qdrant-haystack 8.1.0__py3-none-any.whl → 9.1.0__py3-none-any.whl

haystack_integrations/document_stores/qdrant/document_store.py

@@ -1,11 +1,10 @@
 import inspect
-import logging
 from itertools import islice
-from typing import Any, ClassVar, Dict, Generator, List, Optional, Set, Union
+from typing import Any, AsyncGenerator, ClassVar, Dict, Generator, List, Optional, Set, Union
 
 import numpy as np
 import qdrant_client
-from haystack import default_from_dict, default_to_dict
+from haystack import default_from_dict, default_to_dict, logging
 from haystack.dataclasses import Document
 from haystack.dataclasses.sparse_embedding import SparseEmbedding
 from haystack.document_stores.errors import DocumentStoreError, DuplicateDocumentError
@@ -217,6 +216,7 @@ class QdrantDocumentStore:
         """
 
         self._client = None
+        self._async_client = None
 
         # Store the Qdrant client specific attributes
         self.location = location
@@ -258,24 +258,10 @@ class QdrantDocumentStore:
         self.write_batch_size = write_batch_size
         self.scroll_size = scroll_size
 
-    @property
-    def client(self):
-        if not self._client:
-            self._client = qdrant_client.QdrantClient(
-                location=self.location,
-                url=self.url,
-                port=self.port,
-                grpc_port=self.grpc_port,
-                prefer_grpc=self.prefer_grpc,
-                https=self.https,
-                api_key=self.api_key.resolve_value() if self.api_key else None,
-                prefix=self.prefix,
-                timeout=self.timeout,
-                host=self.host,
-                path=self.path,
-                metadata=self.metadata,
-                force_disable_check_same_thread=self.force_disable_check_same_thread,
-            )
+    def _initialize_client(self):
+        if self._client is None:
+            client_params = self._prepare_client_params()
+            self._client = qdrant_client.QdrantClient(**client_params)
             # Make sure the collection is properly set up
             self._set_up_collection(
                 self.index,
@@ -287,14 +273,52 @@ class QdrantDocumentStore:
                 self.on_disk,
                 self.payload_fields_to_index,
             )
-        return self._client
+
+    async def _initialize_async_client(self):
+        """
+        Returns the asynchronous Qdrant client, initializing it if necessary.
+        """
+        if self._async_client is None:
+            client_params = self._prepare_client_params()
+            self._async_client = qdrant_client.AsyncQdrantClient(
+                **client_params,
+            )
+            await self._set_up_collection_async(
+                self.index,
+                self.embedding_dim,
+                self.recreate_index,
+                self.similarity,
+                self.use_sparse_embeddings,
+                self.sparse_idf,
+                self.on_disk,
+                self.payload_fields_to_index,
+            )
 
     def count_documents(self) -> int:
         """
         Returns the number of documents present in the Document Store.
         """
+        self._initialize_client()
+        assert self._client is not None
+        try:
+            response = self._client.count(
+                collection_name=self.index,
+            )
+            return response.count
+        except (UnexpectedResponse, ValueError):
+            # Qdrant local raises ValueError if the collection is not found, but
+            # with the remote server UnexpectedResponse is raised. Until that's unified,
+            # we need to catch both.
+            return 0
+
+    async def count_documents_async(self) -> int:
+        """
+        Asynchronously returns the number of documents present in the document dtore.
+        """
+        await self._initialize_async_client()
+        assert self._async_client is not None
         try:
-            response = self.client.count(
+            response = await self._async_client.count(
                 collection_name=self.index,
             )
             return response.count
@@ -317,19 +341,29 @@ class QdrantDocumentStore:
         :param filters: The filters to apply to the document list.
         :returns: A list of documents that match the given filters.
         """
-        if filters and not isinstance(filters, dict) and not isinstance(filters, rest.Filter):
-            msg = "Filter must be a dictionary or an instance of `qdrant_client.http.models.Filter`"
-            raise ValueError(msg)
+        # No need to initialize client here as _get_documents_generator
+        # will handle client initialization internally
 
-        if filters and not isinstance(filters, rest.Filter) and "operator" not in filters:
-            msg = "Invalid filter syntax. See https://docs.haystack.deepset.ai/docs/metadata-filtering for details."
-            raise ValueError(msg)
+        self._validate_filters(filters)
         return list(
-            self.get_documents_generator(
+            self._get_documents_generator(
                 filters,
             )
         )
 
+    async def filter_documents_async(
+        self,
+        filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
+    ) -> List[Document]:
+        """
+        Asynchronously returns the documents that match the provided filters.
+        """
+        # No need to initialize client here as _get_documents_generator_async
+        # will handle client initialization internally
+
+        self._validate_filters(filters)
+        return [doc async for doc in self._get_documents_generator_async(filters)]
+
     def write_documents(
         self,
         documents: List[Document],
@@ -348,13 +382,14 @@ class QdrantDocumentStore:
 
         :returns: The number of documents written to the document store.
         """
+
+        self._initialize_client()
+        assert self._client is not None
+
         for doc in documents:
             if not isinstance(doc, Document):
                 msg = f"DocumentStore.write_documents() expects a list of Documents but got an element of {type(doc)}."
                 raise ValueError(msg)
-        self._set_up_collection(
-            self.index, self.embedding_dim, False, self.similarity, self.use_sparse_embeddings, self.sparse_idf
-        )
 
         if len(documents) == 0:
             logger.warning("Calling QdrantDocumentStore.write_documents() with empty list")
@@ -373,7 +408,61 @@ class QdrantDocumentStore:
                     use_sparse_embeddings=self.use_sparse_embeddings,
                 )
 
