All Submissions:

• Contributions should target the dev branch. Did you create your branch from dev?
• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same update/change?

New Feature Submissions:

1. Does your submission pass tests?
2. Have you formatted your code locally using cargo +nightly fmt --all command prior to submission?
3. Have you checked your code using cargo clippy --workspace --all-features command?

Changes to Core Features:

• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your core changes, as applicable?
• Have you successfully ran tests with your changes locally?

Description

This PR adds an optional enable_hnsw parameter to all payload field index types, providing fine-grained control over HNSW graph construction for individual indexed fields.

Changes

Core Data Types

• Added enable_hnsw: Option<bool> field to all 8 payload index parameter types:
  • KeywordIndexParams
  • IntegerIndexParams
  • FloatIndexParams
  • GeoIndexParams
  • TextIndexParams
  • BoolIndexParams
  • DatetimeIndexParams
  • UuidIndexParams

API Layer

• gRPC: Updated proto definitions in collections.proto with documentation
• gRPC Conversions: Updated all From/TryFrom implementations in conversions.rs to handle new field
• OpenAPI: Added field definition to all index param schemas in openapi.json
• Generated Code: Regenerated qdrant.rs with new field definitions

Core Logic

• Modified HNSW index builder in hnsw.rs to filter fields based on enable_hnsw flag
• Added helper methods enable_hnsw() to PayloadSchemaParams and PayloadFieldSchema types
• Fields with enable_hnsw=false are skipped during additional HNSW links construction

Tests

• Updated all existing tests to include enable_hnsw: None for backward compatibility
• Modified tests in:
  • payload_index_test.rs
  • segment_on_disk_snapshot.rs
  • mutable_text_index.rs
  • full_text_index/tests/mod.rs
  • tokenizers/mod.rs
  • test_payload_indexing.py

Motivation

When payload_m > 0 is configured in HNSW settings, Qdrant builds additional graph links for all indexed payload fields to improve filtered search performance.

However, when a collection contains both dense and sparse vectors, payload indexes intended only for sparse vector filtering will still affect the dense vector's HNSW graph. This creates additional graph links in the dense vector's HNSW index that will never be used in queries. This causes the following problems:

1. Increased indexing time: Building HNSW links for every indexed field can significantly slow down indexing
2. Higher memory usage: Additional links consume memory
3. Not always beneficial: Some fields may not benefit from HNSW links depending on query patterns

This feature allows users to:

• Selectively disable HNSW graph building for specific fields that don't need it
• Reduce resource consumption in large-scale deployments
• Fine-tune performance based on actual query patterns
• Index fields for filtering without the overhead of HNSW link construction

Behavior

• Default value: true (maintains full backward compatibility)
• When enable_hnsw=false: The field will still be indexed normally for filtering, but won't have additional HNSW links built
• Requirement: Only takes effect when the collection's HNSW config has payload_m > 0
• Log output: When disabled, logs: enable_hnsw=false. Skip building additional index for field {field_name}

Example Usage

REST API

PUT /collections/{collection_name}/index
{
  "field_name": "category",
  "field_schema": {
    "type": "keyword",
    "enable_hnsw": false
  }
}

gRPC

KeywordIndexParams {
  is_tenant: false,
  on_disk: false,
  enable_hnsw: false  // Disable HNSW for this field
}

Backward Compatibility

✅ Fully backward compatible

• Default value is true, preserving existing behavior
• Existing collections and indices work without modification
• Optional field in all APIs (gRPC, REST, OpenAPI)