community : [bugfix] Use document ids as keys in AzureSearch vectorst…

…ore (#25486)

# Description
[Vector store base
class](https://github.com/langchain-ai/langchain/blob/4cdaca67dc51dba887289f56c6fead3c1a52f97d/libs/core/langchain_core/vectorstores/base.py#L65)
currently expects `ids` to be passed in and that is what it passes along
to the AzureSearch vector store when attempting to `add_texts()`.
However AzureSearch expects `keys` to be passed in. When they are not
present, AzureSearch `add_embeddings()` makes up new uuids. This is a
problem when trying to run indexing. [Indexing code
expects](https://github.com/langchain-ai/langchain/blob/b297af5482ae7c6d26779513d637ec657a1cd552/libs/core/langchain_core/indexing/api.py#L371)
the documents to be uploaded using provided ids. Currently AzureSearch
ignores `ids` passed from `indexing` and makes up new ones. Later when
`indexer` attempts to delete removed file, it uses the `id` it had
stored when uploading the document, however it was uploaded under
different `id`.

**Twitter handle: @martintriska1**

libs/community/langchain_community/vectorstores/azuresearch.py

@@ -443,6 +443,12 @@ def add_texts(
             logger.debug("Nothing to insert, skipping.")
             return []
 
+        # when `keys` are not passed in and there is `ids` in kwargs, use those instead
+        # base class expects `ids` passed in rather than `keys`
+        # https://github.com/langchain-ai/langchain/blob/4cdaca67dc51dba887289f56c6fead3c1a52f97d/libs/core/langchain_core/vectorstores/base.py#L65
+        if (not keys) and ("ids" in kwargs) and (len(kwargs["ids"]) == len(embeddings)):
+            keys = kwargs["ids"]
+
         return self.add_embeddings(zip(texts, embeddings), metadatas, keys=keys)
 
     async def aadd_texts(
@@ -467,6 +473,12 @@ async def aadd_texts(
             logger.debug("Nothing to insert, skipping.")
             return []
 
+        # when `keys` are not passed in and there is `ids` in kwargs, use those instead
+        # base class expects `ids` passed in rather than `keys`
+        # https://github.com/langchain-ai/langchain/blob/4cdaca67dc51dba887289f56c6fead3c1a52f97d/libs/core/langchain_core/vectorstores/base.py#L65
+        if (not keys) and ("ids" in kwargs) and (len(kwargs["ids"]) == len(embeddings)):
+            keys = kwargs["ids"]
+
         return await self.aadd_embeddings(zip(texts, embeddings), metadatas, keys=keys)
 
     def add_embeddings(
@@ -483,9 +495,13 @@ def add_embeddings(
         data = []
         for i, (text, embedding) in enumerate(text_embeddings):
             # Use provided key otherwise use default key
-            key = keys[i] if keys else str(uuid.uuid4())
-            # Encoding key for Azure Search valid characters
-            key = base64.urlsafe_b64encode(bytes(key, "utf-8")).decode("ascii")
+            if keys:
+                key = keys[i]
+            else:
+                key = str(uuid.uuid4())
+                # Encoding key for Azure Search valid characters
+                key = base64.urlsafe_b64encode(bytes(key, "utf-8")).decode("ascii")
+
             metadata = metadatas[i] if metadatas else {}
             # Add data to index
             # Additional metadata to fields mapping

libs/community/tests/unit_tests/vectorstores/test_azure_search.py

@@ -190,3 +190,40 @@ def mock_create_index() -> None:
         )
         assert vector_store.client is not None
         assert vector_store.client._api_version == "test"
+
+
+@pytest.mark.requires("azure.search.documents")
+def test_ids_used_correctly() -> None:
+    """Check whether vector store uses the document ids when provided with them."""
+    from azure.search.documents import SearchClient
+    from azure.search.documents.indexes import SearchIndexClient
+    from langchain_core.documents import Document
+
+    class Response:
+        def __init__(self) -> None:
+            self.succeeded: bool = True
+
+    def mock_upload_documents(self, documents: List[object]) -> List[Response]:  # type: ignore[no-untyped-def]
+        # assume all documents uploaded successfuly
+        response = [Response() for _ in documents]
+        return response
+
+    documents = [
+        Document(
+            page_content="page zero Lorem Ipsum",
+            metadata={"source": "document.pdf", "page": 0, "id": "ID-document-1"},
+        ),
+        Document(
+            page_content="page one Lorem Ipsum",
+            metadata={"source": "document.pdf", "page": 1, "id": "ID-document-2"},
+        ),
+    ]
+    ids_provided = [i.metadata.get("id") for i in documents]
+
+    with patch.object(
+        SearchClient, "upload_documents", mock_upload_documents
+    ), patch.object(SearchIndexClient, "get_index", mock_default_index):
+        vector_store = create_vector_store()
+        ids_used_at_upload = vector_store.add_documents(documents, ids=ids_provided)
+        assert len(ids_provided) == len(ids_used_at_upload)
+        assert ids_provided == ids_used_at_upload