Fix: FAISSDocumentStore - make write_documents properly work in combination w update_embeddings (#5221)

* Update VERSION.txt

* first draft

* simplify method and test

* rm unnecessary pb.close

* integrate feedback

haystack/document_stores/faiss.py

@@ -257,31 +257,33 @@ class FAISSDocumentStore(SQLDocumentStore):
         document_objects = self._handle_duplicate_documents(
             documents=document_objects, index=index, duplicate_documents=duplicate_documents
         )
-        if len(document_objects) > 0:
-            add_vectors = all(doc.embedding is not None for doc in document_objects)
 
-            if self.duplicate_documents == "overwrite" and add_vectors:
+        if len(document_objects) == 0:
-                logger.warning(
+            return
-                    "You have to provide `duplicate_documents = 'overwrite'` arg and "
-                    "`FAISSDocumentStore` does not support update in existing `faiss_index`.\n"
-                    "Please call `update_embeddings` method to repopulate `faiss_index`"
-                )
 
         vector_id = self.faiss_indexes[index].ntotal
+        add_vectors = all(doc.embedding is not None for doc in document_objects)
+
+        if vector_id > 0 and self.duplicate_documents == "overwrite" and add_vectors:
+            logger.warning(
+                "`FAISSDocumentStore` is adding new vectors to an existing `faiss_index`.\n"
+                "Please call `update_embeddings` method to correctly repopulate `faiss_index`"
+            )
+
         with tqdm(
             total=len(document_objects), disable=not self.progress_bar, position=0, desc="Writing Documents"
         ) as progress_bar:
             for i in range(0, len(document_objects), batch_size):
+                batch_documents = document_objects[i : i + batch_size]
                 if add_vectors:
                     if not self.faiss_indexes[index].is_trained:
                         raise ValueError(
-                                "FAISS index of type {} must be trained before adding vectors. Call `train_index()` "
+                            f"FAISS index of type {self.faiss_index_factory_str} must be trained before adding vectors. Call `train_index()` "
                             "method before adding the vectors. For details, refer to the documentation: "
                             "[FAISSDocumentStore API](https://docs.haystack.deepset.ai/reference/document-store-api#faissdocumentstoretrain_index)."
-                                "".format(self.faiss_index_factory_str)
                         )
 
-                        embeddings = [doc.embedding for doc in document_objects[i : i + batch_size]]
+                    embeddings = [doc.embedding for doc in batch_documents]
                     embeddings_to_index = np.array(embeddings, dtype="float32")
 
                     if self.similarity == "cosine":
@@ -289,22 +291,29 @@ class FAISSDocumentStore(SQLDocumentStore):
 
                     self.faiss_indexes[index].add(embeddings_to_index)
 
+                # write_documents method (duplicate_documents="overwrite") should properly work in combination with
+                # update_embeddings method (update_existing_embeddings=False).
+                # If no new embeddings are provided, we save the existing FAISS vector ids
+                elif self.duplicate_documents == "overwrite":
+                    existing_docs = self.get_documents_by_id(ids=[doc.id for doc in batch_documents], index=index)
+                    existing_docs_vector_ids = {
+                        doc.id: doc.meta["vector_id"] for doc in existing_docs if doc.meta and "vector_id" in doc.meta
+                    }
+
                 docs_to_write_in_sql = []
-                    for doc in document_objects[i : i + batch_size]:
+                for doc in batch_documents:
                     meta = doc.meta
                     if add_vectors:
                         meta["vector_id"] = vector_id
                         vector_id += 1
+                    elif self.duplicate_documents == "overwrite" and doc.id in existing_docs_vector_ids:
+                        meta["vector_id"] = existing_docs_vector_ids[doc.id]
                     docs_to_write_in_sql.append(doc)
 
                 super(FAISSDocumentStore, self).write_documents(
-                        docs_to_write_in_sql,
+                    docs_to_write_in_sql, index=index, duplicate_documents=duplicate_documents, batch_size=batch_size
-                        index=index,
-                        duplicate_documents=duplicate_documents,
-                        batch_size=batch_size,
                 )
                 progress_bar.update(batch_size)
-            progress_bar.close()
 
     def _create_document_field_map(self) -> Dict:
         return {self.index: self.embedding_field}

test/document_stores/test_faiss.py

@@ -204,6 +204,25 @@ class TestFAISSDocumentStore(DocumentStoreBaseTestAbstract):
         assert len(docs_from_index_b) == len(docs_b)
         assert {int(doc.meta["vector_id"]) for doc in docs_from_index_b} == {0, 1, 2, 3}
 
+    @pytest.mark.integration
+    def test_dont_update_existing_embeddings(self, ds, docs):
+        retriever = MockDenseRetriever(document_store=ds)
+        first_doc_id = docs[0].id
+
+        for i in range(1, 4):
+            ds.write_documents(docs[:i])
+            ds.update_embeddings(retriever=retriever, update_existing_embeddings=False)
+
+            assert ds.get_document_count() == i
+            assert ds.get_embedding_count() == i
+            assert ds.get_document_by_id(id=first_doc_id).meta["vector_id"] == "0"
+
+            # Check if the embeddings of the first document remain unchanged after multiple updates
+            if i == 1:
+                first_doc_embedding = ds.get_document_by_id(id=first_doc_id).embedding
+            else:
+                assert np.array_equal(ds.get_document_by_id(id=first_doc_id).embedding, first_doc_embedding)
+
     @pytest.mark.integration
     def test_passing_index_from_outside(self, documents_with_embeddings, tmp_path):
         d = 768