-                self.client.upsert(
+                self._client.upsert(
+                    collection_name=self.index,
+                    points=batch,
+                    wait=self.wait_result_from_api,
+                )
+
+                progress_bar.update(self.write_batch_size)
+        return len(document_objects)
+
+    async def write_documents_async(
+        self,
+        documents: List[Document],
+        policy: DuplicatePolicy = DuplicatePolicy.FAIL,
+    ) -> int:
+        """
+        Asynchronously writes documents to Qdrant using the specified policy.
+        The QdrantDocumentStore can handle duplicate documents based on the given policy.
+        The available policies are:
+        - `FAIL`: The operation will raise an error if any document already exists.
+        - `OVERWRITE`: Existing documents will be overwritten with the new ones.
+        - `SKIP`: Existing documents will be skipped, and only new documents will be added.
+
+        :param documents: A list of Document objects to write to Qdrant.
+        :param policy: The policy for handling duplicate documents.
+
+        :returns: The number of documents written to the document store.
+        """
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
+        for doc in documents:
+            if not isinstance(doc, Document):
+                msg = f"""DocumentStore.write_documents_async() expects a list of
+                Documents but got an element of {type(doc)}."""
+                raise ValueError(msg)
+
+        if len(documents) == 0:
+            logger.warning("Calling QdrantDocumentStore.write_documents_async() with empty list")
+            return 0
+
+        document_objects = await self._handle_duplicate_documents_async(
+            documents=documents,
+            policy=policy,
+        )
+
+        batched_documents = get_batches_from_generator(document_objects, self.write_batch_size)
+        with tqdm(total=len(document_objects), disable=not self.progress_bar) as progress_bar:
+            for document_batch in batched_documents:
+                batch = convert_haystack_documents_to_qdrant_points(
+                    document_batch,
+                    use_sparse_embeddings=self.use_sparse_embeddings,
+                )
+
+                await self._async_client.upsert(
                     collection_name=self.index,
                     points=batch,
                     wait=self.wait_result_from_api,
@@ -388,9 +477,13 @@ class QdrantDocumentStore:
 
         :param document_ids: the document ids to delete
         """
+
+        self._initialize_client()
+        assert self._client is not None
+
         ids = [convert_id(_id) for _id in document_ids]
         try:
-            self.client.delete(
+            self._client.delete(
                 collection_name=self.index,
                 points_selector=ids,
                 wait=self.wait_result_from_api,
@@ -400,6 +493,28 @@ class QdrantDocumentStore:
                 "Called QdrantDocumentStore.delete_documents() on a non-existing ID",
             )
 
+    async def delete_documents_async(self, document_ids: List[str]) -> None:
+        """
+        Asynchronously deletes documents that match the provided `document_ids` from the document store.
+
+        :param document_ids: the document ids to delete
+        """
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
+        ids = [convert_id(_id) for _id in document_ids]
+        try:
+            await self._async_client.delete(
+                collection_name=self.index,
+                points_selector=ids,
+                wait=self.wait_result_from_api,
+            )
+        except KeyError:
+            logger.warning(
+                "Called QdrantDocumentStore.delete_documents_async() on a non-existing ID",
+            )
+
     @classmethod
     def from_dict(cls, data: Dict[str, Any]) -> "QdrantDocumentStore":
         """
@@ -430,7 +545,7 @@ class QdrantDocumentStore:
             **init_params,
         )
 
-    def get_documents_generator(
+    def _get_documents_generator(
         self,
         filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
     ) -> Generator[Document, None, None]:
@@ -441,13 +556,53 @@ class QdrantDocumentStore:
         :returns: A generator that yields documents retrieved from Qdrant.
         """
 
+        self._initialize_client()
+        assert self._client is not None
+
+        index = self.index
+        qdrant_filters = convert_filters_to_qdrant(filters)
+
+        next_offset = None
+        stop_scrolling = False
+        while not stop_scrolling:
+            records, next_offset = self._client.scroll(
+                collection_name=index,
+                scroll_filter=qdrant_filters,
+                limit=self.scroll_size,
+                offset=next_offset,
+                with_payload=True,
+                with_vectors=True,
+            )
+            stop_scrolling = next_offset is None or (
+                isinstance(next_offset, grpc.PointId) and next_offset.num == 0 and next_offset.uuid == ""
+            )
+
+            for record in records:
+                yield convert_qdrant_point_to_haystack_document(
+                    record, use_sparse_embeddings=self.use_sparse_embeddings
+                )
+
+    async def _get_documents_generator_async(
+        self,
+        filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
+    ) -> AsyncGenerator[Document, None]:
+        """
+        Returns an asynchronous generator that yields documents from Qdrant based on the provided filters.
+
+        :param filters: Filters applied to the retrieved documents.
+        :returns: An asynchronous generator that yields documents retrieved from Qdrant.
+        """
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
         index = self.index
         qdrant_filters = convert_filters_to_qdrant(filters)
 
         next_offset = None
         stop_scrolling = False
         while not stop_scrolling:
-            records, next_offset = self.client.scroll(
+            records, next_offset = await self._async_client.scroll(
                 collection_name=index,
                 scroll_filter=qdrant_filters,
                 limit=self.scroll_size,
@@ -480,8 +635,44 @@ class QdrantDocumentStore:
         """
         documents: List[Document] = []
 
