[Feat] Implement keyword search in Qdrant

This commit implements keyword search in Qdrant.

Signed-off-by: Varsha Prasad Narsing <[email protected]>

Author: varshaprasad96

llama_stack/providers/remote/vector_io/qdrant/qdrant.py

@@ -128,13 +128,35 @@ async def query_vector(self, embedding: NDArray, k: int, score_threshold: float)
 
         return QueryChunksResponse(chunks=chunks, scores=scores)
 
-    async def query_keyword(
-        self,
-        query_string: str,
-        k: int,
-        score_threshold: float,
-    ) -> QueryChunksResponse:
-        raise NotImplementedError("Keyword search is not supported in Qdrant")
+    async def query_keyword(self, query_string: str, k: int, score_threshold: float) -> QueryChunksResponse:
+        results = (
+            await self.client.query_points(
+                collection_name=self.collection_name,
+                query_filter=models.Filter(
+                    must=[models.FieldCondition(key="chunk_content.content", match=models.MatchText(text=query_string))]
+                ),
+                limit=k,
+                with_payload=True,
+                with_vectors=False,
+                score_threshold=score_threshold,
+            )
+        ).points
+
+        chunks, scores = [], []
+        for point in results:
+            assert isinstance(point, models.ScoredPoint)
+            assert point.payload is not None
+
+            try:
+                chunk = Chunk(**point.payload["chunk_content"])
+            except Exception:
+                log.exception("Failed to parse chunk")
+                continue
+
+            chunks.append(chunk)
+            scores.append(point.score)
+
+        return QueryChunksResponse(chunks=chunks, scores=scores)
 
     async def query_hybrid(
         self,

tests/integration/vector_io/test_openai_vector_stores.py

@@ -55,6 +55,8 @@ def skip_if_provider_doesnt_support_openai_vector_stores_search(client_with_mode
         ],
         "keyword": [
             "inline::sqlite-vec",
+            "inline::qdrant",
+            "remote::qdrant",
             "remote::milvus",
         ],
         "hybrid": [

tests/unit/providers/vector_io/conftest.py

@@ -4,7 +4,9 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import os
 import random
+import tempfile
 
 import numpy as np
 import pytest
@@ -17,7 +19,7 @@
 from llama_stack.providers.inline.vector_io.faiss.config import FaissVectorIOConfig
 from llama_stack.providers.inline.vector_io.faiss.faiss import FaissIndex, FaissVectorIOAdapter
 from llama_stack.providers.inline.vector_io.milvus.config import MilvusVectorIOConfig, SqliteKVStoreConfig
-from llama_stack.providers.inline.vector_io.qdrant import QdrantVectorIOConfig
+from llama_stack.providers.inline.vector_io.qdrant import QdrantVectorIOConfig as InlineQdrantVectorIOConfig
 from llama_stack.providers.inline.vector_io.sqlite_vec import SQLiteVectorIOConfig
 from llama_stack.providers.inline.vector_io.sqlite_vec.sqlite_vec import SQLiteVecIndex, SQLiteVecVectorIOAdapter
 from llama_stack.providers.remote.vector_io.chroma.chroma import ChromaIndex, ChromaVectorIOAdapter, maybe_await
@@ -29,7 +31,7 @@
 MILVUS_ALIAS = "test_milvus"
 
 
-@pytest.fixture(params=["milvus", "sqlite_vec", "faiss", "chroma"])
+@pytest.fixture(params=["milvus", "sqlite_vec", "faiss", "chroma", "qdrant"])
 def vector_provider(request):
     return request.param
 
@@ -283,40 +285,53 @@ async def chroma_vec_adapter(chroma_vec_db_path, mock_inference_api, embedding_d
 
 
 @pytest.fixture
-def qdrant_vec_db_path(tmp_path_factory):
+def qdrant_vec_db_path(tmp_path):
+    """Use tmp_path with additional isolation to ensure unique path per test."""
     import uuid
 
-    db_path = str(tmp_path_factory.getbasetemp() / f"test_qdrant_{uuid.uuid4()}.db")
-    return db_path
+    # Create a completely isolated temporary directory
+    temp_dir = tempfile.mkdtemp(prefix=f"qdrant_test_{uuid.uuid4()}_")
+    return temp_dir
 
 
 @pytest.fixture
 async def qdrant_vec_adapter(qdrant_vec_db_path, mock_inference_api, embedding_dimension):
+    import shutil
     import uuid
 
-    config = QdrantVectorIOConfig(
-        db_path=qdrant_vec_db_path,
+    config = InlineQdrantVectorIOConfig(
+        path=qdrant_vec_db_path,
         kvstore=SqliteKVStoreConfig(),
     )
     adapter = QdrantVectorIOAdapter(
         config=config,
         inference_api=mock_inference_api,
         files_api=None,
     )
-    collection_id = f"qdrant_test_collection_{uuid.uuid4()}"
+
+    original_initialize = adapter.initialize
+
+    async def safe_initialize():
+        if not hasattr(adapter, "_initialized") or not adapter._initialized:
+            await original_initialize()
+            adapter._initialized = True
+
+    adapter.initialize = safe_initialize
     await adapter.initialize()
-    await adapter.register_vector_db(
-        VectorDB(
-            identifier=collection_id,
-            provider_id="test_provider",
-            embedding_model="test_model",
-            embedding_dimension=embedding_dimension,
-        )
-    )
+
+    collection_id = f"qdrant_test_collection_{uuid.uuid4()}"
     adapter.test_collection_id = collection_id
+    adapter._test_db_path = qdrant_vec_db_path
     yield adapter
+
     await adapter.shutdown()
 
+    try:
+        if os.path.exists(qdrant_vec_db_path):
+            shutil.rmtree(qdrant_vec_db_path, ignore_errors=True)
+    except Exception:
+        pass
+
 
 @pytest.fixture
 async def qdrant_vec_index(qdrant_vec_db_path, embedding_dimension):

tests/unit/providers/vector_io/test_qdrant.py

@@ -136,3 +136,89 @@ async def test_qdrant_register_and_unregister_vector_db(
     await qdrant_adapter.unregister_vector_db(vector_db_id)
     assert not (await qdrant_adapter.client.collection_exists(vector_db_id))
     assert len((await qdrant_adapter.client.get_collections()).collections) == 0
+
+
+# Keyword search tests
+async def test_query_chunks_keyword_search(qdrant_vec_index, sample_chunks, sample_embeddings):
+    """Test keyword search functionality in Qdrant."""
+    await qdrant_vec_index.add_chunks(sample_chunks, sample_embeddings)
+    query_string = "Sentence 5"
+    response = await qdrant_vec_index.query_keyword(query_string=query_string, k=3, score_threshold=0.0)
+
+    assert isinstance(response, QueryChunksResponse)
+    assert len(response.chunks) > 0, f"Expected some chunks, but got {len(response.chunks)}"
+
+    non_existent_query_str = "blablabla"
+    response_no_results = await qdrant_vec_index.query_keyword(
+        query_string=non_existent_query_str, k=1, score_threshold=0.0
+    )
+
+    assert isinstance(response_no_results, QueryChunksResponse)
+    assert len(response_no_results.chunks) == 0, f"Expected 0 results, but got {len(response_no_results.chunks)}"
+
+
+async def test_query_chunks_keyword_search_k_greater_than_results(qdrant_vec_index, sample_chunks, sample_embeddings):
+    """Test keyword search when k is greater than available results."""
+    await qdrant_vec_index.add_chunks(sample_chunks, sample_embeddings)
+
+    query_str = "Sentence 1 from document 0"  # Should match only one chunk
+    response = await qdrant_vec_index.query_keyword(k=5, score_threshold=0.0, query_string=query_str)
+
+    assert isinstance(response, QueryChunksResponse)
+    assert 0 < len(response.chunks) < 5, f"Expected results between [1, 4], got {len(response.chunks)}"
+    assert any("Sentence 1 from document 0" in chunk.content for chunk in response.chunks), "Expected chunk not found"
+
+
+async def test_query_chunks_keyword_search_score_threshold(qdrant_vec_index, sample_chunks, sample_embeddings):
+    """Test keyword search with score threshold filtering."""
+    await qdrant_vec_index.add_chunks(sample_chunks, sample_embeddings)
+
+    query_string = "Sentence 5"
+
+    # Test with low threshold (should return results)
+    response_low_threshold = await qdrant_vec_index.query_keyword(query_string=query_string, k=3, score_threshold=0.0)
+    assert len(response_low_threshold.chunks) > 0
+
+    # Test with negative threshold (should return results since scores are 0.0)
+    response_negative_threshold = await qdrant_vec_index.query_keyword(
+        query_string=query_string, k=3, score_threshold=-1.0
+    )
+    assert len(response_negative_threshold.chunks) > 0
+
+
+async def test_query_chunks_keyword_search_edge_cases(qdrant_vec_index, sample_chunks, sample_embeddings):
+    """Test keyword search edge cases."""
+    await qdrant_vec_index.add_chunks(sample_chunks, sample_embeddings)
+
+    # Test with empty string
+    response_empty = await qdrant_vec_index.query_keyword(query_string="", k=3, score_threshold=0.0)
+    assert isinstance(response_empty, QueryChunksResponse)
+
+    # Test with very long query string
+    long_query = "a" * 1000
+    response_long = await qdrant_vec_index.query_keyword(query_string=long_query, k=3, score_threshold=0.0)
+    assert isinstance(response_long, QueryChunksResponse)
+
+    # Test with special characters
+    special_query = "!@#$%^&*()_+-=[]{}|;':\",./<>?"
+    response_special = await qdrant_vec_index.query_keyword(query_string=special_query, k=3, score_threshold=0.0)
+    assert isinstance(response_special, QueryChunksResponse)
+
+
+async def test_query_chunks_keyword_search_metadata_preservation(
+    qdrant_vec_index, sample_chunks_with_metadata, sample_embeddings_with_metadata
+):
+    """Test that keyword search preserves chunk metadata."""
+    await qdrant_vec_index.add_chunks(sample_chunks_with_metadata, sample_embeddings_with_metadata)
+
+    query_string = "Sentence 0"
+    response = await qdrant_vec_index.query_keyword(query_string=query_string, k=2, score_threshold=0.0)
+
+    assert len(response.chunks) > 0
+    for chunk in response.chunks:
+        # Check that metadata is preserved
+        assert hasattr(chunk, "metadata") or hasattr(chunk, "chunk_metadata")
+        if hasattr(chunk, "chunk_metadata") and chunk.chunk_metadata:
+            assert chunk.chunk_metadata.document_id is not None
+            assert chunk.chunk_metadata.chunk_id is not None
+            assert chunk.chunk_metadata.source is not None