from __future__ import annotations import uuid import warnings from concurrent.futures import ThreadPoolExecutor from typing import ( Any, Callable, Dict, Iterable, List, Optional, Set, Tuple, Type, TypeVar, ) import numpy as np from langchain_core.documents import Document from langchain_core.embeddings import Embeddings from langchain_core.utils.iter import batch_iterate from langchain_core.vectorstores import VectorStore from langchain_community.vectorstores.utils import maximal_marginal_relevance ADBVST = TypeVar("ADBVST", bound="AstraDB") T = TypeVar("T") U = TypeVar("U") DocDict = Dict[str, Any] # dicts expressing entries to insert # Batch/concurrency default values (if parameters not provided): # Size of batches for bulk insertions: # (20 is the max batch size for the HTTP API at the time of writing) DEFAULT_BATCH_SIZE = 20 # Number of threads to insert batches concurrently: DEFAULT_BULK_INSERT_BATCH_CONCURRENCY = 16 # Number of threads in a batch to insert pre-existing entries: DEFAULT_BULK_INSERT_OVERWRITE_CONCURRENCY = 10 # Number of threads (for deleting multiple rows concurrently): DEFAULT_BULK_DELETE_CONCURRENCY = 20 def _unique_list(lst: List[T], key: Callable[[T], U]) -> List[T]: visited_keys: Set[U] = set() new_lst = [] for item in lst: item_key = key(item) if item_key not in visited_keys: visited_keys.add(item_key) new_lst.append(item) return new_lst class AstraDB(VectorStore): """Wrapper around DataStax Astra DB for vector-store workloads. To use it, you need a recent installation of the `astrapy` library and an Astra DB cloud database. For quickstart and details, visit: docs.datastax.com/en/astra/home/astra.html Example: .. code-block:: python from langchain_community.vectorstores import AstraDB from langchain_community.embeddings.openai import OpenAIEmbeddings embeddings = OpenAIEmbeddings() vectorstore = AstraDB( embedding=embeddings, collection_name="my_store", token="AstraCS:...", api_endpoint="https://-us-east1.apps.astra.datastax.com" ) vectorstore.add_texts(["Giraffes", "All good here"]) results = vectorstore.similarity_search("Everything's ok", k=1) Constructor Args (only keyword-arguments accepted): embedding (Embeddings): embedding function to use. collection_name (str): name of the Astra DB collection to create/use. token (Optional[str]): API token for Astra DB usage. api_endpoint (Optional[str]): full URL to the API endpoint, such as "https://-us-east1.apps.astra.datastax.com". astra_db_client (Optional[Any]): *alternative to token+api_endpoint*, you can pass an already-created 'astrapy.db.AstraDB' instance. namespace (Optional[str]): namespace (aka keyspace) where the collection is created. Defaults to the database's "default namespace". metric (Optional[str]): similarity function to use out of those available in Astra DB. If left out, it will use Astra DB API's defaults (i.e. "cosine" - but, for performance reasons, "dot_product" is suggested if embeddings are normalized to one). Advanced arguments (coming with sensible defaults): batch_size (Optional[int]): Size of batches for bulk insertions. bulk_insert_batch_concurrency (Optional[int]): Number of threads to insert batches concurrently. bulk_insert_overwrite_concurrency (Optional[int]): Number of threads in a batch to insert pre-existing entries. bulk_delete_concurrency (Optional[int]): Number of threads (for deleting multiple rows concurrently). pre_delete_collection (Optional[bool]): whether to delete the collection before creating it. If False and the collection already exists, the collection will be used as is. A note on concurrency: as a rule of thumb, on a typical client machine it is suggested to keep the quantity bulk_insert_batch_concurrency * bulk_insert_overwrite_concurrency much below 1000 to avoid exhausting the client multithreading/networking resources. The hardcoded defaults are somewhat conservative to meet most machines' specs, but a sensible choice to test may be: bulk_insert_batch_concurrency = 80 bulk_insert_overwrite_concurrency = 10 A bit of experimentation is required to nail the best results here, depending on both the machine/network specs and the expected workload (specifically, how often a write is an update of an existing id). Remember you can pass concurrency settings to individual calls to add_texts and add_documents as well. """ @staticmethod def _filter_to_metadata(filter_dict: Optional[Dict[str, Any]]) -> Dict[str, Any]: if filter_dict is None: return {} else: metadata_filter = {} for k, v in filter_dict.items(): if k and k[0] == "$": if isinstance(v, list): metadata_filter[k] = [AstraDB._filter_to_metadata(f) for f in v] else: metadata_filter[k] = AstraDB._filter_to_metadata(v) else: metadata_filter[f"metadata.