+        self._initialize_client()
+        assert self._client is not None
+
+        ids = [convert_id(_id) for _id in ids]
+        records = self._client.retrieve(
+            collection_name=self.index,
+            ids=ids,
+            with_payload=True,
+            with_vectors=True,
+        )
+
+        for record in records:
+            documents.append(
+                convert_qdrant_point_to_haystack_document(record, use_sparse_embeddings=self.use_sparse_embeddings)
+            )
+        return documents
+
+    async def get_documents_by_id_async(
+        self,
+        ids: List[str],
+    ) -> List[Document]:
+        """
+        Retrieves documents from Qdrant by their IDs.
+
+        :param ids:
+            A list of document IDs to retrieve.
+        :param index:
+            The name of the index to retrieve documents from.
+        :returns:
+            A list of documents.
+        """
+        documents: List[Document] = []
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
         ids = [convert_id(_id) for _id in ids]
-        records = self.client.retrieve(
+        records = await self._async_client.retrieve(
             collection_name=self.index,
             ids=ids,
             with_payload=True,
@@ -527,6 +718,8 @@ class QdrantDocumentStore:
         :raises QdrantStoreError:
             If the Document Store was initialized with `use_sparse_embeddings=False`.
         """
+        self._initialize_client()
+        assert self._client is not None
 
         if not self.use_sparse_embeddings:
             message = (
@@ -539,7 +732,7 @@ class QdrantDocumentStore:
         query_indices = query_sparse_embedding.indices
         query_values = query_sparse_embedding.values
         if group_by:
-            groups = self.client.query_points_groups(
+            groups = self._client.query_points_groups(
                 collection_name=self.index,
                 query=rest.SparseVector(
                     indices=query_indices,
@@ -553,17 +746,9 @@ class QdrantDocumentStore:
                 with_vectors=return_embedding,
                 score_threshold=score_threshold,
             ).groups
-            results = (
-                [
-                    convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
-                    for group in groups
-                    for point in group.hits
-                ]
-                if groups
-                else []
-            )
+            return self._process_group_results(groups)
         else:
-            points = self.client.query_points(
+            points = self._client.query_points(
                 collection_name=self.index,
                 query=rest.SparseVector(
                     indices=query_indices,
@@ -575,16 +760,7 @@ class QdrantDocumentStore:
                 with_vectors=return_embedding,
                 score_threshold=score_threshold,
             ).points
-            results = [
-                convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
-                for point in points
-            ]
-        if scale_score:
-            for document in results:
-                score = document.score
-                score = float(1 / (1 + np.exp(-score / 100)))
-                document.score = score
-        return results
+            return self._process_query_point_results(points, scale_score=scale_score)
 
     def _query_by_embedding(
         self,
@@ -616,9 +792,12 @@ class QdrantDocumentStore:
 
         :returns: List of documents that are most similar to `query_embedding`.
         """
+        self._initialize_client()
+        assert self._client is not None
+
         qdrant_filters = convert_filters_to_qdrant(filters)
         if group_by:
-            groups = self.client.query_points_groups(
+            groups = self._client.query_points_groups(
                 collection_name=self.index,
                 query=query_embedding,
                 using=DENSE_VECTORS_NAME if self.use_sparse_embeddings else None,
@@ -629,17 +808,10 @@ class QdrantDocumentStore:
                 with_vectors=return_embedding,
                 score_threshold=score_threshold,
             ).groups
-            results = (
-                [
-                    convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
-                    for group in groups
-                    for point in group.hits
-                ]
-                if groups
-                else []
-            )
+            return self._process_group_results(groups)
+
         else:
-            points = self.client.query_points(
+            points = self._client.query_points(
                 collection_name=self.index,
                 query=query_embedding,
                 using=DENSE_VECTORS_NAME if self.use_sparse_embeddings else None,
@@ -648,20 +820,7 @@ class QdrantDocumentStore:
                 with_vectors=return_embedding,
                 score_threshold=score_threshold,
             ).points
-            results = [
-                convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
-                for point in points
-            ]
-
-        if scale_score:
-            for document in results:
-                score = document.score
-                if self.similarity == "cosine":
-                    score = (score + 1) / 2
-                else:
-                    score = float(1 / (1 + np.exp(-score / 100)))
-                document.score = score
-        return results
+            return self._process_query_point_results(points, scale_score=scale_score)
 
     def _query_hybrid(
         self,
@@ -702,6 +861,10 @@ class QdrantDocumentStore:
 
         # This implementation is based on the code from the Python Qdrant client:
         # https://github.com/qdrant/qdrant-client/blob/8e3ea58f781e4110d11c0a6985b5e6bb66b85d33/qdrant_client/qdrant_fastembed.py#L519
+
+        self._initialize_client()
+        assert self._client is not None
+
         if not self.use_sparse_embeddings:
             message = (
                 "You are trying to query using sparse embeddings, but the Document Store "
@@ -713,7 +876,7 @@ class QdrantDocumentStore:
 
         try:
             if group_by:
-                groups = self.client.query_points_groups(
+                groups = self._client.query_points_groups(
                     collection_name=self.index,
                     prefetch=[
                         rest.Prefetch(
@@ -739,7 +902,7 @@ class QdrantDocumentStore:
                     with_vectors=return_embedding,
                 ).groups
             else:
-                points = self.client.query_points(
+                points = self._client.query_points(
                     collection_name=self.index,
                     prefetch=[
                         rest.Prefetch(
@@ -768,71 +931,339 @@ class QdrantDocumentStore:
             raise QdrantStoreError(msg) from e
 
         if group_by:
-            results = (
-                [
-                    convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
-                    for group in groups
-                    for point in group.hits
-                ]
-                if groups
-                else []
-            )
+            return self._process_group_results(groups)
         else:
-            results = [convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=True) for point in points]
+            return self._process_query_point_results(points)
 
-        return results
-
-    def get_distance(self, similarity: str) -> rest.Distance:
+    async def _query_by_sparse_async(
+        self,
+        query_sparse_embedding: SparseEmbedding,
+        filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
+        top_k: int = 10,
+        scale_score: bool = False,
+        return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
+        group_by: Optional[str] = None,
+        group_size: Optional[int] = None,
+    ) -> List[Document]:
         """
-        Retrieves the distance metric for the specified similarity measure.
+        Asynchronously queries Qdrant using a sparse embedding and returns the most relevant documents.
+
+        :param query_sparse_embedding: Sparse embedding of the query.
+        :param filters: Filters applied to the retrieved documents.
+        :param top_k: Maximum number of documents to return. If using `group_by` parameters, maximum number of
+             groups to return.
+        :param scale_score: Whether to scale the scores of the retrieved documents.
+        :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
+        :param group_by: Payload field to group by, must be a string or number field. If the field contains more than 1
+             value, all values will be used for grouping. One point can be in multiple groups.
+        :param group_size: Maximum amount of points to return per group. Default is 3.
+
+        :returns: List of documents that are most similar to `query_sparse_embedding`.
 
