faiss-haystack 3.0.0__tar.gz → 3.0.1__tar.gz

{faiss_haystack-3.0.0 → faiss_haystack-3.0.1}/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## [integrations/faiss-v3.0.0] - 2026-10-16
+
+### 🧪 Testing
+
+- Use Python 3.11 in tests instead of 3.10 (#4103)
+- Test compatible integrations with Python 3.15; update pyproject (#4104)
+
+### ⚙️ CI
+
+- Improve changelog generation; fix existing changelogs (#3883)
+
+### 🧹 Chores
+
+- Switch remaining document stores to haystack.logging (#4011)
+- [**breaking**] Drop Python 3.10 support (#4105)
+
+
 ## [integrations/faiss-v2.1.0] - 2026-09-01
 
 ### 🐛 Bug Fixes

{faiss_haystack-3.0.0 → faiss_haystack-3.0.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: faiss-haystack
-Version: 3.0.0
+Version: 3.0.1
 Summary: An integration of FAISS (library for efficient similarity search of dense vectors) with Haystack
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/faiss#readme
 Project-URL: Issues, https://github.com/deepset-ai/haystack-core-integrations/issues

{faiss_haystack-3.0.0 → faiss_haystack-3.0.1}/src/haystack_integrations/document_stores/faiss/document_store.py

@@ -152,9 +152,9 @@ class FAISSDocumentStore:
                     msg = f"Document with id '{doc.id}' already exists."
                     raise DuplicateDocumentError(msg)
 
-        # Process documents
-        ids_to_add_to_index = []
-        vectors_to_add = []
+        # Process documents. Pending vectors are keyed by document ID so that a later duplicate in the same
+        # batch replaces the earlier one instead of leaving an orphaned vector in the FAISS index.
+        pending_vectors: dict[str, tuple[int, list[float]]] = {}
 
         docs_written = 0
 
@@ -168,6 +168,7 @@ class FAISSDocumentStore:
                 # FAISS doesn't support easy update. We'd have to remove and add.
                 # For MVP, let's implement remove then add for overwrite.
                 self.delete_documents([doc.id])
+                pending_vectors.pop(doc.id, None)
 
             self.documents[doc.id] = doc
 
@@ -179,15 +180,14 @@ class FAISSDocumentStore:
                 self.id_map[int_id] = doc.id
                 self.inverse_id_map[doc.id] = int_id
 
-                ids_to_add_to_index.append(int_id)
-                vectors_to_add.append(doc.embedding)
+                pending_vectors[doc.id] = (int_id, doc.embedding)
 
             docs_written += 1
 
         # Add to FAISS
-        if vectors_to_add:
-            vectors = np.array(vectors_to_add, dtype="float32")
-            ids = np.array(ids_to_add_to_index, dtype="int64")
+        if pending_vectors:
+            vectors = np.array([vector for _, vector in pending_vectors.values()], dtype="float32")
+            ids = np.array([int_id for int_id, _ in pending_vectors.values()], dtype="int64")
             index = self._get_index_or_raise()
             index.add_with_ids(vectors, ids)
 

{faiss_haystack-3.0.0 → faiss_haystack-3.0.1}/tests/test_document_store.py

@@ -55,6 +55,21 @@ class TestFAISSDocumentStoreUnit:
         with pytest.raises(DuplicateDocumentError, match="already exists"):
             ds.write_documents([Document(id="dup", content="second")], policy=DuplicatePolicy.FAIL)
 
+    def test_write_documents_overwrite_duplicate_in_same_batch_leaves_no_orphan_vector(self):
+        ds = FAISSDocumentStore(embedding_dim=2)
+        docs = [
+            Document(id="a", content="old", embedding=[0.0, 0.0]),
+            Document(id="a", content="new", embedding=[0.1, 0.1]),
+            Document(id="b", content="b", embedding=[5.0, 5.0]),
+        ]
+        ds.write_documents(docs, policy=DuplicatePolicy.OVERWRITE)
+
+        assert ds.count_documents() == 2
+        assert ds.index.ntotal == 2
+        results = ds.search([0.0, 0.0], top_k=2)
+        assert [doc.id for doc in results] == ["a", "b"]
+        assert results[0].content == "new"
+
     @pytest.mark.parametrize(
         ("filters", "error_match"),
         [