{k}"] = v return metadata_filter def __init__( self, *, embedding: Embeddings, collection_name: str, token: Optional[str] = None, api_endpoint: Optional[str] = None, astra_db_client: Optional[Any] = None, # 'astrapy.db.AstraDB' if passed namespace: Optional[str] = None, metric: Optional[str] = None, batch_size: Optional[int] = None, bulk_insert_batch_concurrency: Optional[int] = None, bulk_insert_overwrite_concurrency: Optional[int] = None, bulk_delete_concurrency: Optional[int] = None, pre_delete_collection: bool = False, ) -> None: """ Create an AstraDB vector store object. See class docstring for help. """ try: from astrapy.db import ( AstraDB as LibAstraDB, ) from astrapy.db import ( AstraDBCollection as LibAstraDBCollection, ) except (ImportError, ModuleNotFoundError): raise ImportError( "Could not import a recent astrapy python package. " "Please install it with `pip install --upgrade astrapy`." ) # Conflicting-arg checks: if astra_db_client is not None: if token is not None or api_endpoint is not None: raise ValueError( "You cannot pass 'astra_db_client' to AstraDB if passing " "'token' and 'api_endpoint'." ) self.embedding = embedding self.collection_name = collection_name self.token = token self.api_endpoint = api_endpoint self.namespace = namespace # Concurrency settings self.batch_size: int = batch_size or DEFAULT_BATCH_SIZE self.bulk_insert_batch_concurrency: int = ( bulk_insert_batch_concurrency or DEFAULT_BULK_INSERT_BATCH_CONCURRENCY ) self.bulk_insert_overwrite_concurrency: int = ( bulk_insert_overwrite_concurrency or DEFAULT_BULK_INSERT_OVERWRITE_CONCURRENCY ) self.bulk_delete_concurrency: int = ( bulk_delete_concurrency or DEFAULT_BULK_DELETE_CONCURRENCY ) # "vector-related" settings self._embedding_dimension: Optional[int] = None self.metric = metric if astra_db_client is not None: self.astra_db = astra_db_client else: self.astra_db = LibAstraDB( token=self.token, api_endpoint=self.api_endpoint, namespace=self.namespace, ) if not pre_delete_collection: self._provision_collection() else: self.clear() self.collection = LibAstraDBCollection( collection_name=self.collection_name, astra_db=self.astra_db, ) def _get_embedding_dimension(self) -> int: if self._embedding_dimension is None: self._embedding_dimension = len( self.embedding.embed_query("This is a sample sentence.") ) return self._embedding_dimension def _drop_collection(self) -> None: """ Drop the collection from storage. This is meant as an internal-usage method, no members are set other than actual deletion on the backend. """ _ = self.astra_db.delete_collection( collection_name=self.collection_name, ) return None def _provision_collection(self) -> None: """ Run the API invocation to create the collection on the backend. Internal-usage method, no object members are set, other than working on the underlying actual storage. """ _ = self.astra_db.create_collection( dimension=self._get_embedding_dimension(), collection_name=self.collection_name, metric=self.metric, ) return None @property def embeddings(self) -> Embeddings: return self.embedding @staticmethod def _dont_flip_the_cos_score(similarity0to1: float) -> float: """Keep similarity from client unchanged ad it's in [0:1] already.""" return similarity0to1 def _select_relevance_score_fn(self) -> Callable[[float], float]: """ The underlying API calls already returns a "score proper", i.e. one in [0, 1] where higher means more *similar*, so here the final score transformation is not reversing the interval: """ return self._dont_flip_the_cos_score def clear(self) -> None: """Empty the collection of all its stored entries.""" self._drop_collection() self._provision_collection() return None def delete_by_document_id(self, document_id: str) -> bool: """ Remove a single document from the store, given its document_id (str). Return True if a document has indeed been deleted, False if ID not found. """ deletion_response = self.collection.delete(document_id) return ((deletion_response or {}).get("status") or {}).get( "deletedCount", 0 ) == 1 def delete( self, ids: Optional[List[str]] = None, concurrency: Optional[int] = None, **kwargs: Any, ) -> Optional[bool]: """Delete by vector ids. Args: ids (Optional[List[str]]): List of ids to delete. concurrency (Optional[int]): max number of threads issuing single-doc delete requests. Defaults to instance-level setting. Returns: Optional[bool]: True if deletion is successful, False otherwise, None if not implemented. """ if kwargs: warnings.warn( "Method 'delete' of AstraDB vector store invoked with " f"unsupported arguments ({', '.join(sorted(kwargs.keys()))}), " "which will be ignored." ) if ids is None: raise ValueError("No ids provided to delete.") _max_workers = concurrency or self.bulk_delete_concurrency with ThreadPoolExecutor(max_workers=_max_workers) as tpe: _ = list( tpe.map( self.delete_by_document_id, ids, ) ) return True def delete_collection(self) -> None: """ Completely delete the collection from the database (as opposed to 'clear()', which empties it only). Stored data is lost and unrecoverable, resources are freed. Use with caution. """ self._drop_collection() return None def add_texts( self, texts: Iterable[str], metadatas: Optional[List[dict]] = None, ids: Optional[List[str]] = None, *, batch_size: Optional[int] = None, batch_concurrency: Optional[int] = None, overwrite_concurrency: Optional[int] = None, **kwargs: Any, ) -> List[str]: """Run texts through the embeddings and add them to the vectorstore. If passing explicit ids, those entries whose id is in the store already will be replaced. Args: texts (Iterable[str]): Texts to add to the vectorstore. metadatas (Optional[List[dict]], optional): Optional list of metadatas. ids (Optional[List[str]], optional): Optional list of ids. batch_size (Optional[int]): Number of documents in each API call. Check the underlying Astra DB HTTP API specs for the max value (20 at the time of writing this). If not provided, defaults to the instance-level setting. batch_concurrency (Optional[int]): number of threads to process insertion batches concurrently. Defaults to instance-level setting if not provided. overwrite_concurrency (Optional[int]): number of threads to process pre-existing documents in each batch (which require individual API calls). Defaults to instance-level setting if not provided. A note on metadata: there are constraints on the allowed field names in this dictionary, coming from the underlying Astra DB API. For instance, the `$` (dollar sign) cannot be used in the dict keys. See this document for details: docs.datastax.com/en/astra-serverless/docs/develop/dev-with-json.html Returns: List[str]: List of ids of the added texts. """ if kwargs: warnings.warn( "Method 'add_texts' of AstraDB vector store invoked with " f"unsupported arguments ({', '.join(sorted(kwargs.keys()))}), " "which will be ignored." ) _texts = list(texts) if ids is None: ids = [uuid.uuid4().hex for _ in _texts] if metadatas is None: metadatas = [{} for _ in _texts] # embedding_vectors = self.embedding.embed_documents(_texts) documents_to_insert = [ { "content": b_txt, "_id": b_id, "$vector": b_emb, "metadata": b_md, } for b_txt, b_emb, b_id, b_md in zip( _texts, embedding_vectors, ids, metadatas, ) ] # make unique by id, keeping the last uniqued_documents_to_insert = _unique_list( documents_to_insert[::-1], lambda document: document["_id"], )[::-1] all_ids = [] def _handle_batch(document_batch: List[DocDict]) -> List[str]: im_result = self.collection.insert_many( documents=document_batch, options={"ordered": False}, partial_failures_allowed=True, ) if "status" not in im_result: raise ValueError( f"API Exception while running bulk insertion: {str(im_result)}" ) batch_inserted = im_result["status"]["insertedIds"] # estimation of the preexisting documents that failed missed_inserted_ids = { document["_id"] for document in document_batch } - set(batch_inserted) errors = im_result.get("errors", []) # careful for other sources of error other than "doc already exists" num_errors = len(errors) unexpected_errors = any( error.get("errorCode") != "DOCUMENT_ALREADY_EXISTS" for error in errors ) if num_errors != len(missed_inserted_ids) or unexpected_errors: raise ValueError( f"API