-        :param similarity:
-            The similarity measure to retrieve the distance.
-        :returns:
-            The corresponding rest.Distance object.
         :raises QdrantStoreError:
-            If the provided similarity measure is not supported.
+            If the Document Store was initialized with `use_sparse_embeddings=False`.
         """
-        try:
-            return self.SIMILARITY[similarity]
-        except KeyError as ke:
-            msg = (
-                f"Provided similarity '{similarity}' is not supported by Qdrant "
-                f"document store. Please choose one of the options: "
-                f"{', '.join(self.SIMILARITY.keys())}"
-            )
-            raise QdrantStoreError(msg) from ke
 
-    def _create_payload_index(self, collection_name: str, payload_fields_to_index: Optional[List[dict]] = None):
-        """
-        Create payload index for the collection if payload_fields_to_index is provided
-        See: https://qdrant.tech/documentation/concepts/indexing/#payload-index
-        """
-        if payload_fields_to_index is not None:
-            for payload_index in payload_fields_to_index:
-                self.client.create_payload_index(
-                    collection_name=collection_name,
-                    field_name=payload_index["field_name"],
-                    field_schema=payload_index["field_schema"],
-                )
+        await self._initialize_async_client()
+        assert self._async_client is not None
 
-    def _set_up_collection(
-        self,
-        collection_name: str,
-        embedding_dim: int,
-        recreate_collection: bool,
-        similarity: str,
-        use_sparse_embeddings: bool,
-        sparse_idf: bool,
-        on_disk: bool = False,
-        payload_fields_to_index: Optional[List[dict]] = None,
-    ):
-        """
-        Sets up the Qdrant collection with the specified parameters.
-        :param collection_name:
-            The name of the collection to set up.
-        :param embedding_dim:
-            The dimension of the embeddings.
+        if not self.use_sparse_embeddings:
+            message = (
+                "You are trying to query using sparse embeddings, but the Document Store "
+                "was initialized with `use_sparse_embeddings=False`. "
+            )
+            raise QdrantStoreError(message)
+
+        qdrant_filters = convert_filters_to_qdrant(filters)
+        query_indices = query_sparse_embedding.indices
+        query_values = query_sparse_embedding.values
+        if group_by:
+            response = await self._async_client.query_points_groups(
+                collection_name=self.index,
+                query=rest.SparseVector(
+                    indices=query_indices,
+                    values=query_values,
+                ),
+                using=SPARSE_VECTORS_NAME,
+                query_filter=qdrant_filters,
+                limit=top_k,
+                group_by=group_by,
+                group_size=group_size,
+                with_vectors=return_embedding,
+                score_threshold=score_threshold,
+            )
+            groups = response.groups
+            return self._process_group_results(groups)
+        else:
+            response = await self._async_client.query_points(
+                collection_name=self.index,
+                query=rest.SparseVector(
+                    indices=query_indices,
+                    values=query_values,
+                ),
+                using=SPARSE_VECTORS_NAME,
+                query_filter=qdrant_filters,
+                limit=top_k,
+                with_vectors=return_embedding,
+                score_threshold=score_threshold,
+            )
+            points = response.points
+            return self._process_query_point_results(points, scale_score=scale_score)
+
+    async def _query_by_embedding_async(
+        self,
+        query_embedding: List[float],
+        filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
+        top_k: int = 10,
+        scale_score: bool = False,
+        return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
+        group_by: Optional[str] = None,
+        group_size: Optional[int] = None,
+    ) -> List[Document]:
+        """
+        Asynchronously queries Qdrant using a dense embedding and returns the most relevant documents.
+
+        :param query_embedding: Dense embedding of the query.
+        :param filters: Filters applied to the retrieved documents.
+        :param top_k: Maximum number of documents to return. If using `group_by` parameters, maximum number of
+             groups to return.
+        :param scale_score: Whether to scale the scores of the retrieved documents.
+        :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
+        :param group_by: Payload field to group by, must be a string or number field. If the field contains more than 1
+             value, all values will be used for grouping. One point can be in multiple groups.
+        :param group_size: Maximum amount of points to return per group. Default is 3.
+
+        :returns: List of documents that are most similar to `query_embedding`.
+        """
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
+        qdrant_filters = convert_filters_to_qdrant(filters)
+        if group_by:
+            response = await self._async_client.query_points_groups(
+                collection_name=self.index,
+                query=query_embedding,
+                using=DENSE_VECTORS_NAME if self.use_sparse_embeddings else None,
+                query_filter=qdrant_filters,
+                limit=top_k,
+                group_by=group_by,
+                group_size=group_size,
+                with_vectors=return_embedding,
+                score_threshold=score_threshold,
+            )
+            groups = response.groups
+            return self._process_group_results(groups)
+        else:
+            response = await self._async_client.query_points(
+                collection_name=self.index,
+                query=query_embedding,
+                using=DENSE_VECTORS_NAME if self.use_sparse_embeddings else None,
+                query_filter=qdrant_filters,
+                limit=top_k,
+                with_vectors=return_embedding,
+                score_threshold=score_threshold,
+            )
+            points = response.points
+            return self._process_query_point_results(points, scale_score=scale_score)
+
+    async def _query_hybrid_async(
+        self,
+        query_embedding: List[float],
+        query_sparse_embedding: SparseEmbedding,
+        filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
+        top_k: int = 10,
+        return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
+        group_by: Optional[str] = None,
+        group_size: Optional[int] = None,
+    ) -> List[Document]:
+        """
+        Asynchronously retrieves documents based on dense and sparse embeddings and fuses
+        the results using Reciprocal Rank Fusion.
+
+        This method is not part of the public interface of `QdrantDocumentStore` and shouldn't be used directly.
+        Use the `QdrantHybridRetriever` instead.
+
+        :param query_embedding: Dense embedding of the query.
+        :param query_sparse_embedding: Sparse embedding of the query.
