feat: Support Vector Search in Valkey#354
Merged
Manisha4 merged 22 commits intofeature/vector-storefrom Apr 7, 2026
Merged
Conversation
…batch functionality
- 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>
…h-support-reads Resolves merge conflicts and incorporates Option A implementation: - One index per vector field with feature name in index name - Float64 to Float32 conversion (Valkey limitation) - Vector fields use original name for hset keys Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
| """ | ||
| # Build KNN query with project filter | ||
| # Format: "(@__project__:{project})=>[KNN {top_k} @{field} $vec AS distance]" | ||
| query_str = ( |
Collaborator
There was a problem hiding this comment.
Do we allow hyphens in project name? If yes, then Iyou would have to escape it. Redisearch interprets it as negation.
| search_results = [] | ||
| for doc in results.docs: | ||
| doc_key = doc.id.encode() if isinstance(doc.id, str) else doc.id | ||
| distance = float(getattr(doc, "__distance__", 0.0)) |
Collaborator
There was a problem hiding this comment.
Distance of 0.0 means its a perfect match right? I think we should use something else as default.
| query = ( | ||
| Query(query_str) | ||
| .return_fields("__distance__") | ||
| .sort_by("__distance__") |
Collaborator
There was a problem hiding this comment.
sort_by("distance") would be ascending or descending?
Collaborator
Author
There was a problem hiding this comment.
sort_by() defaults to ascending order, I'll make it explicit
Collaborator
Author
There was a problem hiding this comment.
I changed it a bit to infer sort order from distance metric
| table: FeatureView, | ||
| requested_features: List[str], | ||
| search_results: List[Tuple[bytes, float]], | ||
| vector_field: Field, |
Collaborator
There was a problem hiding this comment.
Where is this being used?
| embedding: Query embedding vector | ||
| top_k: Number of results to return | ||
| distance_metric: Optional override for distance metric (COSINE, L2, IP) | ||
| query_string: Not supported in V1 (reserved for future BM25 search) |
Collaborator
There was a problem hiding this comment.
Suggested change
| query_string: Not supported in V1 (reserved for future BM25 search) | |
| query_string: Not supported in V2(reserved for future BM25 search) |
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>
| entity_hset = dict() | ||
| entity_hset[ts_key] = ts.SerializeToString() | ||
| # Store project and entity key for vector search | ||
| entity_hset["__project__"] = project.encode() |
Collaborator
There was a problem hiding this comment.
This is producing bytes but line 1138 expects a string. Can you test it out?
Base automatically changed from
add/valkey-search-support-writes
to
feature/vector-store
April 6, 2026 23:34
added 6 commits
April 6, 2026 16:39
Resolved conflicts: - eg_valkey.py: Keep both deserialize_entity_key import (for reads) and Query import (for FT.SEARCH) - test_valkey.py: Keep read/search test classes from HEAD
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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:
Adds vector similarity search support to the EG Valkey online store, enabling semantic search use cases for ML features.
Which issue(s) this PR fixes:
Misc