Exception while running bulk insertion: {str(errors)}" ) # deal with the missing insertions as upserts missing_from_batch = [ document for document in document_batch if document["_id"] in missed_inserted_ids ] def _handle_missing_document(missing_document: DocDict) -> str: replacement_result = self.collection.find_one_and_replace( filter={"_id": missing_document["_id"]}, replacement=missing_document, ) return replacement_result["data"]["document"]["_id"] _u_max_workers = ( overwrite_concurrency or self.bulk_insert_overwrite_concurrency ) with ThreadPoolExecutor(max_workers=_u_max_workers) as tpe2: batch_replaced = list( tpe2.map( _handle_missing_document, missing_from_batch, ) ) upsert_ids = batch_inserted + batch_replaced return upsert_ids _b_max_workers = batch_concurrency or self.bulk_insert_batch_concurrency with ThreadPoolExecutor(max_workers=_b_max_workers) as tpe: all_ids_nested = tpe.map( _handle_batch, batch_iterate( batch_size or self.batch_size, uniqued_documents_to_insert, ), ) all_ids = [iid for id_list in all_ids_nested for iid in id_list] return all_ids def similarity_search_with_score_id_by_vector( self, embedding: List[float], k: int = 4, filter: Optional[Dict[str, Any]] = None, ) -> List[Tuple[Document, float, str]]: """Return docs most similar to embedding vector. Args: embedding (str): Embedding to look up documents similar to. k (int): Number of Documents to return. Defaults to 4. Returns: List of (Document, score, id), the most similar to the query vector. """ metadata_parameter = self._filter_to_metadata(filter) # hits = list( self.collection.paginated_find( filter=metadata_parameter, sort={"$vector": embedding}, options={"limit": k, "includeSimilarity": True}, projection={ "_id": 1, "content": 1, "metadata": 1, }, ) ) # return [ ( Document( page_content=hit["content"], metadata=hit["metadata"], ), hit["$similarity"], hit["_id"], ) for hit in hits ] def similarity_search_with_score_id( self, query: str, k: int = 4, filter: Optional[Dict[str, Any]] = None, ) -> List[Tuple[Document, float, str]]: embedding_vector = self.embedding.embed_query(query) return self.similarity_search_with_score_id_by_vector( embedding=embedding_vector, k=k, filter=filter, ) def similarity_search_with_score_by_vector( self, embedding: List[float], k: int = 4, filter: Optional[Dict[str, Any]] = None, ) -> List[Tuple[Document, float]]: """Return docs most similar to embedding vector. Args: embedding (str): Embedding to look up documents similar to. k (int): Number of Documents to return. Defaults to 4. Returns: List of (Document, score), the most similar to the query vector. """ return [ (doc, score) for (doc, score, doc_id) in self.similarity_search_with_score_id_by_vector( embedding=embedding, k=k, filter=filter, ) ] def similarity_search( self, query: str, k: int = 4, filter: Optional[Dict[str, Any]] = None, **kwargs: Any, ) -> List[Document]: embedding_vector = self.embedding.embed_query(query) return self.similarity_search_by_vector( embedding_vector, k, filter=filter, ) def similarity_search_by_vector( self, embedding: List[float], k: int = 4, filter: Optional[Dict[str, Any]] = None, **kwargs: Any, ) -> List[Document]: return [ doc for doc, _ in self.similarity_search_with_score_by_vector( embedding, k, filter=filter, ) ] def similarity_search_with_score( self, query: str, k: int = 4, filter: Optional[Dict[str, Any]] = None, ) -> List[Tuple[Document, float]]: embedding_vector = self.embedding.embed_query(query) return self.similarity_search_with_score_by_vector( embedding_vector, k, filter=filter, ) def max_marginal_relevance_search_by_vector( self, embedding: List[float], k: int = 4, fetch_k: int = 20, lambda_mult: float = 0.5, filter: Optional[Dict[str, Any]] = None, **kwargs: Any, ) -> List[Document]: """Return docs selected using the maximal marginal relevance. Maximal marginal relevance optimizes for similarity to query AND diversity among selected documents. Args: embedding: Embedding to look up documents similar to. k: Number of Documents to return. fetch_k: Number of Documents to fetch to pass to MMR algorithm. lambda_mult: Number between 0 and 1 that determines the degree of diversity among the results with 0 corresponding to maximum diversity and 1 to minimum diversity. Returns: List of Documents selected by maximal