+        :param filters: Filters applied to the retrieved documents.
+        :param top_k: Maximum number of documents to return. If using `group_by` parameters, maximum number of
+             groups to return.
+        :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
+        :param group_by: Payload field to group by, must be a string or number field. If the field contains more than 1
+             value, all values will be used for grouping. One point can be in multiple groups.
+        :param group_size: Maximum amount of points to return per group. Default is 3.
+
+        :returns: List of Document that are most similar to `query_embedding` and `query_sparse_embedding`.
+
+        :raises QdrantStoreError:
+            If the Document Store was initialized with `use_sparse_embeddings=False`.
+        """
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
+
+        if not self.use_sparse_embeddings:
+            message = (
+                "You are trying to query using sparse embeddings, but the Document Store "
+                "was initialized with `use_sparse_embeddings=False`. "
+            )
+            raise QdrantStoreError(message)
+
+        qdrant_filters = convert_filters_to_qdrant(filters)
+
+        try:
+            if group_by:
+                response = await self._async_client.query_points_groups(
+                    collection_name=self.index,
+                    prefetch=[
+                        rest.Prefetch(
+                            query=rest.SparseVector(
+                                indices=query_sparse_embedding.indices,
+                                values=query_sparse_embedding.values,
+                            ),
+                            using=SPARSE_VECTORS_NAME,
+                            filter=qdrant_filters,
+                        ),
+                        rest.Prefetch(
+                            query=query_embedding,
+                            using=DENSE_VECTORS_NAME,
+                            filter=qdrant_filters,
+                        ),
+                    ],
+                    query=rest.FusionQuery(fusion=rest.Fusion.RRF),
+                    limit=top_k,
+                    group_by=group_by,
+                    group_size=group_size,
+                    score_threshold=score_threshold,
+                    with_payload=True,
+                    with_vectors=return_embedding,
+                )
+                groups = response.groups
+            else:
+                response = await self._async_client.query_points(
+                    collection_name=self.index,
+                    prefetch=[
+                        rest.Prefetch(
+                            query=rest.SparseVector(
+                                indices=query_sparse_embedding.indices,
+                                values=query_sparse_embedding.values,
+                            ),
+                            using=SPARSE_VECTORS_NAME,
+                            filter=qdrant_filters,
+                        ),
+                        rest.Prefetch(
+                            query=query_embedding,
+                            using=DENSE_VECTORS_NAME,
+                            filter=qdrant_filters,
+                        ),
+                    ],
+                    query=rest.FusionQuery(fusion=rest.Fusion.RRF),
+                    limit=top_k,
+                    score_threshold=score_threshold,
+                    with_payload=True,
+                    with_vectors=return_embedding,
+                )
+                points = response.points
+
+        except Exception as e:
+            msg = "Error during hybrid search"
+            raise QdrantStoreError(msg) from e
+
+        if group_by:
+            return self._process_group_results(groups)
+        else:
+            return self._process_query_point_results(points)
+
+    def get_distance(self, similarity: str) -> rest.Distance:
+        """
+        Retrieves the distance metric for the specified similarity measure.
+
+        :param similarity:
+            The similarity measure to retrieve the distance.
+        :returns:
+            The corresponding rest.Distance object.
+        :raises QdrantStoreError:
+            If the provided similarity measure is not supported.
+        """
+        try:
+            return self.SIMILARITY[similarity]
+        except KeyError as ke:
+            msg = (
+                f"Provided similarity '{similarity}' is not supported by Qdrant "
+                f"document store. Please choose one of the options: "
+                f"{', '.join(self.SIMILARITY.keys())}"
+            )
+            raise QdrantStoreError(msg) from ke
+
+    def _create_payload_index(self, collection_name: str, payload_fields_to_index: Optional[List[dict]] = None):
+        """
+        Create payload index for the collection if payload_fields_to_index is provided
+        See: https://qdrant.tech/documentation/concepts/indexing/#payload-index
+        """
+        if payload_fields_to_index is not None:
+            for payload_index in payload_fields_to_index:
+                # self._client is initialized at this point
+                # since _initialize_client() is called before this method is executed
+
+                assert self._client is not None
+                self._client.create_payload_index(
+                    collection_name=collection_name,
+                    field_name=payload_index["field_name"],
+                    field_schema=payload_index["field_schema"],
+                )
+
+    async def _create_payload_index_async(
+        self, collection_name: str, payload_fields_to_index: Optional[List[dict]] = None
+    ):
+        """
+        Asynchronously create payload index for the collection if payload_fields_to_index is provided
+        See: https://qdrant.tech/documentation/concepts/indexing/#payload-index
+        """
+        if payload_fields_to_index is not None:
+            for payload_index in payload_fields_to_index:
+
+                # self._async_client is initialized at this point
+                # since _initialize_async_client() is called before this method is executed
+                assert self._async_client is not None
+
+                await self._async_client.create_payload_index(
+                    collection_name=collection_name,
+                    field_name=payload_index["field_name"],
+                    field_schema=payload_index["field_schema"],
+                )
+
+    def _set_up_collection(
+        self,
+        collection_name: str,
+        embedding_dim: int,
+        recreate_collection: bool,
+        similarity: str,
+        use_sparse_embeddings: bool,
+        sparse_idf: bool,
+        on_disk: bool = False,
+        payload_fields_to_index: Optional[List[dict]] = None,
+    ):
+        """
+        Sets up the Qdrant collection with the specified parameters.
+        :param collection_name:
+            The name of the collection to set up.
+        :param embedding_dim:
+            The dimension of the embeddings.
         :param recreate_collection:
             Whether to recreate the collection if it already exists.
         :param similarity:
@@ -852,9 +1283,13 @@ class QdrantDocumentStore:
             If the collection exists with a different similarity measure or embedding dimension.
 
