to bytes for consistency with other hash values - Decode doc_key bytes to string for hmget compatibility - Fix code formatting: break long lines and remove extra blank lines - Remove tests for multiple vector fields (Feast enforces one vector per feature view) - Fix config type: use 'eg-valkey' (hyphen) not 'eg_valkey' (underscore) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * addressing PR comments * addressing PR comments * fixing linting * Fix missing feature_name argument in retrieve_online_documents_v2 Add the third argument (vector_field.name) to _get_vector_index_name call to match the updated function signature. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * addressing comments, PR changes for some fixes and merge conflicts * fixing tests * fixing tests * fixing linting * fixing linting --------- Co-authored-by: Manisha4 <Manisha4@github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix: Valkey vector search - remove unsupported SORTBY (#356) * fix: Valkey vector search - remove unsupported SORTBY and fix tag filter syntax Valkey Search KNN queries return results pre-sorted by distance, so explicit SORTBY is not supported and causes a ResponseError. This removes the .sort_by() call from the query builder. Additionally, fixes the project tag filter to use unquoted syntax with backslash escaping for special characters (e.g. hyphens, dots) instead of the quoted syntax which was returning empty results. Updates unit tests to reflect both changes: replaces three metric-specific sort order tests with a single test asserting no SORTBY is set, and updates escaping assertions to match the new backslash-escape approach. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style: apply ruff format to eg_valkey.py and test_valkey.py Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Manisha4 <Manisha4@github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Manisha4 <Manisha4@github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Adding support for Valkey Search, adding changes to the online_write_…

4f04a75

…batch functionality

vanitabhagwat reviewed Mar 31, 2026

View reviewed changes

Manisha4 added 3 commits April 1, 2026 23:45

Addressing PR comments

c6512cc

addressing linting error

59c7753

fix tests

0b7ad6c

vanitabhagwat reviewed Apr 2, 2026

View reviewed changes

piket reviewed Apr 2, 2026

View reviewed changes

Manisha4 added 3 commits April 2, 2026 22:28

addressing PR comments

ad3987a

addressing PR comments

24df6ab

fixing linting

ba1b6d7

piket approved these changes Apr 3, 2026

View reviewed changes

Manisha4 merged commit bd19925 into feature/vector-store Apr 6, 2026
28 of 29 checks passed

Manisha4 deleted the add/valkey-search-support-writes branch April 6, 2026 23:34

Conversation

Manisha4 commented Mar 31, 2026

What this PR does / why we need it:

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vanitabhagwat left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

piket left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants