import config from db.dense_embeddings import DenseEmbeddings from langchain_qdrant import QdrantVectorStore, FastEmbedSparse, RetrievalMode from qdrant_client import QdrantClient from qdrant_client.http import models as qmodels class VectorDbManager: __client: QdrantClient __dense_embeddings: DenseEmbeddings __sparse_embeddings: FastEmbedSparse def __init__(self): self.__client = QdrantClient(path=config.QDRANT_DB_PATH) self.__dense_embeddings = DenseEmbeddings(model_name=config.DENSE_MODEL) self.__sparse_embeddings = FastEmbedSparse(model_name=config.SPARSE_MODEL) def _dense_vector_size(self): return self.__dense_embeddings.vector_size @staticmethod def _collection_vector_size(collection_info): vectors_config = collection_info.config.params.vectors if hasattr(vectors_config, "size"): return vectors_config.size if isinstance(vectors_config, dict) and vectors_config: first_vector = next(iter(vectors_config.values())) return getattr(first_vector, "size", None) return None def create_collection(self, collection_name): expected_size = self._dense_vector_size() if not self.__client.collection_exists(collection_name): print(f"Creating collection: {collection_name}...") self.__client.create_collection( collection_name=collection_name, vectors_config=qmodels.VectorParams(size=expected_size, distance=qmodels.Distance.COSINE), sparse_vectors_config={config.SPARSE_VECTOR_NAME: qmodels.SparseVectorParams()}, ) print(f"✓ Collection created: {collection_name}") else: collection_info = self.__client.get_collection(collection_name) vectors_config = collection_info.config.params.vectors dense_config = ( vectors_config.get("") if isinstance(vectors_config, dict) else vectors_config ) if dense_config is None: raise ValueError(f"Qdrant collection '{collection_name}' must have an unnamed dense vector.") if dense_config.distance != qmodels.Distance.COSINE: raise ValueError(f"Qdrant collection '{collection_name}' must use cosine distance.") sparse_config = collection_info.config.params.sparse_vectors or {} if config.SPARSE_VECTOR_NAME not in sparse_config: raise ValueError( f"Qdrant collection '{collection_name}' is missing sparse vector '{config.SPARSE_VECTOR_NAME}'." ) existing_size = self._collection_vector_size(collection_info) if existing_size and existing_size != expected_size: raise ValueError( f"Qdrant collection '{collection_name}' has dense vector size " f"{existing_size}, but '{config.DENSE_MODEL}' produces size " f"{expected_size}. Clear and re-index the collection after " "changing embedding models." ) print(f"✓ Collection already exists: {collection_name}") def delete_collection(self, collection_name): try: if self.__client.collection_exists(collection_name): print(f"Removing existing Qdrant collection: {collection_name}") self.__client.delete_collection(collection_name) except Exception as e: raise RuntimeError(f"Unable to delete Qdrant collection '{collection_name}'.") from e def get_collection(self, collection_name) -> QdrantVectorStore: try: return QdrantVectorStore( client=self.__client, collection_name=collection_name, embedding=self.__dense_embeddings, sparse_embedding=self.__sparse_embeddings, retrieval_mode=RetrievalMode.HYBRID, sparse_vector_name=config.SPARSE_VECTOR_NAME, # Schema is checked above without requesting GPU inference at startup. validate_collection_config=False, ) except Exception as e: raise RuntimeError(f"Unable to initialize Qdrant collection '{collection_name}'.") from e