         """
+
+        self._initialize_client()
+        assert self._client is not None
+
         distance = self.get_distance(similarity)
 
-        if recreate_collection or not self.client.collection_exists(collection_name):
+        if recreate_collection or not self._client.collection_exists(collection_name):
             # There is no need to verify the current configuration of that
             # collection. It might be just recreated again or does not exist yet.
             self.recreate_collection(
@@ -864,64 +1299,65 @@ class QdrantDocumentStore:
             self._create_payload_index(collection_name, payload_fields_to_index)
             return
 
-        collection_info = self.client.get_collection(collection_name)
+        collection_info = self._client.get_collection(collection_name)
 
-        has_named_vectors = isinstance(collection_info.config.params.vectors, dict)
+        self._validate_collection_compatibility(collection_name, collection_info, distance, embedding_dim)
 
-        if has_named_vectors and DENSE_VECTORS_NAME not in collection_info.config.params.vectors:
-            msg = (
-                f"Collection '{collection_name}' already exists in Qdrant, "
-                f"but it has been originally created outside of Haystack and is not supported. "
-                f"If possible, you should create a new Document Store with Haystack. "
-                f"In case you want to migrate the existing collection, see an example script in "
-                f"https://github.com/deepset-ai/haystack-core-integrations/blob/main/integrations/qdrant/src/"
-                f"haystack_integrations/document_stores/qdrant/migrate_to_sparse.py."
-            )
-            raise QdrantStoreError(msg)
+    async def _set_up_collection_async(
+        self,
+        collection_name: str,
+        embedding_dim: int,
+        recreate_collection: bool,
+        similarity: str,
+        use_sparse_embeddings: bool,
+        sparse_idf: bool,
+        on_disk: bool = False,
+        payload_fields_to_index: Optional[List[dict]] = None,
+    ):
+        """
+        Asynchronously sets up the Qdrant collection with the specified parameters.
+        :param collection_name:
+            The name of the collection to set up.
+        :param embedding_dim:
+            The dimension of the embeddings.
+        :param recreate_collection:
+            Whether to recreate the collection if it already exists.
+        :param similarity:
+            The similarity measure to use.
+        :param use_sparse_embeddings:
+            Whether to use sparse embeddings.
+        :param sparse_idf:
+            Whether to compute the Inverse Document Frequency (IDF) when using sparse embeddings. Required for BM42.
+        :param on_disk:
+            Whether to store the collection on disk.
+        :param payload_fields_to_index:
+            List of payload fields to index.
 
-        if self.use_sparse_embeddings and not has_named_vectors:
-            msg = (
-                f"Collection '{collection_name}' already exists in Qdrant, "
-                f"but it has been originally created without sparse embedding vectors. "
-                f"If you want to use that collection, you can set `use_sparse_embeddings=False`. "
-                f"To use sparse embeddings, you need to recreate the collection or migrate the existing one. "
-                f"See `migrate_to_sparse_embeddings_support` function in "
-                f"`haystack_integrations.document_stores.qdrant`."
-            )
-            raise QdrantStoreError(msg)
+        :raises QdrantStoreError:
+            If the collection exists with incompatible settings.
+        :raises ValueError:
+            If the collection exists with a different similarity measure or embedding dimension.
 
-        if not self.use_sparse_embeddings and has_named_vectors:
-            msg = (
-                f"Collection '{collection_name}' already exists in Qdrant, "
-                f"but it has been originally created with sparse embedding vectors."
-                f"If you want to use that collection, please set `use_sparse_embeddings=True`."
-            )
-            raise QdrantStoreError(msg)
+        """
 
-        if self.use_sparse_embeddings:
-            current_distance = collection_info.config.params.vectors[DENSE_VECTORS_NAME].distance
-            current_vector_size = collection_info.config.params.vectors[DENSE_VECTORS_NAME].size
-        else:
-            current_distance = collection_info.config.params.vectors.distance
-            current_vector_size = collection_info.config.params.vectors.size
+        await self._initialize_async_client()
+        assert self._async_client is not None
 
-        if current_distance != distance:
-            msg = (
-                f"Collection '{collection_name}' already exists in Qdrant, "
-                f"but it is configured with a similarity '{current_distance.name}'. "
-                f"If you want to use that collection, but with a different "
-                f"similarity, please set `recreate_collection=True` argument."
-            )
-            raise ValueError(msg)
+        distance = self.get_distance(similarity)
 
-        if current_vector_size != embedding_dim:
-            msg = (
-                f"Collection '{collection_name}' already exists in Qdrant, "
-                f"but it is configured with a vector size '{current_vector_size}'. "
-                f"If you want to use that collection, but with a different "
-                f"vector size, please set `recreate_collection=True` argument."
+        if recreate_collection or not await self._async_client.collection_exists(collection_name):
+            # There is no need to verify the current configuration of that
+            # collection. It might be just recreated again or does not exist yet.
+            await self.recreate_collection_async(
+                collection_name, distance, embedding_dim, on_disk, use_sparse_embeddings, sparse_idf
             )
-            raise ValueError(msg)
+            # Create Payload index if payload_fields_to_index is provided
+            await self._create_payload_index_async(collection_name, payload_fields_to_index)
+            return
+
+        collection_info = await self._async_client.get_collection(collection_name)
+
+        self._validate_collection_compatibility(collection_name, collection_info, distance, embedding_dim)
 
     def recreate_collection(
         self,
@@ -948,44 +1384,65 @@ class QdrantDocumentStore:
         :param sparse_idf:
             Whether to compute the Inverse Document Frequency (IDF) when using sparse embeddings. Required for BM42.
         """
-        if on_disk is None:
-            on_disk = self.on_disk
+        vectors_config, sparse_vectors_config = self._prepare_collection_config(
+            embedding_dim, distance, on_disk, use_sparse_embeddings, sparse_idf
+        )
+        collection_params = self._prepare_collection_params()
 