marginal relevance. """ metadata_parameter = self._filter_to_metadata(filter) prefetch_hits = list( self.collection.paginated_find( filter=metadata_parameter, sort={"$vector": embedding}, options={"limit": fetch_k, "includeSimilarity": True}, projection={ "_id": 1, "content": 1, "metadata": 1, "$vector": 1, }, ) ) mmr_chosen_indices = maximal_marginal_relevance( np.array(embedding, dtype=np.float32), [prefetch_hit["$vector"] for prefetch_hit in prefetch_hits], k=k, lambda_mult=lambda_mult, ) mmr_hits = [ prefetch_hit for prefetch_index, prefetch_hit in enumerate(prefetch_hits) if prefetch_index in mmr_chosen_indices ] return [ Document( page_content=hit["content"], metadata=hit["metadata"], ) for hit in mmr_hits ] def max_marginal_relevance_search( self, query: str, k: int = 4, fetch_k: int = 20, lambda_mult: float = 0.5, filter: Optional[Dict[str, Any]] = None, **kwargs: Any, ) -> List[Document]: """Return docs selected using the maximal marginal relevance. Maximal marginal relevance optimizes for similarity to query AND diversity among selected documents. Args: query (str): Text to look up documents similar to. k (int = 4): Number of Documents to return. fetch_k (int = 20): Number of Documents to fetch to pass to MMR algorithm. lambda_mult (float = 0.5): Number between 0 and 1 that determines the degree of diversity among the results with 0 corresponding to maximum diversity and 1 to minimum diversity. Optional. Returns: List of Documents selected by maximal marginal relevance. """ embedding_vector = self.embedding.embed_query(query) return self.max_marginal_relevance_search_by_vector( embedding_vector, k, fetch_k, lambda_mult=lambda_mult, filter=filter, ) @classmethod def from_texts( cls: Type[ADBVST], texts: List[str], embedding: Embeddings, metadatas: Optional[List[dict]] = None, ids: Optional[List[str]] = None, **kwargs: Any, ) -> ADBVST: """Create an Astra DB vectorstore from raw texts. Args: texts (List[str]): the texts to insert. embedding (Embeddings): the embedding function to use in the store. metadatas (Optional[List[dict]]): metadata dicts for the texts. ids (Optional[List[str]]): ids to associate to the texts. *Additional arguments*: you can pass any argument that you would to 'add_texts' and/or to the 'AstraDB' class constructor (see these methods for details). These arguments will be routed to the respective methods as they are. Returns: an `AstraDb` vectorstore. """ known_kwargs = { "collection_name", "token", "api_endpoint", "astra_db_client", "namespace", "metric", "batch_size", "bulk_insert_batch_concurrency", "bulk_insert_overwrite_concurrency", "bulk_delete_concurrency", "batch_concurrency", "overwrite_concurrency", } if kwargs: unknown_kwargs = set(kwargs.keys()) - known_kwargs if unknown_kwargs: warnings.warn( "Method 'from_texts' of AstraDB vector store invoked with " f"unsupported arguments ({', '.join(sorted(unknown_kwargs))}), " "which will be ignored." ) collection_name: str = kwargs["collection_name"] token = kwargs.get("token") api_endpoint = kwargs.get("api_endpoint") astra_db_client = kwargs.get("astra_db_client") namespace = kwargs.get("namespace") metric = kwargs.get("metric") astra_db_store = cls( embedding=embedding, collection_name=collection_name, token=token, api_endpoint=api_endpoint, astra_db_client=astra_db_client, namespace=namespace, metric=metric, batch_size=kwargs.get("batch_size"), bulk_insert_batch_concurrency=kwargs.get("bulk_insert_batch_concurrency"), bulk_insert_overwrite_concurrency=kwargs.get( "bulk_insert_overwrite_concurrency" ), bulk_delete_concurrency=kwargs.get("bulk_delete_concurrency"), ) astra_db_store.add_texts( texts=texts, metadatas=metadatas, ids=ids, batch_size=kwargs.get("batch_size"), batch_concurrency=kwargs.get("batch_concurrency"), overwrite_concurrency=kwargs.get("overwrite_concurrency"), ) return astra_db_store @classmethod def from_documents( cls: Type[ADBVST], documents: List[Document], embedding: Embeddings, **kwargs: Any, ) -> ADBVST: """Create an Astra DB vectorstore from a document list. Utility method that defers to 'from_texts' (see that one). Args: see 'from_texts', except here you have to supply 'documents' in place of 'texts' and 'metadatas'. Returns: an `AstraDB` vectorstore. """ return super().from_documents(documents, embedding, **kwargs)