-        if use_sparse_embeddings is None:
-            use_sparse_embeddings = self.use_sparse_embeddings
+        self._initialize_client()
+        assert self._client is not None
 
-        # dense vectors configuration
-        vectors_config = rest.VectorParams(size=embedding_dim, on_disk=on_disk, distance=distance)
+        if self._client.collection_exists(collection_name):
+            self._client.delete_collection(collection_name)
 
-        if use_sparse_embeddings:
-            # in this case, we need to define named vectors
-            vectors_config = {DENSE_VECTORS_NAME: vectors_config}
+        self._client.create_collection(
+            collection_name=collection_name,
+            vectors_config=vectors_config,
+            sparse_vectors_config=sparse_vectors_config,
+            **collection_params,
+        )
 
-            sparse_vectors_config = {
-                SPARSE_VECTORS_NAME: rest.SparseVectorParams(
-                    index=rest.SparseIndexParams(
-                        on_disk=on_disk,
-                    ),
-                    modifier=rest.Modifier.IDF if sparse_idf else None,
-                ),
-            }
+    async def recreate_collection_async(
+        self,
+        collection_name: str,
+        distance,
+        embedding_dim: int,
+        on_disk: Optional[bool] = None,
+        use_sparse_embeddings: Optional[bool] = None,
+        sparse_idf: bool = False,
+    ):
+        """
+        Asynchronously recreates the Qdrant collection with the specified parameters.
+
+        :param collection_name:
+            The name of the collection to recreate.
+        :param distance:
+            The distance metric to use for the collection.
+        :param embedding_dim:
+            The dimension of the embeddings.
+        :param on_disk:
+            Whether to store the collection on disk.
+        :param use_sparse_embeddings:
+            Whether to use sparse embeddings.
+        :param sparse_idf:
+            Whether to compute the Inverse Document Frequency (IDF) when using sparse embeddings. Required for BM42.
+        """
+        vectors_config, sparse_vectors_config = self._prepare_collection_config(
+            embedding_dim, distance, on_disk, use_sparse_embeddings, sparse_idf
+        )
+        collection_params = self._prepare_collection_params()
+
+        await self._initialize_async_client()
+        assert self._async_client is not None
 
-        if self.client.collection_exists(collection_name):
-            self.client.delete_collection(collection_name)
+        if await self._async_client.collection_exists(collection_name):
+            await self._async_client.delete_collection(collection_name)
 
-        self.client.create_collection(
+        await self._async_client.create_collection(
             collection_name=collection_name,
             vectors_config=vectors_config,
-            sparse_vectors_config=sparse_vectors_config if use_sparse_embeddings else None,
-            shard_number=self.shard_number,
-            replication_factor=self.replication_factor,
-            write_consistency_factor=self.write_consistency_factor,
-            on_disk_payload=self.on_disk_payload,
-            hnsw_config=self.hnsw_config,
-            optimizers_config=self.optimizers_config,
-            wal_config=self.wal_config,
-            quantization_config=self.quantization_config,
-            init_from=self.init_from,
+            sparse_vectors_config=sparse_vectors_config,
+            **collection_params,
         )
 
     def _handle_duplicate_documents(
@@ -1015,12 +1472,38 @@ class QdrantDocumentStore:
 
         return documents
 
-    def _drop_duplicate_documents(self, documents: List[Document]) -> List[Document]:
+    async def _handle_duplicate_documents_async(
+        self,
+        documents: List[Document],
+        policy: DuplicatePolicy = None,
+    ):
         """
-        Drop duplicate documents based on same hash ID.
+        Asynchronously checks whether any of the passed documents is already existing
+        in the chosen index and returns a list of
+        documents that are not in the index yet.
 
         :param documents: A list of Haystack Document objects.
+        :param policy: The duplicate policy to use when writing documents.
         :returns: A list of Haystack Document objects.
+        """
+
+        if policy in (DuplicatePolicy.SKIP, DuplicatePolicy.FAIL):
+            documents = self._drop_duplicate_documents(documents)
+            documents_found = await self.get_documents_by_id_async(ids=[doc.id for doc in documents])
+            ids_exist_in_db: List[str] = [doc.id for doc in documents_found]
+
+            if len(ids_exist_in_db) > 0 and policy == DuplicatePolicy.FAIL:
+                msg = f"Document with ids '{', '.join(ids_exist_in_db)} already exists in index = '{self.index}'."
+                raise DuplicateDocumentError(msg)
+
+            documents = list(filter(lambda doc: doc.id not in ids_exist_in_db, documents))
+
+        return documents
+
+    def _drop_duplicate_documents(self, documents: List[Document]) -> List[Document]:
+        """
+        Drop duplicate documents based on same hash ID.
+
         """
         _hash_ids: Set = set()
         _documents: List[Document] = []
@@ -1037,3 +1520,193 @@ class QdrantDocumentStore:
             _hash_ids.add(document.id)
 
         return _documents
+
+    def _prepare_collection_params(self):
+        """
+        Prepares the common parameters for collection creation.
+        """
+        return {
+            "shard_number": self.shard_number,
+            "replication_factor": self.replication_factor,
+            "write_consistency_factor": self.write_consistency_factor,
+            "on_disk_payload": self.on_disk_payload,
+            "hnsw_config": self.hnsw_config,
+            "optimizers_config": self.optimizers_config,
+            "wal_config": self.wal_config,
+            "quantization_config": self.quantization_config,
+            "init_from": self.init_from,
+        }
+
+    def _prepare_client_params(self):
+        """
+        Prepares the common parameters for client initialization.
+
+        """
+        return {
+            "location": self.location,
+            "url": self.url,
+            "port": self.port,
+            "grpc_port": self.grpc_port,
+            "prefer_grpc": self.prefer_grpc,
+            "https": self.https,
+            "api_key": self.api_key.resolve_value() if self.api_key else None,
+            "prefix": self.prefix,
+            "timeout": self.timeout,
+            "host": self.host,
+            "path": self.path,
+            "metadata": self.metadata,
+            "force_disable_check_same_thread": self.force_disable_check_same_thread,
+        }
+
+    def _prepare_collection_config(
+        self,
+        embedding_dim: int,
+        distance,
+        on_disk: Optional[bool] = None,
+        use_sparse_embeddings: Optional[bool] = None,
+        sparse_idf: bool = False,
+    ):
+        """
+        Prepares the configuration for creating or recreating a Qdrant collection.
+
+        """
+        if on_disk is None:
+            on_disk = self.on_disk
+
+        if use_sparse_embeddings is None:
+            use_sparse_embeddings = self.use_sparse_embeddings
+
+        # dense vectors configuration
+        vectors_config = rest.VectorParams(size=embedding_dim, on_disk=on_disk, distance=distance)
+        sparse_vectors_config = None
+
+        if use_sparse_embeddings:
+            # in this case, we need to define named vectors
+            vectors_config = {DENSE_VECTORS_NAME: vectors_config}
+
+            sparse_vectors_config = {
+                SPARSE_VECTORS_NAME: rest.SparseVectorParams(
+                    index=rest.SparseIndexParams(
+                        on_disk=on_disk,
+                    ),
+                    modifier=rest.Modifier.IDF if sparse_idf else None,
+                ),
+            }
+
+        return vectors_config, sparse_vectors_config
+
+    def _validate_filters(self, filters: Optional[Union[Dict[str, Any], rest.Filter]] = None):
+        """
+        Validates the filters provided for querying.
+        """
+        if filters and not isinstance(filters, dict) and not isinstance(filters, rest.Filter):
+            msg = "Filter must be a dictionary or an instance of `qdrant_client.http.models.Filter`"
+            raise ValueError(msg)
+
+        if filters and not isinstance(filters, rest.Filter) and "operator" not in filters:
+            msg = "Invalid filter syntax. See https://docs.haystack.deepset.ai/docs/metadata-filtering for details."
+            raise ValueError(msg)
+
+    def _process_query_point_results(self, results, scale_score: bool = False):
+        """
+        Processes query results from Qdrant.
+        """
+        documents = [
+            convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
+            for point in results
+        ]
+
+        if scale_score:
+            for document in documents:
+                score = document.score
+                if self.similarity == "cosine":
+                    score = (score + 1) / 2
+                else:
+                    score = float(1 / (1 + np.exp(-score / 100)))
+                document.score = score
+
+        return documents
+
+    def _process_group_results(self, groups):
+        """
+        Processes grouped query results from Qdrant.
+
+        """
+        if not groups:
+            return []
+
+        return [
+            convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
+            for group in groups
+            for point in group.hits
+        ]
+
+    def _validate_collection_compatibility(
+        self,
+        collection_name: str,
+        collection_info,
+        distance,
+        embedding_dim: int,
+    ):
+        """
+        Validates that an existing collection is compatible with the current configuration.
+        """
+        has_named_vectors = isinstance(collection_info.config.params.vectors, dict)
+
+        if has_named_vectors and DENSE_VECTORS_NAME not in collection_info.config.params.vectors:
+            msg = (
+                f"Collection '{collection_name}' already exists in Qdrant, "
+                f"but it has been originally created outside of Haystack and is not supported. "
+                f"If possible, you should create a new Document Store with Haystack. "
+                f"In case you want to migrate the existing collection, see an example script in "
+                f"https://github.com/deepset-ai/haystack-core-integrations/blob/main/integrations/qdrant/src/"
+                f"haystack_integrations/document_stores/qdrant/migrate_to_sparse.py."
+            )
+            raise QdrantStoreError(msg)
+
+        if self.use_sparse_embeddings and not has_named_vectors:
+            msg = (
+                f"Collection '{collection_name}' already exists in Qdrant, "
+                f"but it has been originally created without sparse embedding vectors. "
+                f"If you want to use that collection, you can set `use_sparse_embeddings=False`. "
+                f"To use sparse embeddings, you need to recreate the collection or migrate the existing one. "
+                f"See `migrate_to_sparse_embeddings_support` function in "
+                f"`haystack_integrations.document_stores.qdrant`."
+            )
+            raise QdrantStoreError(msg)
+
+        if not self.use_sparse_embeddings and has_named_vectors:
+            msg = (
+                f"Collection '{collection_name}' already exists in Qdrant, "
+                f"but it has been originally created with sparse embedding vectors."
+                f"If you want to use that collection, please set `use_sparse_embeddings=True`."
+            )
+            raise QdrantStoreError(msg)
+
+        # Get current distance and vector size based on collection configuration
+        if self.use_sparse_embeddings:
+            current_distance = collection_info.config.params.vectors[DENSE_VECTORS_NAME].distance
+            current_vector_size = collection_info.config.params.vectors[DENSE_VECTORS_NAME].size
+        else:
+            current_distance = collection_info.config.params.vectors.distance
+            current_vector_size = collection_info.config.params.vectors.size
+
+        # Validate distance metric
+        if current_distance != distance:
+            msg = (
+                f"Collection '{collection_name}' already exists in Qdrant, "
+                f"but it is configured with a similarity '{current_distance.name}'. "
+                f"If you want to use that collection, but with a different "
+                f"similarity, please set `recreate_collection=True` argument."
+            )
+            raise ValueError(msg)
+
+        # Validate embedding dimension
+        if current_vector_size != embedding_dim:
+            msg = (
+                f"Collection '{collection_name}' already exists in Qdrant, "
+                f"but it is configured with a vector size '{current_vector_size}'. "
+                f"If you want to use that collection, but with a different "
+                f"vector size, please set `recreate_collection=True` argument."
+            )
+            raise ValueError(msg)